llama.cpp eval-callback — ggml op level
Every row below is a real ggml tensor operation that fired inside llama.cpp's own inference engine for this exact forward pass — not module-level, the actual matmuls, norms, views and the fused flash-attention kernel, in firing order.
| # | Tensor | Op | Operands | Output shape | |
|---|---|---|---|---|---|
| 1 | embd | GET_ROWS | token_embd.weight{4096, 151936, 1, 1} , inp_tokens{5, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 2 | norm-0 | RMS_NORM | embd{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 3 | attn_norm-0 | MUL | norm-0{4096, 5, 1, 1} , blk.0.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 4 | Qcur-0 | MUL_MAT | blk.0.attn_q.weight{4096, 4096, 1, 1} , attn_norm-0{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 5 | Qcur-0 | RESHAPE | Qcur-0{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 6 | norm-0 | RMS_NORM | Qcur-0{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 7 | Qcur_normed-0 | MUL | norm-0{128, 32, 5, 1} , blk.0.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 8 | Qcur-0 | ROPE | Qcur_normed-0{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 9 | Vcur-0 | MUL_MAT | blk.0.attn_v.weight{4096, 1024, 1, 1} , attn_norm-0{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 10 | Vcur-0 | RESHAPE | Vcur-0{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 11 | Kcur-0 | MUL_MAT | blk.0.attn_k.weight{4096, 1024, 1, 1} , attn_norm-0{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 12 | Kcur-0 | RESHAPE | Kcur-0{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 13 | norm-0 | RMS_NORM | Kcur-0{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 14 | Kcur_normed-0 | MUL | norm-0{128, 8, 5, 1} , blk.0.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 15 | Kcur-0 | ROPE | Kcur_normed-0{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 16 | Kcur-0 (view) | VIEW | Kcur-0{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 17 | cache_k_l0 (view) | SET_ROWS | Kcur-0 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 18 | Vcur-0 (view) | VIEW | Vcur-0{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 19 | cache_v_l0 (view) | SET_ROWS | Vcur-0 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 20 | Qcur-0 (view) | VIEW | Qcur-0{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 21 | Qcur-0 (view) (permuted) | PERMUTE | Qcur-0 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 22 | cache_k_l0 (view) | VIEW | cache_k_l0{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 23 | cache_k_l0 (view) (permuted) | PERMUTE | cache_k_l0 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 24 | cache_v_l0 (view) | VIEW | cache_v_l0{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 25 | cache_v_l0 (view) (permuted) | PERMUTE | cache_v_l0 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 26 | node_25 | FLASH_ATTN_EXT | Qcur-0 (view) (permuted){128, 5, 32, 1} , cache_k_l0 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 27 | kqv_out-0 | RESHAPE | node_25{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 28 | node_27 | MUL_MAT | blk.0.attn_output.weight{4096, 4096, 1, 1} , kqv_out-0{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 29 | ffn_inp-0 | ADD | node_27{4096, 5, 1, 1} , embd{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 30 | norm-0 | RMS_NORM | ffn_inp-0{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 31 | ffn_norm-0 | MUL | norm-0{4096, 5, 1, 1} , blk.0.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 32 | ffn_gate-0 | MUL_MAT | blk.0.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-0{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 33 | ffn_up-0 | MUL_MAT | blk.0.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-0{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 34 | ffn_swiglu-0 | SWIGLU | ffn_gate-0{12288, 5, 1, 1} , ffn_up-0{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 35 | ffn_out-0 | MUL_MAT | blk.0.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-0{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 36 | l_out-0 | ADD | ffn_out-0{4096, 5, 1, 1} , ffn_inp-0{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 37 | norm-1 | RMS_NORM | l_out-0{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 38 | attn_norm-1 | MUL | norm-1{4096, 5, 1, 1} , blk.1.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 39 | Qcur-1 | MUL_MAT | blk.1.attn_q.weight{4096, 4096, 1, 1} , attn_norm-1{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 40 | Qcur-1 | RESHAPE | Qcur-1{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 41 | norm-1 | RMS_NORM | Qcur-1{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 42 | Qcur_normed-1 | MUL | norm-1{128, 32, 5, 1} , blk.1.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 43 | Qcur-1 | ROPE | Qcur_normed-1{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 44 | Vcur-1 | MUL_MAT | blk.1.attn_v.weight{4096, 1024, 1, 1} , attn_norm-1{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 45 | Vcur-1 | RESHAPE | Vcur-1{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 46 | Kcur-1 | MUL_MAT | blk.1.attn_k.weight{4096, 1024, 1, 1} , attn_norm-1{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 47 | Kcur-1 | RESHAPE | Kcur-1{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 48 | norm-1 | RMS_NORM | Kcur-1{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 49 | Kcur_normed-1 | MUL | norm-1{128, 8, 5, 1} , blk.1.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 50 | Kcur-1 | ROPE | Kcur_normed-1{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 51 | Kcur-1 (view) | VIEW | Kcur-1{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 52 | cache_k_l1 (view) | SET_ROWS | Kcur-1 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 53 | Vcur-1 (view) | VIEW | Vcur-1{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 54 | cache_v_l1 (view) | SET_ROWS | Vcur-1 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 55 | Qcur-1 (view) | VIEW | Qcur-1{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 56 | Qcur-1 (view) (permuted) | PERMUTE | Qcur-1 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 57 | cache_k_l1 (view) | VIEW | cache_k_l1{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 58 | cache_k_l1 (view) (permuted) | PERMUTE | cache_k_l1 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 59 | cache_v_l1 (view) | VIEW | cache_v_l1{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 60 | cache_v_l1 (view) (permuted) | PERMUTE | cache_v_l1 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 61 | node_60 | FLASH_ATTN_EXT | Qcur-1 (view) (permuted){128, 5, 32, 1} , cache_k_l1 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 62 | kqv_out-1 | RESHAPE | node_60{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 63 | node_62 | MUL_MAT | blk.1.attn_output.weight{4096, 4096, 1, 1} , kqv_out-1{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 64 | ffn_inp-1 | ADD | node_62{4096, 5, 1, 1} , l_out-0{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 65 | norm-1 | RMS_NORM | ffn_inp-1{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 66 | ffn_norm-1 | MUL | norm-1{4096, 5, 1, 1} , blk.1.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 67 | ffn_gate-1 | MUL_MAT | blk.1.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-1{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 68 | ffn_up-1 | MUL_MAT | blk.1.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-1{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 69 | ffn_swiglu-1 | SWIGLU | ffn_gate-1{12288, 5, 1, 1} , ffn_up-1{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 70 | ffn_out-1 | MUL_MAT | blk.1.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-1{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 71 | l_out-1 | ADD | ffn_out-1{4096, 5, 1, 1} , ffn_inp-1{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 72 | norm-2 | RMS_NORM | l_out-1{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 73 | attn_norm-2 | MUL | norm-2{4096, 5, 1, 1} , blk.2.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 74 | Qcur-2 | MUL_MAT | blk.2.attn_q.weight{4096, 4096, 1, 1} , attn_norm-2{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 75 | Qcur-2 | RESHAPE | Qcur-2{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 76 | norm-2 | RMS_NORM | Qcur-2{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 77 | Qcur_normed-2 | MUL | norm-2{128, 32, 5, 1} , blk.2.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 78 | Qcur-2 | ROPE | Qcur_normed-2{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 79 | Vcur-2 | MUL_MAT | blk.2.attn_v.weight{4096, 1024, 1, 1} , attn_norm-2{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 80 | Vcur-2 | RESHAPE | Vcur-2{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 81 | Kcur-2 | MUL_MAT | blk.2.attn_k.weight{4096, 1024, 1, 1} , attn_norm-2{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 82 | Kcur-2 | RESHAPE | Kcur-2{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 83 | norm-2 | RMS_NORM | Kcur-2{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 84 | Kcur_normed-2 | MUL | norm-2{128, 8, 5, 1} , blk.2.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 85 | Kcur-2 | ROPE | Kcur_normed-2{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 86 | Kcur-2 (view) | VIEW | Kcur-2{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 87 | cache_k_l2 (view) | SET_ROWS | Kcur-2 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 88 | Vcur-2 (view) | VIEW | Vcur-2{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 89 | cache_v_l2 (view) | SET_ROWS | Vcur-2 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 90 | Qcur-2 (view) | VIEW | Qcur-2{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 91 | Qcur-2 (view) (permuted) | PERMUTE | Qcur-2 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 92 | cache_k_l2 (view) | VIEW | cache_k_l2{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 93 | cache_k_l2 (view) (permuted) | PERMUTE | cache_k_l2 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 94 | cache_v_l2 (view) | VIEW | cache_v_l2{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 95 | cache_v_l2 (view) (permuted) | PERMUTE | cache_v_l2 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 96 | node_95 | FLASH_ATTN_EXT | Qcur-2 (view) (permuted){128, 5, 32, 1} , cache_k_l2 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 97 | kqv_out-2 | RESHAPE | node_95{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 98 | node_97 | MUL_MAT | blk.2.attn_output.weight{4096, 4096, 1, 1} , kqv_out-2{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 99 | ffn_inp-2 | ADD | node_97{4096, 5, 1, 1} , l_out-1{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 100 | norm-2 | RMS_NORM | ffn_inp-2{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 101 | ffn_norm-2 | MUL | norm-2{4096, 5, 1, 1} , blk.2.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 102 | ffn_gate-2 | MUL_MAT | blk.2.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-2{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 103 | ffn_up-2 | MUL_MAT | blk.2.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-2{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 104 | ffn_swiglu-2 | SWIGLU | ffn_gate-2{12288, 5, 1, 1} , ffn_up-2{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 105 | ffn_out-2 | MUL_MAT | blk.2.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-2{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 106 | l_out-2 | ADD | ffn_out-2{4096, 5, 1, 1} , ffn_inp-2{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 107 | norm-3 | RMS_NORM | l_out-2{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 108 | attn_norm-3 | MUL | norm-3{4096, 5, 1, 1} , blk.3.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 109 | Qcur-3 | MUL_MAT | blk.3.attn_q.weight{4096, 4096, 1, 1} , attn_norm-3{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 110 | Qcur-3 | RESHAPE | Qcur-3{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 111 | norm-3 | RMS_NORM | Qcur-3{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 112 | Qcur_normed-3 | MUL | norm-3{128, 32, 5, 1} , blk.3.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 113 | Qcur-3 | ROPE | Qcur_normed-3{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 114 | Vcur-3 | MUL_MAT | blk.3.attn_v.weight{4096, 1024, 1, 1} , attn_norm-3{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 115 | Vcur-3 | RESHAPE | Vcur-3{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 116 | Kcur-3 | MUL_MAT | blk.3.attn_k.weight{4096, 1024, 1, 1} , attn_norm-3{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 117 | Kcur-3 | RESHAPE | Kcur-3{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 118 | norm-3 | RMS_NORM | Kcur-3{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 119 | Kcur_normed-3 | MUL | norm-3{128, 8, 5, 1} , blk.3.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 120 | Kcur-3 | ROPE | Kcur_normed-3{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 121 | Kcur-3 (view) | VIEW | Kcur-3{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 122 | cache_k_l3 (view) | SET_ROWS | Kcur-3 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 123 | Vcur-3 (view) | VIEW | Vcur-3{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 124 | cache_v_l3 (view) | SET_ROWS | Vcur-3 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 125 | Qcur-3 (view) | VIEW | Qcur-3{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 126 | Qcur-3 (view) (permuted) | PERMUTE | Qcur-3 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 127 | cache_k_l3 (view) | VIEW | cache_k_l3{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 128 | cache_k_l3 (view) (permuted) | PERMUTE | cache_k_l3 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 129 | cache_v_l3 (view) | VIEW | cache_v_l3{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 130 | cache_v_l3 (view) (permuted) | PERMUTE | cache_v_l3 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 131 | node_130 | FLASH_ATTN_EXT | Qcur-3 (view) (permuted){128, 5, 32, 1} , cache_k_l3 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 132 | kqv_out-3 | RESHAPE | node_130{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 133 | node_132 | MUL_MAT | blk.3.attn_output.weight{4096, 4096, 1, 1} , kqv_out-3{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 134 | ffn_inp-3 | ADD | node_132{4096, 5, 1, 1} , l_out-2{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 135 | norm-3 | RMS_NORM | ffn_inp-3{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 136 | ffn_norm-3 | MUL | norm-3{4096, 5, 1, 1} , blk.3.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 137 | ffn_gate-3 | MUL_MAT | blk.3.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-3{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 138 | ffn_up-3 | MUL_MAT | blk.3.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-3{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 139 | ffn_swiglu-3 | SWIGLU | ffn_gate-3{12288, 5, 1, 1} , ffn_up-3{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 140 | ffn_out-3 | MUL_MAT | blk.3.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-3{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 141 | l_out-3 | ADD | ffn_out-3{4096, 5, 1, 1} , ffn_inp-3{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 142 | norm-4 | RMS_NORM | l_out-3{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 143 | attn_norm-4 | MUL | norm-4{4096, 5, 1, 1} , blk.4.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 144 | Qcur-4 | MUL_MAT | blk.4.attn_q.weight{4096, 4096, 1, 1} , attn_norm-4{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 145 | Qcur-4 | RESHAPE | Qcur-4{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 146 | norm-4 | RMS_NORM | Qcur-4{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 147 | Qcur_normed-4 | MUL | norm-4{128, 32, 5, 1} , blk.4.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 148 | Qcur-4 | ROPE | Qcur_normed-4{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 149 | Vcur-4 | MUL_MAT | blk.4.attn_v.weight{4096, 1024, 1, 1} , attn_norm-4{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 150 | Vcur-4 | RESHAPE | Vcur-4{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 151 | Kcur-4 | MUL_MAT | blk.4.attn_k.weight{4096, 1024, 1, 1} , attn_norm-4{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 152 | Kcur-4 | RESHAPE | Kcur-4{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 153 | norm-4 | RMS_NORM | Kcur-4{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 154 | Kcur_normed-4 | MUL | norm-4{128, 8, 5, 1} , blk.4.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 155 | Kcur-4 | ROPE | Kcur_normed-4{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 156 | Kcur-4 (view) | VIEW | Kcur-4{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 157 | cache_k_l4 (view) | SET_ROWS | Kcur-4 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 158 | Vcur-4 (view) | VIEW | Vcur-4{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 159 | cache_v_l4 (view) | SET_ROWS | Vcur-4 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 160 | Qcur-4 (view) | VIEW | Qcur-4{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 161 | Qcur-4 (view) (permuted) | PERMUTE | Qcur-4 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 162 | cache_k_l4 (view) | VIEW | cache_k_l4{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 163 | cache_k_l4 (view) (permuted) | PERMUTE | cache_k_l4 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 164 | cache_v_l4 (view) | VIEW | cache_v_l4{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 165 | cache_v_l4 (view) (permuted) | PERMUTE | cache_v_l4 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 166 | node_165 | FLASH_ATTN_EXT | Qcur-4 (view) (permuted){128, 5, 32, 1} , cache_k_l4 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 167 | kqv_out-4 | RESHAPE | node_165{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 168 | node_167 | MUL_MAT | blk.4.attn_output.weight{4096, 4096, 1, 1} , kqv_out-4{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 169 | ffn_inp-4 | ADD | node_167{4096, 5, 1, 1} , l_out-3{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 170 | norm-4 | RMS_NORM | ffn_inp-4{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 171 | ffn_norm-4 | MUL | norm-4{4096, 5, 1, 1} , blk.4.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 172 | ffn_gate-4 | MUL_MAT | blk.4.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-4{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 173 | ffn_up-4 | MUL_MAT | blk.4.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-4{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 174 | ffn_swiglu-4 | SWIGLU | ffn_gate-4{12288, 5, 1, 1} , ffn_up-4{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 175 | ffn_out-4 | MUL_MAT | blk.4.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-4{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 176 | l_out-4 | ADD | ffn_out-4{4096, 5, 1, 1} , ffn_inp-4{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 177 | norm-5 | RMS_NORM | l_out-4{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 178 | attn_norm-5 | MUL | norm-5{4096, 5, 1, 1} , blk.5.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 179 | Qcur-5 | MUL_MAT | blk.5.attn_q.weight{4096, 4096, 1, 1} , attn_norm-5{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 180 | Qcur-5 | RESHAPE | Qcur-5{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 181 | norm-5 | RMS_NORM | Qcur-5{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 182 | Qcur_normed-5 | MUL | norm-5{128, 32, 5, 1} , blk.5.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 183 | Qcur-5 | ROPE | Qcur_normed-5{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 184 | Vcur-5 | MUL_MAT | blk.5.attn_v.weight{4096, 1024, 1, 1} , attn_norm-5{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 185 | Vcur-5 | RESHAPE | Vcur-5{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 186 | Kcur-5 | MUL_MAT | blk.5.attn_k.weight{4096, 1024, 1, 1} , attn_norm-5{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 187 | Kcur-5 | RESHAPE | Kcur-5{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 188 | norm-5 | RMS_NORM | Kcur-5{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 189 | Kcur_normed-5 | MUL | norm-5{128, 8, 5, 1} , blk.5.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 190 | Kcur-5 | ROPE | Kcur_normed-5{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 191 | Kcur-5 (view) | VIEW | Kcur-5{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 192 | cache_k_l5 (view) | SET_ROWS | Kcur-5 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 193 | Vcur-5 (view) | VIEW | Vcur-5{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 194 | cache_v_l5 (view) | SET_ROWS | Vcur-5 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 195 | Qcur-5 (view) | VIEW | Qcur-5{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 196 | Qcur-5 (view) (permuted) | PERMUTE | Qcur-5 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 197 | cache_k_l5 (view) | VIEW | cache_k_l5{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 198 | cache_k_l5 (view) (permuted) | PERMUTE | cache_k_l5 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 199 | cache_v_l5 (view) | VIEW | cache_v_l5{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 200 | cache_v_l5 (view) (permuted) | PERMUTE | cache_v_l5 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 201 | node_200 | FLASH_ATTN_EXT | Qcur-5 (view) (permuted){128, 5, 32, 1} , cache_k_l5 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 202 | kqv_out-5 | RESHAPE | node_200{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 203 | node_202 | MUL_MAT | blk.5.attn_output.weight{4096, 4096, 1, 1} , kqv_out-5{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 204 | ffn_inp-5 | ADD | node_202{4096, 5, 1, 1} , l_out-4{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 205 | norm-5 | RMS_NORM | ffn_inp-5{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 206 | ffn_norm-5 | MUL | norm-5{4096, 5, 1, 1} , blk.5.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 207 | ffn_gate-5 | MUL_MAT | blk.5.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-5{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 208 | ffn_up-5 | MUL_MAT | blk.5.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-5{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 209 | ffn_swiglu-5 | SWIGLU | ffn_gate-5{12288, 5, 1, 1} , ffn_up-5{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 210 | ffn_out-5 | MUL_MAT | blk.5.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-5{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 211 | l_out-5 | ADD | ffn_out-5{4096, 5, 1, 1} , ffn_inp-5{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 212 | norm-6 | RMS_NORM | l_out-5{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 213 | attn_norm-6 | MUL | norm-6{4096, 5, 1, 1} , blk.6.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 214 | Qcur-6 | MUL_MAT | blk.6.attn_q.weight{4096, 4096, 1, 1} , attn_norm-6{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 215 | Qcur-6 | RESHAPE | Qcur-6{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 216 | norm-6 | RMS_NORM | Qcur-6{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 217 | Qcur_normed-6 | MUL | norm-6{128, 32, 5, 1} , blk.6.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 218 | Qcur-6 | ROPE | Qcur_normed-6{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 219 | Vcur-6 | MUL_MAT | blk.6.attn_v.weight{4096, 1024, 1, 1} , attn_norm-6{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 220 | Vcur-6 | RESHAPE | Vcur-6{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 221 | Kcur-6 | MUL_MAT | blk.6.attn_k.weight{4096, 1024, 1, 1} , attn_norm-6{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 222 | Kcur-6 | RESHAPE | Kcur-6{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 223 | norm-6 | RMS_NORM | Kcur-6{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 224 | Kcur_normed-6 | MUL | norm-6{128, 8, 5, 1} , blk.6.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 225 | Kcur-6 | ROPE | Kcur_normed-6{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 226 | Kcur-6 (view) | VIEW | Kcur-6{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 227 | cache_k_l6 (view) | SET_ROWS | Kcur-6 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 228 | Vcur-6 (view) | VIEW | Vcur-6{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 229 | cache_v_l6 (view) | SET_ROWS | Vcur-6 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 230 | Qcur-6 (view) | VIEW | Qcur-6{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 231 | Qcur-6 (view) (permuted) | PERMUTE | Qcur-6 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 232 | cache_k_l6 (view) | VIEW | cache_k_l6{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 233 | cache_k_l6 (view) (permuted) | PERMUTE | cache_k_l6 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 234 | cache_v_l6 (view) | VIEW | cache_v_l6{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 235 | cache_v_l6 (view) (permuted) | PERMUTE | cache_v_l6 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 236 | node_235 | FLASH_ATTN_EXT | Qcur-6 (view) (permuted){128, 5, 32, 1} , cache_k_l6 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 237 | kqv_out-6 | RESHAPE | node_235{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 238 | node_237 | MUL_MAT | blk.6.attn_output.weight{4096, 4096, 1, 1} , kqv_out-6{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 239 | ffn_inp-6 | ADD | node_237{4096, 5, 1, 1} , l_out-5{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 240 | norm-6 | RMS_NORM | ffn_inp-6{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 241 | ffn_norm-6 | MUL | norm-6{4096, 5, 1, 1} , blk.6.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 242 | ffn_gate-6 | MUL_MAT | blk.6.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-6{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 243 | ffn_up-6 | MUL_MAT | blk.6.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-6{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 244 | ffn_swiglu-6 | SWIGLU | ffn_gate-6{12288, 5, 1, 1} , ffn_up-6{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 245 | ffn_out-6 | MUL_MAT | blk.6.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-6{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 246 | l_out-6 | ADD | ffn_out-6{4096, 5, 1, 1} , ffn_inp-6{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 247 | norm-7 | RMS_NORM | l_out-6{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 248 | attn_norm-7 | MUL | norm-7{4096, 5, 1, 1} , blk.7.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 249 | Qcur-7 | MUL_MAT | blk.7.attn_q.weight{4096, 4096, 1, 1} , attn_norm-7{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 250 | Qcur-7 | RESHAPE | Qcur-7{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 251 | norm-7 | RMS_NORM | Qcur-7{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 252 | Qcur_normed-7 | MUL | norm-7{128, 32, 5, 1} , blk.7.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 253 | Qcur-7 | ROPE | Qcur_normed-7{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 254 | Vcur-7 | MUL_MAT | blk.7.attn_v.weight{4096, 1024, 1, 1} , attn_norm-7{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 255 | Vcur-7 | RESHAPE | Vcur-7{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 256 | Kcur-7 | MUL_MAT | blk.7.attn_k.weight{4096, 1024, 1, 1} , attn_norm-7{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 257 | Kcur-7 | RESHAPE | Kcur-7{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 258 | norm-7 | RMS_NORM | Kcur-7{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 259 | Kcur_normed-7 | MUL | norm-7{128, 8, 5, 1} , blk.7.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 260 | Kcur-7 | ROPE | Kcur_normed-7{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 261 | Kcur-7 (view) | VIEW | Kcur-7{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 262 | cache_k_l7 (view) | SET_ROWS | Kcur-7 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 263 | Vcur-7 (view) | VIEW | Vcur-7{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 264 | cache_v_l7 (view) | SET_ROWS | Vcur-7 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 265 | Qcur-7 (view) | VIEW | Qcur-7{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 266 | Qcur-7 (view) (permuted) | PERMUTE | Qcur-7 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 267 | cache_k_l7 (view) | VIEW | cache_k_l7{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 268 | cache_k_l7 (view) (permuted) | PERMUTE | cache_k_l7 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 269 | cache_v_l7 (view) | VIEW | cache_v_l7{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 270 | cache_v_l7 (view) (permuted) | PERMUTE | cache_v_l7 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 271 | node_270 | FLASH_ATTN_EXT | Qcur-7 (view) (permuted){128, 5, 32, 1} , cache_k_l7 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 272 | kqv_out-7 | RESHAPE | node_270{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 273 | node_272 | MUL_MAT | blk.7.attn_output.weight{4096, 4096, 1, 1} , kqv_out-7{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 274 | ffn_inp-7 | ADD | node_272{4096, 5, 1, 1} , l_out-6{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 275 | norm-7 | RMS_NORM | ffn_inp-7{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 276 | ffn_norm-7 | MUL | norm-7{4096, 5, 1, 1} , blk.7.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 277 | ffn_gate-7 | MUL_MAT | blk.7.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-7{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 278 | ffn_up-7 | MUL_MAT | blk.7.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-7{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 279 | ffn_swiglu-7 | SWIGLU | ffn_gate-7{12288, 5, 1, 1} , ffn_up-7{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 280 | ffn_out-7 | MUL_MAT | blk.7.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-7{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 281 | l_out-7 | ADD | ffn_out-7{4096, 5, 1, 1} , ffn_inp-7{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 282 | norm-8 | RMS_NORM | l_out-7{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 283 | attn_norm-8 | MUL | norm-8{4096, 5, 1, 1} , blk.8.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 284 | Qcur-8 | MUL_MAT | blk.8.attn_q.weight{4096, 4096, 1, 1} , attn_norm-8{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 285 | Qcur-8 | RESHAPE | Qcur-8{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 286 | norm-8 | RMS_NORM | Qcur-8{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 287 | Qcur_normed-8 | MUL | norm-8{128, 32, 5, 1} , blk.8.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 288 | Qcur-8 | ROPE | Qcur_normed-8{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 289 | Vcur-8 | MUL_MAT | blk.8.attn_v.weight{4096, 1024, 1, 1} , attn_norm-8{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 290 | Vcur-8 | RESHAPE | Vcur-8{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 291 | Kcur-8 | MUL_MAT | blk.8.attn_k.weight{4096, 1024, 1, 1} , attn_norm-8{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 292 | Kcur-8 | RESHAPE | Kcur-8{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 293 | norm-8 | RMS_NORM | Kcur-8{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 294 | Kcur_normed-8 | MUL | norm-8{128, 8, 5, 1} , blk.8.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 295 | Kcur-8 | ROPE | Kcur_normed-8{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 296 | Kcur-8 (view) | VIEW | Kcur-8{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 297 | cache_k_l8 (view) | SET_ROWS | Kcur-8 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 298 | Vcur-8 (view) | VIEW | Vcur-8{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 299 | cache_v_l8 (view) | SET_ROWS | Vcur-8 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 300 | Qcur-8 (view) | VIEW | Qcur-8{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 301 | Qcur-8 (view) (permuted) | PERMUTE | Qcur-8 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 302 | cache_k_l8 (view) | VIEW | cache_k_l8{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 303 | cache_k_l8 (view) (permuted) | PERMUTE | cache_k_l8 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 304 | cache_v_l8 (view) | VIEW | cache_v_l8{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 305 | cache_v_l8 (view) (permuted) | PERMUTE | cache_v_l8 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 306 | node_305 | FLASH_ATTN_EXT | Qcur-8 (view) (permuted){128, 5, 32, 1} , cache_k_l8 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 307 | kqv_out-8 | RESHAPE | node_305{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 308 | node_307 | MUL_MAT | blk.8.attn_output.weight{4096, 4096, 1, 1} , kqv_out-8{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 309 | ffn_inp-8 | ADD | node_307{4096, 5, 1, 1} , l_out-7{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 310 | norm-8 | RMS_NORM | ffn_inp-8{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 311 | ffn_norm-8 | MUL | norm-8{4096, 5, 1, 1} , blk.8.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 312 | ffn_gate-8 | MUL_MAT | blk.8.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-8{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 313 | ffn_up-8 | MUL_MAT | blk.8.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-8{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 314 | ffn_swiglu-8 | SWIGLU | ffn_gate-8{12288, 5, 1, 1} , ffn_up-8{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 315 | ffn_out-8 | MUL_MAT | blk.8.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-8{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 316 | l_out-8 | ADD | ffn_out-8{4096, 5, 1, 1} , ffn_inp-8{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 317 | norm-9 | RMS_NORM | l_out-8{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 318 | attn_norm-9 | MUL | norm-9{4096, 5, 1, 1} , blk.9.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 319 | Qcur-9 | MUL_MAT | blk.9.attn_q.weight{4096, 4096, 1, 1} , attn_norm-9{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 320 | Qcur-9 | RESHAPE | Qcur-9{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 321 | norm-9 | RMS_NORM | Qcur-9{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 322 | Qcur_normed-9 | MUL | norm-9{128, 32, 5, 1} , blk.9.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 323 | Qcur-9 | ROPE | Qcur_normed-9{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 324 | Vcur-9 | MUL_MAT | blk.9.attn_v.weight{4096, 1024, 1, 1} , attn_norm-9{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 325 | Vcur-9 | RESHAPE | Vcur-9{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 326 | Kcur-9 | MUL_MAT | blk.9.attn_k.weight{4096, 1024, 1, 1} , attn_norm-9{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 327 | Kcur-9 | RESHAPE | Kcur-9{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 328 | norm-9 | RMS_NORM | Kcur-9{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 329 | Kcur_normed-9 | MUL | norm-9{128, 8, 5, 1} , blk.9.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 330 | Kcur-9 | ROPE | Kcur_normed-9{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 331 | Kcur-9 (view) | VIEW | Kcur-9{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 332 | cache_k_l9 (view) | SET_ROWS | Kcur-9 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 333 | Vcur-9 (view) | VIEW | Vcur-9{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 334 | cache_v_l9 (view) | SET_ROWS | Vcur-9 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 335 | Qcur-9 (view) | VIEW | Qcur-9{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 336 | Qcur-9 (view) (permuted) | PERMUTE | Qcur-9 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 337 | cache_k_l9 (view) | VIEW | cache_k_l9{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 338 | cache_k_l9 (view) (permuted) | PERMUTE | cache_k_l9 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 339 | cache_v_l9 (view) | VIEW | cache_v_l9{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 340 | cache_v_l9 (view) (permuted) | PERMUTE | cache_v_l9 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 341 | node_340 | FLASH_ATTN_EXT | Qcur-9 (view) (permuted){128, 5, 32, 1} , cache_k_l9 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 342 | kqv_out-9 | RESHAPE | node_340{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 343 | node_342 | MUL_MAT | blk.9.attn_output.weight{4096, 4096, 1, 1} , kqv_out-9{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 344 | ffn_inp-9 | ADD | node_342{4096, 5, 1, 1} , l_out-8{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 345 | norm-9 | RMS_NORM | ffn_inp-9{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 346 | ffn_norm-9 | MUL | norm-9{4096, 5, 1, 1} , blk.9.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 347 | ffn_gate-9 | MUL_MAT | blk.9.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-9{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 348 | ffn_up-9 | MUL_MAT | blk.9.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-9{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 349 | ffn_swiglu-9 | SWIGLU | ffn_gate-9{12288, 5, 1, 1} , ffn_up-9{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 350 | ffn_out-9 | MUL_MAT | blk.9.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-9{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 351 | l_out-9 | ADD | ffn_out-9{4096, 5, 1, 1} , ffn_inp-9{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 352 | norm-10 | RMS_NORM | l_out-9{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 353 | attn_norm-10 | MUL | norm-10{4096, 5, 1, 1} , blk.10.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 354 | Qcur-10 | MUL_MAT | blk.10.attn_q.weight{4096, 4096, 1, 1} , attn_norm-10{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 355 | Qcur-10 | RESHAPE | Qcur-10{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 356 | norm-10 | RMS_NORM | Qcur-10{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 357 | Qcur_normed-10 | MUL | norm-10{128, 32, 5, 1} , blk.10.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 358 | Qcur-10 | ROPE | Qcur_normed-10{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 359 | Vcur-10 | MUL_MAT | blk.10.attn_v.weight{4096, 1024, 1, 1} , attn_norm-10{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 360 | Vcur-10 | RESHAPE | Vcur-10{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 361 | Kcur-10 | MUL_MAT | blk.10.attn_k.weight{4096, 1024, 1, 1} , attn_norm-10{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 362 | Kcur-10 | RESHAPE | Kcur-10{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 363 | norm-10 | RMS_NORM | Kcur-10{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 364 | Kcur_normed-10 | MUL | norm-10{128, 8, 5, 1} , blk.10.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 365 | Kcur-10 | ROPE | Kcur_normed-10{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 366 | Kcur-10 (view) | VIEW | Kcur-10{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 367 | cache_k_l10 (view) | SET_ROWS | Kcur-10 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 368 | Vcur-10 (view) | VIEW | Vcur-10{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 369 | cache_v_l10 (view) | SET_ROWS | Vcur-10 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 370 | Qcur-10 (view) | VIEW | Qcur-10{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 371 | Qcur-10 (view) (permuted) | PERMUTE | Qcur-10 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 372 | cache_k_l10 (view) | VIEW | cache_k_l10{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 373 | cache_k_l10 (view) (permuted) | PERMUTE | cache_k_l10 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 374 | cache_v_l10 (view) | VIEW | cache_v_l10{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 375 | cache_v_l10 (view) (permuted) | PERMUTE | cache_v_l10 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 376 | node_375 | FLASH_ATTN_EXT | Qcur-10 (view) (permuted){128, 5, 32, 1} , cache_k_l10 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 377 | kqv_out-10 | RESHAPE | node_375{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 378 | node_377 | MUL_MAT | blk.10.attn_output.weight{4096, 4096, 1, 1} , kqv_out-10{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 379 | ffn_inp-10 | ADD | node_377{4096, 5, 1, 1} , l_out-9{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 380 | norm-10 | RMS_NORM | ffn_inp-10{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 381 | ffn_norm-10 | MUL | norm-10{4096, 5, 1, 1} , blk.10.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 382 | ffn_gate-10 | MUL_MAT | blk.10.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-10{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 383 | ffn_up-10 | MUL_MAT | blk.10.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-10{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 384 | ffn_swiglu-10 | SWIGLU | ffn_gate-10{12288, 5, 1, 1} , ffn_up-10{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 385 | ffn_out-10 | MUL_MAT | blk.10.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-10{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 386 | l_out-10 | ADD | ffn_out-10{4096, 5, 1, 1} , ffn_inp-10{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 387 | norm-11 | RMS_NORM | l_out-10{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 388 | attn_norm-11 | MUL | norm-11{4096, 5, 1, 1} , blk.11.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 389 | Qcur-11 | MUL_MAT | blk.11.attn_q.weight{4096, 4096, 1, 1} , attn_norm-11{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 390 | Qcur-11 | RESHAPE | Qcur-11{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 391 | norm-11 | RMS_NORM | Qcur-11{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 392 | Qcur_normed-11 | MUL | norm-11{128, 32, 5, 1} , blk.11.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 393 | Qcur-11 | ROPE | Qcur_normed-11{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 394 | Vcur-11 | MUL_MAT | blk.11.attn_v.weight{4096, 1024, 1, 1} , attn_norm-11{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 395 | Vcur-11 | RESHAPE | Vcur-11{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 396 | Kcur-11 | MUL_MAT | blk.11.attn_k.weight{4096, 1024, 1, 1} , attn_norm-11{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 397 | Kcur-11 | RESHAPE | Kcur-11{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 398 | norm-11 | RMS_NORM | Kcur-11{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 399 | Kcur_normed-11 | MUL | norm-11{128, 8, 5, 1} , blk.11.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 400 | Kcur-11 | ROPE | Kcur_normed-11{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 401 | Kcur-11 (view) | VIEW | Kcur-11{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 402 | cache_k_l11 (view) | SET_ROWS | Kcur-11 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 403 | Vcur-11 (view) | VIEW | Vcur-11{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 404 | cache_v_l11 (view) | SET_ROWS | Vcur-11 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 405 | Qcur-11 (view) | VIEW | Qcur-11{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 406 | Qcur-11 (view) (permuted) | PERMUTE | Qcur-11 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 407 | cache_k_l11 (view) | VIEW | cache_k_l11{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 408 | cache_k_l11 (view) (permuted) | PERMUTE | cache_k_l11 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 409 | cache_v_l11 (view) | VIEW | cache_v_l11{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 410 | cache_v_l11 (view) (permuted) | PERMUTE | cache_v_l11 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 411 | node_410 | FLASH_ATTN_EXT | Qcur-11 (view) (permuted){128, 5, 32, 1} , cache_k_l11 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 412 | kqv_out-11 | RESHAPE | node_410{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 413 | node_412 | MUL_MAT | blk.11.attn_output.weight{4096, 4096, 1, 1} , kqv_out-11{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 414 | ffn_inp-11 | ADD | node_412{4096, 5, 1, 1} , l_out-10{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 415 | norm-11 | RMS_NORM | ffn_inp-11{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 416 | ffn_norm-11 | MUL | norm-11{4096, 5, 1, 1} , blk.11.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 417 | ffn_gate-11 | MUL_MAT | blk.11.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-11{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 418 | ffn_up-11 | MUL_MAT | blk.11.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-11{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 419 | ffn_swiglu-11 | SWIGLU | ffn_gate-11{12288, 5, 1, 1} , ffn_up-11{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 420 | ffn_out-11 | MUL_MAT | blk.11.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-11{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 421 | l_out-11 | ADD | ffn_out-11{4096, 5, 1, 1} , ffn_inp-11{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 422 | norm-12 | RMS_NORM | l_out-11{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 423 | attn_norm-12 | MUL | norm-12{4096, 5, 1, 1} , blk.12.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 424 | Qcur-12 | MUL_MAT | blk.12.attn_q.weight{4096, 4096, 1, 1} , attn_norm-12{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 425 | Qcur-12 | RESHAPE | Qcur-12{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 426 | norm-12 | RMS_NORM | Qcur-12{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 427 | Qcur_normed-12 | MUL | norm-12{128, 32, 5, 1} , blk.12.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 428 | Qcur-12 | ROPE | Qcur_normed-12{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 429 | Vcur-12 | MUL_MAT | blk.12.attn_v.weight{4096, 1024, 1, 1} , attn_norm-12{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 430 | Vcur-12 | RESHAPE | Vcur-12{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 431 | Kcur-12 | MUL_MAT | blk.12.attn_k.weight{4096, 1024, 1, 1} , attn_norm-12{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 432 | Kcur-12 | RESHAPE | Kcur-12{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 433 | norm-12 | RMS_NORM | Kcur-12{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 434 | Kcur_normed-12 | MUL | norm-12{128, 8, 5, 1} , blk.12.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 435 | Kcur-12 | ROPE | Kcur_normed-12{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 436 | Kcur-12 (view) | VIEW | Kcur-12{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 437 | cache_k_l12 (view) | SET_ROWS | Kcur-12 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 438 | Vcur-12 (view) | VIEW | Vcur-12{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 439 | cache_v_l12 (view) | SET_ROWS | Vcur-12 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 440 | Qcur-12 (view) | VIEW | Qcur-12{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 441 | Qcur-12 (view) (permuted) | PERMUTE | Qcur-12 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 442 | cache_k_l12 (view) | VIEW | cache_k_l12{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 443 | cache_k_l12 (view) (permuted) | PERMUTE | cache_k_l12 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 444 | cache_v_l12 (view) | VIEW | cache_v_l12{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 445 | cache_v_l12 (view) (permuted) | PERMUTE | cache_v_l12 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 446 | node_445 | FLASH_ATTN_EXT | Qcur-12 (view) (permuted){128, 5, 32, 1} , cache_k_l12 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 447 | kqv_out-12 | RESHAPE | node_445{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 448 | node_447 | MUL_MAT | blk.12.attn_output.weight{4096, 4096, 1, 1} , kqv_out-12{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 449 | ffn_inp-12 | ADD | node_447{4096, 5, 1, 1} , l_out-11{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 450 | norm-12 | RMS_NORM | ffn_inp-12{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 451 | ffn_norm-12 | MUL | norm-12{4096, 5, 1, 1} , blk.12.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 452 | ffn_gate-12 | MUL_MAT | blk.12.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-12{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 453 | ffn_up-12 | MUL_MAT | blk.12.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-12{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 454 | ffn_swiglu-12 | SWIGLU | ffn_gate-12{12288, 5, 1, 1} , ffn_up-12{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 455 | ffn_out-12 | MUL_MAT | blk.12.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-12{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 456 | l_out-12 | ADD | ffn_out-12{4096, 5, 1, 1} , ffn_inp-12{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 457 | norm-13 | RMS_NORM | l_out-12{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 458 | attn_norm-13 | MUL | norm-13{4096, 5, 1, 1} , blk.13.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 459 | Qcur-13 | MUL_MAT | blk.13.attn_q.weight{4096, 4096, 1, 1} , attn_norm-13{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 460 | Qcur-13 | RESHAPE | Qcur-13{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 461 | norm-13 | RMS_NORM | Qcur-13{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 462 | Qcur_normed-13 | MUL | norm-13{128, 32, 5, 1} , blk.13.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 463 | Qcur-13 | ROPE | Qcur_normed-13{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 464 | Vcur-13 | MUL_MAT | blk.13.attn_v.weight{4096, 1024, 1, 1} , attn_norm-13{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 465 | Vcur-13 | RESHAPE | Vcur-13{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 466 | Kcur-13 | MUL_MAT | blk.13.attn_k.weight{4096, 1024, 1, 1} , attn_norm-13{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 467 | Kcur-13 | RESHAPE | Kcur-13{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 468 | norm-13 | RMS_NORM | Kcur-13{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 469 | Kcur_normed-13 | MUL | norm-13{128, 8, 5, 1} , blk.13.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 470 | Kcur-13 | ROPE | Kcur_normed-13{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 471 | Kcur-13 (view) | VIEW | Kcur-13{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 472 | cache_k_l13 (view) | SET_ROWS | Kcur-13 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 473 | Vcur-13 (view) | VIEW | Vcur-13{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 474 | cache_v_l13 (view) | SET_ROWS | Vcur-13 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 475 | Qcur-13 (view) | VIEW | Qcur-13{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 476 | Qcur-13 (view) (permuted) | PERMUTE | Qcur-13 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 477 | cache_k_l13 (view) | VIEW | cache_k_l13{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 478 | cache_k_l13 (view) (permuted) | PERMUTE | cache_k_l13 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 479 | cache_v_l13 (view) | VIEW | cache_v_l13{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 480 | cache_v_l13 (view) (permuted) | PERMUTE | cache_v_l13 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 481 | node_480 | FLASH_ATTN_EXT | Qcur-13 (view) (permuted){128, 5, 32, 1} , cache_k_l13 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 482 | kqv_out-13 | RESHAPE | node_480{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 483 | node_482 | MUL_MAT | blk.13.attn_output.weight{4096, 4096, 1, 1} , kqv_out-13{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 484 | ffn_inp-13 | ADD | node_482{4096, 5, 1, 1} , l_out-12{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 485 | norm-13 | RMS_NORM | ffn_inp-13{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 486 | ffn_norm-13 | MUL | norm-13{4096, 5, 1, 1} , blk.13.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 487 | ffn_gate-13 | MUL_MAT | blk.13.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-13{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 488 | ffn_up-13 | MUL_MAT | blk.13.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-13{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 489 | ffn_swiglu-13 | SWIGLU | ffn_gate-13{12288, 5, 1, 1} , ffn_up-13{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 490 | ffn_out-13 | MUL_MAT | blk.13.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-13{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 491 | l_out-13 | ADD | ffn_out-13{4096, 5, 1, 1} , ffn_inp-13{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 492 | norm-14 | RMS_NORM | l_out-13{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 493 | attn_norm-14 | MUL | norm-14{4096, 5, 1, 1} , blk.14.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 494 | Qcur-14 | MUL_MAT | blk.14.attn_q.weight{4096, 4096, 1, 1} , attn_norm-14{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 495 | Qcur-14 | RESHAPE | Qcur-14{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 496 | norm-14 | RMS_NORM | Qcur-14{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 497 | Qcur_normed-14 | MUL | norm-14{128, 32, 5, 1} , blk.14.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 498 | Qcur-14 | ROPE | Qcur_normed-14{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 499 | Vcur-14 | MUL_MAT | blk.14.attn_v.weight{4096, 1024, 1, 1} , attn_norm-14{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 500 | Vcur-14 | RESHAPE | Vcur-14{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 501 | Kcur-14 | MUL_MAT | blk.14.attn_k.weight{4096, 1024, 1, 1} , attn_norm-14{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 502 | Kcur-14 | RESHAPE | Kcur-14{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 503 | norm-14 | RMS_NORM | Kcur-14{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 504 | Kcur_normed-14 | MUL | norm-14{128, 8, 5, 1} , blk.14.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 505 | Kcur-14 | ROPE | Kcur_normed-14{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 506 | Kcur-14 (view) | VIEW | Kcur-14{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 507 | cache_k_l14 (view) | SET_ROWS | Kcur-14 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 508 | Vcur-14 (view) | VIEW | Vcur-14{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 509 | cache_v_l14 (view) | SET_ROWS | Vcur-14 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 510 | Qcur-14 (view) | VIEW | Qcur-14{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 511 | Qcur-14 (view) (permuted) | PERMUTE | Qcur-14 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 512 | cache_k_l14 (view) | VIEW | cache_k_l14{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 513 | cache_k_l14 (view) (permuted) | PERMUTE | cache_k_l14 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 514 | cache_v_l14 (view) | VIEW | cache_v_l14{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 515 | cache_v_l14 (view) (permuted) | PERMUTE | cache_v_l14 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 516 | node_515 | FLASH_ATTN_EXT | Qcur-14 (view) (permuted){128, 5, 32, 1} , cache_k_l14 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 517 | kqv_out-14 | RESHAPE | node_515{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 518 | node_517 | MUL_MAT | blk.14.attn_output.weight{4096, 4096, 1, 1} , kqv_out-14{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 519 | ffn_inp-14 | ADD | node_517{4096, 5, 1, 1} , l_out-13{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 520 | norm-14 | RMS_NORM | ffn_inp-14{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 521 | ffn_norm-14 | MUL | norm-14{4096, 5, 1, 1} , blk.14.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 522 | ffn_gate-14 | MUL_MAT | blk.14.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-14{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 523 | ffn_up-14 | MUL_MAT | blk.14.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-14{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 524 | ffn_swiglu-14 | SWIGLU | ffn_gate-14{12288, 5, 1, 1} , ffn_up-14{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 525 | ffn_out-14 | MUL_MAT | blk.14.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-14{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 526 | l_out-14 | ADD | ffn_out-14{4096, 5, 1, 1} , ffn_inp-14{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 527 | norm-15 | RMS_NORM | l_out-14{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 528 | attn_norm-15 | MUL | norm-15{4096, 5, 1, 1} , blk.15.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 529 | Qcur-15 | MUL_MAT | blk.15.attn_q.weight{4096, 4096, 1, 1} , attn_norm-15{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 530 | Qcur-15 | RESHAPE | Qcur-15{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 531 | norm-15 | RMS_NORM | Qcur-15{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 532 | Qcur_normed-15 | MUL | norm-15{128, 32, 5, 1} , blk.15.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 533 | Qcur-15 | ROPE | Qcur_normed-15{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 534 | Vcur-15 | MUL_MAT | blk.15.attn_v.weight{4096, 1024, 1, 1} , attn_norm-15{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 535 | Vcur-15 | RESHAPE | Vcur-15{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 536 | Kcur-15 | MUL_MAT | blk.15.attn_k.weight{4096, 1024, 1, 1} , attn_norm-15{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 537 | Kcur-15 | RESHAPE | Kcur-15{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 538 | norm-15 | RMS_NORM | Kcur-15{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 539 | Kcur_normed-15 | MUL | norm-15{128, 8, 5, 1} , blk.15.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 540 | Kcur-15 | ROPE | Kcur_normed-15{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 541 | Kcur-15 (view) | VIEW | Kcur-15{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 542 | cache_k_l15 (view) | SET_ROWS | Kcur-15 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 543 | Vcur-15 (view) | VIEW | Vcur-15{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 544 | cache_v_l15 (view) | SET_ROWS | Vcur-15 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 545 | Qcur-15 (view) | VIEW | Qcur-15{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 546 | Qcur-15 (view) (permuted) | PERMUTE | Qcur-15 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 547 | cache_k_l15 (view) | VIEW | cache_k_l15{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 548 | cache_k_l15 (view) (permuted) | PERMUTE | cache_k_l15 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 549 | cache_v_l15 (view) | VIEW | cache_v_l15{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 550 | cache_v_l15 (view) (permuted) | PERMUTE | cache_v_l15 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 551 | node_550 | FLASH_ATTN_EXT | Qcur-15 (view) (permuted){128, 5, 32, 1} , cache_k_l15 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 552 | kqv_out-15 | RESHAPE | node_550{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 553 | node_552 | MUL_MAT | blk.15.attn_output.weight{4096, 4096, 1, 1} , kqv_out-15{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 554 | ffn_inp-15 | ADD | node_552{4096, 5, 1, 1} , l_out-14{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 555 | norm-15 | RMS_NORM | ffn_inp-15{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 556 | ffn_norm-15 | MUL | norm-15{4096, 5, 1, 1} , blk.15.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 557 | ffn_gate-15 | MUL_MAT | blk.15.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-15{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 558 | ffn_up-15 | MUL_MAT | blk.15.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-15{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 559 | ffn_swiglu-15 | SWIGLU | ffn_gate-15{12288, 5, 1, 1} , ffn_up-15{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 560 | ffn_out-15 | MUL_MAT | blk.15.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-15{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 561 | l_out-15 | ADD | ffn_out-15{4096, 5, 1, 1} , ffn_inp-15{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 562 | norm-16 | RMS_NORM | l_out-15{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 563 | attn_norm-16 | MUL | norm-16{4096, 5, 1, 1} , blk.16.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 564 | Qcur-16 | MUL_MAT | blk.16.attn_q.weight{4096, 4096, 1, 1} , attn_norm-16{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 565 | Qcur-16 | RESHAPE | Qcur-16{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 566 | norm-16 | RMS_NORM | Qcur-16{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 567 | Qcur_normed-16 | MUL | norm-16{128, 32, 5, 1} , blk.16.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 568 | Qcur-16 | ROPE | Qcur_normed-16{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 569 | Vcur-16 | MUL_MAT | blk.16.attn_v.weight{4096, 1024, 1, 1} , attn_norm-16{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 570 | Vcur-16 | RESHAPE | Vcur-16{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 571 | Kcur-16 | MUL_MAT | blk.16.attn_k.weight{4096, 1024, 1, 1} , attn_norm-16{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 572 | Kcur-16 | RESHAPE | Kcur-16{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 573 | norm-16 | RMS_NORM | Kcur-16{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 574 | Kcur_normed-16 | MUL | norm-16{128, 8, 5, 1} , blk.16.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 575 | Kcur-16 | ROPE | Kcur_normed-16{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 576 | Kcur-16 (view) | VIEW | Kcur-16{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 577 | cache_k_l16 (view) | SET_ROWS | Kcur-16 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 578 | Vcur-16 (view) | VIEW | Vcur-16{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 579 | cache_v_l16 (view) | SET_ROWS | Vcur-16 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 580 | Qcur-16 (view) | VIEW | Qcur-16{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 581 | Qcur-16 (view) (permuted) | PERMUTE | Qcur-16 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 582 | cache_k_l16 (view) | VIEW | cache_k_l16{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 583 | cache_k_l16 (view) (permuted) | PERMUTE | cache_k_l16 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 584 | cache_v_l16 (view) | VIEW | cache_v_l16{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 585 | cache_v_l16 (view) (permuted) | PERMUTE | cache_v_l16 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 586 | node_585 | FLASH_ATTN_EXT | Qcur-16 (view) (permuted){128, 5, 32, 1} , cache_k_l16 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 587 | kqv_out-16 | RESHAPE | node_585{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 588 | node_587 | MUL_MAT | blk.16.attn_output.weight{4096, 4096, 1, 1} , kqv_out-16{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 589 | ffn_inp-16 | ADD | node_587{4096, 5, 1, 1} , l_out-15{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 590 | norm-16 | RMS_NORM | ffn_inp-16{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 591 | ffn_norm-16 | MUL | norm-16{4096, 5, 1, 1} , blk.16.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 592 | ffn_gate-16 | MUL_MAT | blk.16.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-16{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 593 | ffn_up-16 | MUL_MAT | blk.16.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-16{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 594 | ffn_swiglu-16 | SWIGLU | ffn_gate-16{12288, 5, 1, 1} , ffn_up-16{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 595 | ffn_out-16 | MUL_MAT | blk.16.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-16{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 596 | l_out-16 | ADD | ffn_out-16{4096, 5, 1, 1} , ffn_inp-16{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 597 | norm-17 | RMS_NORM | l_out-16{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 598 | attn_norm-17 | MUL | norm-17{4096, 5, 1, 1} , blk.17.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 599 | Qcur-17 | MUL_MAT | blk.17.attn_q.weight{4096, 4096, 1, 1} , attn_norm-17{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 600 | Qcur-17 | RESHAPE | Qcur-17{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 601 | norm-17 | RMS_NORM | Qcur-17{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 602 | Qcur_normed-17 | MUL | norm-17{128, 32, 5, 1} , blk.17.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 603 | Qcur-17 | ROPE | Qcur_normed-17{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 604 | Vcur-17 | MUL_MAT | blk.17.attn_v.weight{4096, 1024, 1, 1} , attn_norm-17{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 605 | Vcur-17 | RESHAPE | Vcur-17{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 606 | Kcur-17 | MUL_MAT | blk.17.attn_k.weight{4096, 1024, 1, 1} , attn_norm-17{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 607 | Kcur-17 | RESHAPE | Kcur-17{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 608 | norm-17 | RMS_NORM | Kcur-17{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 609 | Kcur_normed-17 | MUL | norm-17{128, 8, 5, 1} , blk.17.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 610 | Kcur-17 | ROPE | Kcur_normed-17{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 611 | Kcur-17 (view) | VIEW | Kcur-17{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 612 | cache_k_l17 (view) | SET_ROWS | Kcur-17 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 613 | Vcur-17 (view) | VIEW | Vcur-17{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 614 | cache_v_l17 (view) | SET_ROWS | Vcur-17 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 615 | Qcur-17 (view) | VIEW | Qcur-17{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 616 | Qcur-17 (view) (permuted) | PERMUTE | Qcur-17 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 617 | cache_k_l17 (view) | VIEW | cache_k_l17{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 618 | cache_k_l17 (view) (permuted) | PERMUTE | cache_k_l17 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 619 | cache_v_l17 (view) | VIEW | cache_v_l17{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 620 | cache_v_l17 (view) (permuted) | PERMUTE | cache_v_l17 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 621 | node_620 | FLASH_ATTN_EXT | Qcur-17 (view) (permuted){128, 5, 32, 1} , cache_k_l17 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 622 | kqv_out-17 | RESHAPE | node_620{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 623 | node_622 | MUL_MAT | blk.17.attn_output.weight{4096, 4096, 1, 1} , kqv_out-17{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 624 | ffn_inp-17 | ADD | node_622{4096, 5, 1, 1} , l_out-16{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 625 | norm-17 | RMS_NORM | ffn_inp-17{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 626 | ffn_norm-17 | MUL | norm-17{4096, 5, 1, 1} , blk.17.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 627 | ffn_gate-17 | MUL_MAT | blk.17.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-17{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 628 | ffn_up-17 | MUL_MAT | blk.17.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-17{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 629 | ffn_swiglu-17 | SWIGLU | ffn_gate-17{12288, 5, 1, 1} , ffn_up-17{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 630 | ffn_out-17 | MUL_MAT | blk.17.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-17{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 631 | l_out-17 | ADD | ffn_out-17{4096, 5, 1, 1} , ffn_inp-17{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 632 | norm-18 | RMS_NORM | l_out-17{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 633 | attn_norm-18 | MUL | norm-18{4096, 5, 1, 1} , blk.18.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 634 | Qcur-18 | MUL_MAT | blk.18.attn_q.weight{4096, 4096, 1, 1} , attn_norm-18{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 635 | Qcur-18 | RESHAPE | Qcur-18{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 636 | norm-18 | RMS_NORM | Qcur-18{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 637 | Qcur_normed-18 | MUL | norm-18{128, 32, 5, 1} , blk.18.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 638 | Qcur-18 | ROPE | Qcur_normed-18{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 639 | Vcur-18 | MUL_MAT | blk.18.attn_v.weight{4096, 1024, 1, 1} , attn_norm-18{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 640 | Vcur-18 | RESHAPE | Vcur-18{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 641 | Kcur-18 | MUL_MAT | blk.18.attn_k.weight{4096, 1024, 1, 1} , attn_norm-18{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 642 | Kcur-18 | RESHAPE | Kcur-18{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 643 | norm-18 | RMS_NORM | Kcur-18{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 644 | Kcur_normed-18 | MUL | norm-18{128, 8, 5, 1} , blk.18.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 645 | Kcur-18 | ROPE | Kcur_normed-18{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 646 | Kcur-18 (view) | VIEW | Kcur-18{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 647 | cache_k_l18 (view) | SET_ROWS | Kcur-18 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 648 | Vcur-18 (view) | VIEW | Vcur-18{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 649 | cache_v_l18 (view) | SET_ROWS | Vcur-18 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 650 | Qcur-18 (view) | VIEW | Qcur-18{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 651 | Qcur-18 (view) (permuted) | PERMUTE | Qcur-18 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 652 | cache_k_l18 (view) | VIEW | cache_k_l18{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 653 | cache_k_l18 (view) (permuted) | PERMUTE | cache_k_l18 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 654 | cache_v_l18 (view) | VIEW | cache_v_l18{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 655 | cache_v_l18 (view) (permuted) | PERMUTE | cache_v_l18 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 656 | node_655 | FLASH_ATTN_EXT | Qcur-18 (view) (permuted){128, 5, 32, 1} , cache_k_l18 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 657 | kqv_out-18 | RESHAPE | node_655{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 658 | node_657 | MUL_MAT | blk.18.attn_output.weight{4096, 4096, 1, 1} , kqv_out-18{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 659 | ffn_inp-18 | ADD | node_657{4096, 5, 1, 1} , l_out-17{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 660 | norm-18 | RMS_NORM | ffn_inp-18{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 661 | ffn_norm-18 | MUL | norm-18{4096, 5, 1, 1} , blk.18.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 662 | ffn_gate-18 | MUL_MAT | blk.18.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-18{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 663 | ffn_up-18 | MUL_MAT | blk.18.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-18{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 664 | ffn_swiglu-18 | SWIGLU | ffn_gate-18{12288, 5, 1, 1} , ffn_up-18{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 665 | ffn_out-18 | MUL_MAT | blk.18.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-18{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 666 | l_out-18 | ADD | ffn_out-18{4096, 5, 1, 1} , ffn_inp-18{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 667 | norm-19 | RMS_NORM | l_out-18{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 668 | attn_norm-19 | MUL | norm-19{4096, 5, 1, 1} , blk.19.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 669 | Qcur-19 | MUL_MAT | blk.19.attn_q.weight{4096, 4096, 1, 1} , attn_norm-19{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 670 | Qcur-19 | RESHAPE | Qcur-19{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 671 | norm-19 | RMS_NORM | Qcur-19{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 672 | Qcur_normed-19 | MUL | norm-19{128, 32, 5, 1} , blk.19.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 673 | Qcur-19 | ROPE | Qcur_normed-19{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 674 | Vcur-19 | MUL_MAT | blk.19.attn_v.weight{4096, 1024, 1, 1} , attn_norm-19{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 675 | Vcur-19 | RESHAPE | Vcur-19{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 676 | Kcur-19 | MUL_MAT | blk.19.attn_k.weight{4096, 1024, 1, 1} , attn_norm-19{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 677 | Kcur-19 | RESHAPE | Kcur-19{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 678 | norm-19 | RMS_NORM | Kcur-19{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 679 | Kcur_normed-19 | MUL | norm-19{128, 8, 5, 1} , blk.19.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 680 | Kcur-19 | ROPE | Kcur_normed-19{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 681 | Kcur-19 (view) | VIEW | Kcur-19{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 682 | cache_k_l19 (view) | SET_ROWS | Kcur-19 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 683 | Vcur-19 (view) | VIEW | Vcur-19{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 684 | cache_v_l19 (view) | SET_ROWS | Vcur-19 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 685 | Qcur-19 (view) | VIEW | Qcur-19{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 686 | Qcur-19 (view) (permuted) | PERMUTE | Qcur-19 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 687 | cache_k_l19 (view) | VIEW | cache_k_l19{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 688 | cache_k_l19 (view) (permuted) | PERMUTE | cache_k_l19 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 689 | cache_v_l19 (view) | VIEW | cache_v_l19{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 690 | cache_v_l19 (view) (permuted) | PERMUTE | cache_v_l19 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 691 | node_690 | FLASH_ATTN_EXT | Qcur-19 (view) (permuted){128, 5, 32, 1} , cache_k_l19 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 692 | kqv_out-19 | RESHAPE | node_690{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 693 | node_692 | MUL_MAT | blk.19.attn_output.weight{4096, 4096, 1, 1} , kqv_out-19{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 694 | ffn_inp-19 | ADD | node_692{4096, 5, 1, 1} , l_out-18{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 695 | norm-19 | RMS_NORM | ffn_inp-19{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 696 | ffn_norm-19 | MUL | norm-19{4096, 5, 1, 1} , blk.19.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 697 | ffn_gate-19 | MUL_MAT | blk.19.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-19{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 698 | ffn_up-19 | MUL_MAT | blk.19.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-19{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 699 | ffn_swiglu-19 | SWIGLU | ffn_gate-19{12288, 5, 1, 1} , ffn_up-19{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 700 | ffn_out-19 | MUL_MAT | blk.19.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-19{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 701 | l_out-19 | ADD | ffn_out-19{4096, 5, 1, 1} , ffn_inp-19{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 702 | norm-20 | RMS_NORM | l_out-19{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 703 | attn_norm-20 | MUL | norm-20{4096, 5, 1, 1} , blk.20.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 704 | Qcur-20 | MUL_MAT | blk.20.attn_q.weight{4096, 4096, 1, 1} , attn_norm-20{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 705 | Qcur-20 | RESHAPE | Qcur-20{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 706 | norm-20 | RMS_NORM | Qcur-20{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 707 | Qcur_normed-20 | MUL | norm-20{128, 32, 5, 1} , blk.20.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 708 | Qcur-20 | ROPE | Qcur_normed-20{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 709 | Vcur-20 | MUL_MAT | blk.20.attn_v.weight{4096, 1024, 1, 1} , attn_norm-20{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 710 | Vcur-20 | RESHAPE | Vcur-20{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 711 | Kcur-20 | MUL_MAT | blk.20.attn_k.weight{4096, 1024, 1, 1} , attn_norm-20{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 712 | Kcur-20 | RESHAPE | Kcur-20{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 713 | norm-20 | RMS_NORM | Kcur-20{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 714 | Kcur_normed-20 | MUL | norm-20{128, 8, 5, 1} , blk.20.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 715 | Kcur-20 | ROPE | Kcur_normed-20{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 716 | Kcur-20 (view) | VIEW | Kcur-20{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 717 | cache_k_l20 (view) | SET_ROWS | Kcur-20 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 718 | Vcur-20 (view) | VIEW | Vcur-20{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 719 | cache_v_l20 (view) | SET_ROWS | Vcur-20 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 720 | Qcur-20 (view) | VIEW | Qcur-20{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 721 | Qcur-20 (view) (permuted) | PERMUTE | Qcur-20 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 722 | cache_k_l20 (view) | VIEW | cache_k_l20{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 723 | cache_k_l20 (view) (permuted) | PERMUTE | cache_k_l20 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 724 | cache_v_l20 (view) | VIEW | cache_v_l20{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 725 | cache_v_l20 (view) (permuted) | PERMUTE | cache_v_l20 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 726 | node_725 | FLASH_ATTN_EXT | Qcur-20 (view) (permuted){128, 5, 32, 1} , cache_k_l20 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 727 | kqv_out-20 | RESHAPE | node_725{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 728 | node_727 | MUL_MAT | blk.20.attn_output.weight{4096, 4096, 1, 1} , kqv_out-20{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 729 | ffn_inp-20 | ADD | node_727{4096, 5, 1, 1} , l_out-19{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 730 | norm-20 | RMS_NORM | ffn_inp-20{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 731 | ffn_norm-20 | MUL | norm-20{4096, 5, 1, 1} , blk.20.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 732 | ffn_gate-20 | MUL_MAT | blk.20.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-20{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 733 | ffn_up-20 | MUL_MAT | blk.20.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-20{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 734 | ffn_swiglu-20 | SWIGLU | ffn_gate-20{12288, 5, 1, 1} , ffn_up-20{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 735 | ffn_out-20 | MUL_MAT | blk.20.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-20{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 736 | l_out-20 | ADD | ffn_out-20{4096, 5, 1, 1} , ffn_inp-20{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 737 | norm-21 | RMS_NORM | l_out-20{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 738 | attn_norm-21 | MUL | norm-21{4096, 5, 1, 1} , blk.21.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 739 | Qcur-21 | MUL_MAT | blk.21.attn_q.weight{4096, 4096, 1, 1} , attn_norm-21{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 740 | Qcur-21 | RESHAPE | Qcur-21{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 741 | norm-21 | RMS_NORM | Qcur-21{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 742 | Qcur_normed-21 | MUL | norm-21{128, 32, 5, 1} , blk.21.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 743 | Qcur-21 | ROPE | Qcur_normed-21{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 744 | Vcur-21 | MUL_MAT | blk.21.attn_v.weight{4096, 1024, 1, 1} , attn_norm-21{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 745 | Vcur-21 | RESHAPE | Vcur-21{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 746 | Kcur-21 | MUL_MAT | blk.21.attn_k.weight{4096, 1024, 1, 1} , attn_norm-21{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 747 | Kcur-21 | RESHAPE | Kcur-21{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 748 | norm-21 | RMS_NORM | Kcur-21{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 749 | Kcur_normed-21 | MUL | norm-21{128, 8, 5, 1} , blk.21.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 750 | Kcur-21 | ROPE | Kcur_normed-21{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 751 | Kcur-21 (view) | VIEW | Kcur-21{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 752 | cache_k_l21 (view) | SET_ROWS | Kcur-21 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 753 | Vcur-21 (view) | VIEW | Vcur-21{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 754 | cache_v_l21 (view) | SET_ROWS | Vcur-21 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 755 | Qcur-21 (view) | VIEW | Qcur-21{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 756 | Qcur-21 (view) (permuted) | PERMUTE | Qcur-21 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 757 | cache_k_l21 (view) | VIEW | cache_k_l21{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 758 | cache_k_l21 (view) (permuted) | PERMUTE | cache_k_l21 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 759 | cache_v_l21 (view) | VIEW | cache_v_l21{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 760 | cache_v_l21 (view) (permuted) | PERMUTE | cache_v_l21 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 761 | node_760 | FLASH_ATTN_EXT | Qcur-21 (view) (permuted){128, 5, 32, 1} , cache_k_l21 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 762 | kqv_out-21 | RESHAPE | node_760{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 763 | node_762 | MUL_MAT | blk.21.attn_output.weight{4096, 4096, 1, 1} , kqv_out-21{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 764 | ffn_inp-21 | ADD | node_762{4096, 5, 1, 1} , l_out-20{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 765 | norm-21 | RMS_NORM | ffn_inp-21{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 766 | ffn_norm-21 | MUL | norm-21{4096, 5, 1, 1} , blk.21.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 767 | ffn_gate-21 | MUL_MAT | blk.21.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-21{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 768 | ffn_up-21 | MUL_MAT | blk.21.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-21{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 769 | ffn_swiglu-21 | SWIGLU | ffn_gate-21{12288, 5, 1, 1} , ffn_up-21{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 770 | ffn_out-21 | MUL_MAT | blk.21.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-21{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 771 | l_out-21 | ADD | ffn_out-21{4096, 5, 1, 1} , ffn_inp-21{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 772 | norm-22 | RMS_NORM | l_out-21{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 773 | attn_norm-22 | MUL | norm-22{4096, 5, 1, 1} , blk.22.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 774 | Qcur-22 | MUL_MAT | blk.22.attn_q.weight{4096, 4096, 1, 1} , attn_norm-22{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 775 | Qcur-22 | RESHAPE | Qcur-22{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 776 | norm-22 | RMS_NORM | Qcur-22{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 777 | Qcur_normed-22 | MUL | norm-22{128, 32, 5, 1} , blk.22.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 778 | Qcur-22 | ROPE | Qcur_normed-22{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 779 | Vcur-22 | MUL_MAT | blk.22.attn_v.weight{4096, 1024, 1, 1} , attn_norm-22{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 780 | Vcur-22 | RESHAPE | Vcur-22{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 781 | Kcur-22 | MUL_MAT | blk.22.attn_k.weight{4096, 1024, 1, 1} , attn_norm-22{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 782 | Kcur-22 | RESHAPE | Kcur-22{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 783 | norm-22 | RMS_NORM | Kcur-22{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 784 | Kcur_normed-22 | MUL | norm-22{128, 8, 5, 1} , blk.22.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 785 | Kcur-22 | ROPE | Kcur_normed-22{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 786 | Kcur-22 (view) | VIEW | Kcur-22{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 787 | cache_k_l22 (view) | SET_ROWS | Kcur-22 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 788 | Vcur-22 (view) | VIEW | Vcur-22{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 789 | cache_v_l22 (view) | SET_ROWS | Vcur-22 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 790 | Qcur-22 (view) | VIEW | Qcur-22{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 791 | Qcur-22 (view) (permuted) | PERMUTE | Qcur-22 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 792 | cache_k_l22 (view) | VIEW | cache_k_l22{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 793 | cache_k_l22 (view) (permuted) | PERMUTE | cache_k_l22 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 794 | cache_v_l22 (view) | VIEW | cache_v_l22{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 795 | cache_v_l22 (view) (permuted) | PERMUTE | cache_v_l22 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 796 | node_795 | FLASH_ATTN_EXT | Qcur-22 (view) (permuted){128, 5, 32, 1} , cache_k_l22 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 797 | kqv_out-22 | RESHAPE | node_795{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 798 | node_797 | MUL_MAT | blk.22.attn_output.weight{4096, 4096, 1, 1} , kqv_out-22{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 799 | ffn_inp-22 | ADD | node_797{4096, 5, 1, 1} , l_out-21{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 800 | norm-22 | RMS_NORM | ffn_inp-22{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 801 | ffn_norm-22 | MUL | norm-22{4096, 5, 1, 1} , blk.22.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 802 | ffn_gate-22 | MUL_MAT | blk.22.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-22{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 803 | ffn_up-22 | MUL_MAT | blk.22.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-22{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 804 | ffn_swiglu-22 | SWIGLU | ffn_gate-22{12288, 5, 1, 1} , ffn_up-22{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 805 | ffn_out-22 | MUL_MAT | blk.22.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-22{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 806 | l_out-22 | ADD | ffn_out-22{4096, 5, 1, 1} , ffn_inp-22{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 807 | norm-23 | RMS_NORM | l_out-22{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 808 | attn_norm-23 | MUL | norm-23{4096, 5, 1, 1} , blk.23.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 809 | Qcur-23 | MUL_MAT | blk.23.attn_q.weight{4096, 4096, 1, 1} , attn_norm-23{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 810 | Qcur-23 | RESHAPE | Qcur-23{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 811 | norm-23 | RMS_NORM | Qcur-23{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 812 | Qcur_normed-23 | MUL | norm-23{128, 32, 5, 1} , blk.23.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 813 | Qcur-23 | ROPE | Qcur_normed-23{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 814 | Vcur-23 | MUL_MAT | blk.23.attn_v.weight{4096, 1024, 1, 1} , attn_norm-23{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 815 | Vcur-23 | RESHAPE | Vcur-23{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 816 | Kcur-23 | MUL_MAT | blk.23.attn_k.weight{4096, 1024, 1, 1} , attn_norm-23{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 817 | Kcur-23 | RESHAPE | Kcur-23{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 818 | norm-23 | RMS_NORM | Kcur-23{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 819 | Kcur_normed-23 | MUL | norm-23{128, 8, 5, 1} , blk.23.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 820 | Kcur-23 | ROPE | Kcur_normed-23{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 821 | Kcur-23 (view) | VIEW | Kcur-23{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 822 | cache_k_l23 (view) | SET_ROWS | Kcur-23 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 823 | Vcur-23 (view) | VIEW | Vcur-23{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 824 | cache_v_l23 (view) | SET_ROWS | Vcur-23 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 825 | Qcur-23 (view) | VIEW | Qcur-23{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 826 | Qcur-23 (view) (permuted) | PERMUTE | Qcur-23 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 827 | cache_k_l23 (view) | VIEW | cache_k_l23{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 828 | cache_k_l23 (view) (permuted) | PERMUTE | cache_k_l23 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 829 | cache_v_l23 (view) | VIEW | cache_v_l23{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 830 | cache_v_l23 (view) (permuted) | PERMUTE | cache_v_l23 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 831 | node_830 | FLASH_ATTN_EXT | Qcur-23 (view) (permuted){128, 5, 32, 1} , cache_k_l23 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 832 | kqv_out-23 | RESHAPE | node_830{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 833 | node_832 | MUL_MAT | blk.23.attn_output.weight{4096, 4096, 1, 1} , kqv_out-23{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 834 | ffn_inp-23 | ADD | node_832{4096, 5, 1, 1} , l_out-22{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 835 | norm-23 | RMS_NORM | ffn_inp-23{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 836 | ffn_norm-23 | MUL | norm-23{4096, 5, 1, 1} , blk.23.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 837 | ffn_gate-23 | MUL_MAT | blk.23.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-23{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 838 | ffn_up-23 | MUL_MAT | blk.23.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-23{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 839 | ffn_swiglu-23 | SWIGLU | ffn_gate-23{12288, 5, 1, 1} , ffn_up-23{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 840 | ffn_out-23 | MUL_MAT | blk.23.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-23{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 841 | l_out-23 | ADD | ffn_out-23{4096, 5, 1, 1} , ffn_inp-23{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 842 | norm-24 | RMS_NORM | l_out-23{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 843 | attn_norm-24 | MUL | norm-24{4096, 5, 1, 1} , blk.24.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 844 | Qcur-24 | MUL_MAT | blk.24.attn_q.weight{4096, 4096, 1, 1} , attn_norm-24{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 845 | Qcur-24 | RESHAPE | Qcur-24{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 846 | norm-24 | RMS_NORM | Qcur-24{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 847 | Qcur_normed-24 | MUL | norm-24{128, 32, 5, 1} , blk.24.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 848 | Qcur-24 | ROPE | Qcur_normed-24{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 849 | Vcur-24 | MUL_MAT | blk.24.attn_v.weight{4096, 1024, 1, 1} , attn_norm-24{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 850 | Vcur-24 | RESHAPE | Vcur-24{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 851 | Kcur-24 | MUL_MAT | blk.24.attn_k.weight{4096, 1024, 1, 1} , attn_norm-24{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 852 | Kcur-24 | RESHAPE | Kcur-24{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 853 | norm-24 | RMS_NORM | Kcur-24{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 854 | Kcur_normed-24 | MUL | norm-24{128, 8, 5, 1} , blk.24.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 855 | Kcur-24 | ROPE | Kcur_normed-24{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 856 | Kcur-24 (view) | VIEW | Kcur-24{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 857 | cache_k_l24 (view) | SET_ROWS | Kcur-24 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 858 | Vcur-24 (view) | VIEW | Vcur-24{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 859 | cache_v_l24 (view) | SET_ROWS | Vcur-24 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 860 | Qcur-24 (view) | VIEW | Qcur-24{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 861 | Qcur-24 (view) (permuted) | PERMUTE | Qcur-24 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 862 | cache_k_l24 (view) | VIEW | cache_k_l24{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 863 | cache_k_l24 (view) (permuted) | PERMUTE | cache_k_l24 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 864 | cache_v_l24 (view) | VIEW | cache_v_l24{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 865 | cache_v_l24 (view) (permuted) | PERMUTE | cache_v_l24 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 866 | node_865 | FLASH_ATTN_EXT | Qcur-24 (view) (permuted){128, 5, 32, 1} , cache_k_l24 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 867 | kqv_out-24 | RESHAPE | node_865{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 868 | node_867 | MUL_MAT | blk.24.attn_output.weight{4096, 4096, 1, 1} , kqv_out-24{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 869 | ffn_inp-24 | ADD | node_867{4096, 5, 1, 1} , l_out-23{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 870 | norm-24 | RMS_NORM | ffn_inp-24{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 871 | ffn_norm-24 | MUL | norm-24{4096, 5, 1, 1} , blk.24.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 872 | ffn_gate-24 | MUL_MAT | blk.24.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-24{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 873 | ffn_up-24 | MUL_MAT | blk.24.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-24{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 874 | ffn_swiglu-24 | SWIGLU | ffn_gate-24{12288, 5, 1, 1} , ffn_up-24{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 875 | ffn_out-24 | MUL_MAT | blk.24.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-24{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 876 | l_out-24 | ADD | ffn_out-24{4096, 5, 1, 1} , ffn_inp-24{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 877 | norm-25 | RMS_NORM | l_out-24{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 878 | attn_norm-25 | MUL | norm-25{4096, 5, 1, 1} , blk.25.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 879 | Qcur-25 | MUL_MAT | blk.25.attn_q.weight{4096, 4096, 1, 1} , attn_norm-25{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 880 | Qcur-25 | RESHAPE | Qcur-25{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 881 | norm-25 | RMS_NORM | Qcur-25{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 882 | Qcur_normed-25 | MUL | norm-25{128, 32, 5, 1} , blk.25.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 883 | Qcur-25 | ROPE | Qcur_normed-25{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 884 | Vcur-25 | MUL_MAT | blk.25.attn_v.weight{4096, 1024, 1, 1} , attn_norm-25{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 885 | Vcur-25 | RESHAPE | Vcur-25{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 886 | Kcur-25 | MUL_MAT | blk.25.attn_k.weight{4096, 1024, 1, 1} , attn_norm-25{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 887 | Kcur-25 | RESHAPE | Kcur-25{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 888 | norm-25 | RMS_NORM | Kcur-25{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 889 | Kcur_normed-25 | MUL | norm-25{128, 8, 5, 1} , blk.25.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 890 | Kcur-25 | ROPE | Kcur_normed-25{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 891 | Kcur-25 (view) | VIEW | Kcur-25{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 892 | cache_k_l25 (view) | SET_ROWS | Kcur-25 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 893 | Vcur-25 (view) | VIEW | Vcur-25{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 894 | cache_v_l25 (view) | SET_ROWS | Vcur-25 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 895 | Qcur-25 (view) | VIEW | Qcur-25{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 896 | Qcur-25 (view) (permuted) | PERMUTE | Qcur-25 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 897 | cache_k_l25 (view) | VIEW | cache_k_l25{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 898 | cache_k_l25 (view) (permuted) | PERMUTE | cache_k_l25 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 899 | cache_v_l25 (view) | VIEW | cache_v_l25{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 900 | cache_v_l25 (view) (permuted) | PERMUTE | cache_v_l25 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 901 | node_900 | FLASH_ATTN_EXT | Qcur-25 (view) (permuted){128, 5, 32, 1} , cache_k_l25 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 902 | kqv_out-25 | RESHAPE | node_900{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 903 | node_902 | MUL_MAT | blk.25.attn_output.weight{4096, 4096, 1, 1} , kqv_out-25{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 904 | ffn_inp-25 | ADD | node_902{4096, 5, 1, 1} , l_out-24{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 905 | norm-25 | RMS_NORM | ffn_inp-25{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 906 | ffn_norm-25 | MUL | norm-25{4096, 5, 1, 1} , blk.25.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 907 | ffn_gate-25 | MUL_MAT | blk.25.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-25{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 908 | ffn_up-25 | MUL_MAT | blk.25.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-25{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 909 | ffn_swiglu-25 | SWIGLU | ffn_gate-25{12288, 5, 1, 1} , ffn_up-25{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 910 | ffn_out-25 | MUL_MAT | blk.25.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-25{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 911 | l_out-25 | ADD | ffn_out-25{4096, 5, 1, 1} , ffn_inp-25{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 912 | norm-26 | RMS_NORM | l_out-25{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 913 | attn_norm-26 | MUL | norm-26{4096, 5, 1, 1} , blk.26.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 914 | Qcur-26 | MUL_MAT | blk.26.attn_q.weight{4096, 4096, 1, 1} , attn_norm-26{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 915 | Qcur-26 | RESHAPE | Qcur-26{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 916 | norm-26 | RMS_NORM | Qcur-26{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 917 | Qcur_normed-26 | MUL | norm-26{128, 32, 5, 1} , blk.26.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 918 | Qcur-26 | ROPE | Qcur_normed-26{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 919 | Vcur-26 | MUL_MAT | blk.26.attn_v.weight{4096, 1024, 1, 1} , attn_norm-26{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 920 | Vcur-26 | RESHAPE | Vcur-26{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 921 | Kcur-26 | MUL_MAT | blk.26.attn_k.weight{4096, 1024, 1, 1} , attn_norm-26{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 922 | Kcur-26 | RESHAPE | Kcur-26{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 923 | norm-26 | RMS_NORM | Kcur-26{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 924 | Kcur_normed-26 | MUL | norm-26{128, 8, 5, 1} , blk.26.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 925 | Kcur-26 | ROPE | Kcur_normed-26{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 926 | Kcur-26 (view) | VIEW | Kcur-26{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 927 | cache_k_l26 (view) | SET_ROWS | Kcur-26 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 928 | Vcur-26 (view) | VIEW | Vcur-26{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 929 | cache_v_l26 (view) | SET_ROWS | Vcur-26 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 930 | Qcur-26 (view) | VIEW | Qcur-26{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 931 | Qcur-26 (view) (permuted) | PERMUTE | Qcur-26 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 932 | cache_k_l26 (view) | VIEW | cache_k_l26{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 933 | cache_k_l26 (view) (permuted) | PERMUTE | cache_k_l26 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 934 | cache_v_l26 (view) | VIEW | cache_v_l26{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 935 | cache_v_l26 (view) (permuted) | PERMUTE | cache_v_l26 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 936 | node_935 | FLASH_ATTN_EXT | Qcur-26 (view) (permuted){128, 5, 32, 1} , cache_k_l26 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 937 | kqv_out-26 | RESHAPE | node_935{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 938 | node_937 | MUL_MAT | blk.26.attn_output.weight{4096, 4096, 1, 1} , kqv_out-26{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 939 | ffn_inp-26 | ADD | node_937{4096, 5, 1, 1} , l_out-25{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 940 | norm-26 | RMS_NORM | ffn_inp-26{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 941 | ffn_norm-26 | MUL | norm-26{4096, 5, 1, 1} , blk.26.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 942 | ffn_gate-26 | MUL_MAT | blk.26.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-26{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 943 | ffn_up-26 | MUL_MAT | blk.26.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-26{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 944 | ffn_swiglu-26 | SWIGLU | ffn_gate-26{12288, 5, 1, 1} , ffn_up-26{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 945 | ffn_out-26 | MUL_MAT | blk.26.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-26{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 946 | l_out-26 | ADD | ffn_out-26{4096, 5, 1, 1} , ffn_inp-26{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 947 | norm-27 | RMS_NORM | l_out-26{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 948 | attn_norm-27 | MUL | norm-27{4096, 5, 1, 1} , blk.27.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 949 | Qcur-27 | MUL_MAT | blk.27.attn_q.weight{4096, 4096, 1, 1} , attn_norm-27{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 950 | Qcur-27 | RESHAPE | Qcur-27{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 951 | norm-27 | RMS_NORM | Qcur-27{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 952 | Qcur_normed-27 | MUL | norm-27{128, 32, 5, 1} , blk.27.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 953 | Qcur-27 | ROPE | Qcur_normed-27{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 954 | Vcur-27 | MUL_MAT | blk.27.attn_v.weight{4096, 1024, 1, 1} , attn_norm-27{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 955 | Vcur-27 | RESHAPE | Vcur-27{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 956 | Kcur-27 | MUL_MAT | blk.27.attn_k.weight{4096, 1024, 1, 1} , attn_norm-27{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 957 | Kcur-27 | RESHAPE | Kcur-27{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 958 | norm-27 | RMS_NORM | Kcur-27{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 959 | Kcur_normed-27 | MUL | norm-27{128, 8, 5, 1} , blk.27.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 960 | Kcur-27 | ROPE | Kcur_normed-27{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 961 | Kcur-27 (view) | VIEW | Kcur-27{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 962 | cache_k_l27 (view) | SET_ROWS | Kcur-27 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 963 | Vcur-27 (view) | VIEW | Vcur-27{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 964 | cache_v_l27 (view) | SET_ROWS | Vcur-27 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 965 | Qcur-27 (view) | VIEW | Qcur-27{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 966 | Qcur-27 (view) (permuted) | PERMUTE | Qcur-27 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 967 | cache_k_l27 (view) | VIEW | cache_k_l27{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 968 | cache_k_l27 (view) (permuted) | PERMUTE | cache_k_l27 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 969 | cache_v_l27 (view) | VIEW | cache_v_l27{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 970 | cache_v_l27 (view) (permuted) | PERMUTE | cache_v_l27 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 971 | node_970 | FLASH_ATTN_EXT | Qcur-27 (view) (permuted){128, 5, 32, 1} , cache_k_l27 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 972 | kqv_out-27 | RESHAPE | node_970{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 973 | node_972 | MUL_MAT | blk.27.attn_output.weight{4096, 4096, 1, 1} , kqv_out-27{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 974 | ffn_inp-27 | ADD | node_972{4096, 5, 1, 1} , l_out-26{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 975 | norm-27 | RMS_NORM | ffn_inp-27{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 976 | ffn_norm-27 | MUL | norm-27{4096, 5, 1, 1} , blk.27.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 977 | ffn_gate-27 | MUL_MAT | blk.27.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-27{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 978 | ffn_up-27 | MUL_MAT | blk.27.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-27{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 979 | ffn_swiglu-27 | SWIGLU | ffn_gate-27{12288, 5, 1, 1} , ffn_up-27{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 980 | ffn_out-27 | MUL_MAT | blk.27.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-27{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 981 | l_out-27 | ADD | ffn_out-27{4096, 5, 1, 1} , ffn_inp-27{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 982 | norm-28 | RMS_NORM | l_out-27{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 983 | attn_norm-28 | MUL | norm-28{4096, 5, 1, 1} , blk.28.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 984 | Qcur-28 | MUL_MAT | blk.28.attn_q.weight{4096, 4096, 1, 1} , attn_norm-28{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 985 | Qcur-28 | RESHAPE | Qcur-28{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 986 | norm-28 | RMS_NORM | Qcur-28{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 987 | Qcur_normed-28 | MUL | norm-28{128, 32, 5, 1} , blk.28.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 988 | Qcur-28 | ROPE | Qcur_normed-28{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 989 | Vcur-28 | MUL_MAT | blk.28.attn_v.weight{4096, 1024, 1, 1} , attn_norm-28{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 990 | Vcur-28 | RESHAPE | Vcur-28{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 991 | Kcur-28 | MUL_MAT | blk.28.attn_k.weight{4096, 1024, 1, 1} , attn_norm-28{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 992 | Kcur-28 | RESHAPE | Kcur-28{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 993 | norm-28 | RMS_NORM | Kcur-28{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 994 | Kcur_normed-28 | MUL | norm-28{128, 8, 5, 1} , blk.28.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 995 | Kcur-28 | ROPE | Kcur_normed-28{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 996 | Kcur-28 (view) | VIEW | Kcur-28{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 997 | cache_k_l28 (view) | SET_ROWS | Kcur-28 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 998 | Vcur-28 (view) | VIEW | Vcur-28{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 999 | cache_v_l28 (view) | SET_ROWS | Vcur-28 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1000 | Qcur-28 (view) | VIEW | Qcur-28{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1001 | Qcur-28 (view) (permuted) | PERMUTE | Qcur-28 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 1002 | cache_k_l28 (view) | VIEW | cache_k_l28{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1003 | cache_k_l28 (view) (permuted) | PERMUTE | cache_k_l28 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1004 | cache_v_l28 (view) | VIEW | cache_v_l28{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1005 | cache_v_l28 (view) (permuted) | PERMUTE | cache_v_l28 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1006 | node_1005 | FLASH_ATTN_EXT | Qcur-28 (view) (permuted){128, 5, 32, 1} , cache_k_l28 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 1007 | kqv_out-28 | RESHAPE | node_1005{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 1008 | node_1007 | MUL_MAT | blk.28.attn_output.weight{4096, 4096, 1, 1} , kqv_out-28{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1009 | ffn_inp-28 | ADD | node_1007{4096, 5, 1, 1} , l_out-27{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1010 | norm-28 | RMS_NORM | ffn_inp-28{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1011 | ffn_norm-28 | MUL | norm-28{4096, 5, 1, 1} , blk.28.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1012 | ffn_gate-28 | MUL_MAT | blk.28.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-28{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1013 | ffn_up-28 | MUL_MAT | blk.28.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-28{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1014 | ffn_swiglu-28 | SWIGLU | ffn_gate-28{12288, 5, 1, 1} , ffn_up-28{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1015 | ffn_out-28 | MUL_MAT | blk.28.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-28{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1016 | l_out-28 | ADD | ffn_out-28{4096, 5, 1, 1} , ffn_inp-28{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1017 | norm-29 | RMS_NORM | l_out-28{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1018 | attn_norm-29 | MUL | norm-29{4096, 5, 1, 1} , blk.29.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1019 | Qcur-29 | MUL_MAT | blk.29.attn_q.weight{4096, 4096, 1, 1} , attn_norm-29{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1020 | Qcur-29 | RESHAPE | Qcur-29{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 1021 | norm-29 | RMS_NORM | Qcur-29{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1022 | Qcur_normed-29 | MUL | norm-29{128, 32, 5, 1} , blk.29.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1023 | Qcur-29 | ROPE | Qcur_normed-29{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1024 | Vcur-29 | MUL_MAT | blk.29.attn_v.weight{4096, 1024, 1, 1} , attn_norm-29{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1025 | Vcur-29 | RESHAPE | Vcur-29{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1026 | Kcur-29 | MUL_MAT | blk.29.attn_k.weight{4096, 1024, 1, 1} , attn_norm-29{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1027 | Kcur-29 | RESHAPE | Kcur-29{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1028 | norm-29 | RMS_NORM | Kcur-29{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 1029 | Kcur_normed-29 | MUL | norm-29{128, 8, 5, 1} , blk.29.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1030 | Kcur-29 | ROPE | Kcur_normed-29{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1031 | Kcur-29 (view) | VIEW | Kcur-29{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1032 | cache_k_l29 (view) | SET_ROWS | Kcur-29 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1033 | Vcur-29 (view) | VIEW | Vcur-29{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1034 | cache_v_l29 (view) | SET_ROWS | Vcur-29 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1035 | Qcur-29 (view) | VIEW | Qcur-29{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1036 | Qcur-29 (view) (permuted) | PERMUTE | Qcur-29 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 1037 | cache_k_l29 (view) | VIEW | cache_k_l29{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1038 | cache_k_l29 (view) (permuted) | PERMUTE | cache_k_l29 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1039 | cache_v_l29 (view) | VIEW | cache_v_l29{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1040 | cache_v_l29 (view) (permuted) | PERMUTE | cache_v_l29 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1041 | node_1040 | FLASH_ATTN_EXT | Qcur-29 (view) (permuted){128, 5, 32, 1} , cache_k_l29 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 1042 | kqv_out-29 | RESHAPE | node_1040{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 1043 | node_1042 | MUL_MAT | blk.29.attn_output.weight{4096, 4096, 1, 1} , kqv_out-29{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1044 | ffn_inp-29 | ADD | node_1042{4096, 5, 1, 1} , l_out-28{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1045 | norm-29 | RMS_NORM | ffn_inp-29{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1046 | ffn_norm-29 | MUL | norm-29{4096, 5, 1, 1} , blk.29.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1047 | ffn_gate-29 | MUL_MAT | blk.29.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-29{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1048 | ffn_up-29 | MUL_MAT | blk.29.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-29{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1049 | ffn_swiglu-29 | SWIGLU | ffn_gate-29{12288, 5, 1, 1} , ffn_up-29{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1050 | ffn_out-29 | MUL_MAT | blk.29.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-29{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1051 | l_out-29 | ADD | ffn_out-29{4096, 5, 1, 1} , ffn_inp-29{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1052 | norm-30 | RMS_NORM | l_out-29{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1053 | attn_norm-30 | MUL | norm-30{4096, 5, 1, 1} , blk.30.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1054 | Qcur-30 | MUL_MAT | blk.30.attn_q.weight{4096, 4096, 1, 1} , attn_norm-30{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1055 | Qcur-30 | RESHAPE | Qcur-30{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 1056 | norm-30 | RMS_NORM | Qcur-30{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1057 | Qcur_normed-30 | MUL | norm-30{128, 32, 5, 1} , blk.30.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1058 | Qcur-30 | ROPE | Qcur_normed-30{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1059 | Vcur-30 | MUL_MAT | blk.30.attn_v.weight{4096, 1024, 1, 1} , attn_norm-30{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1060 | Vcur-30 | RESHAPE | Vcur-30{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1061 | Kcur-30 | MUL_MAT | blk.30.attn_k.weight{4096, 1024, 1, 1} , attn_norm-30{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1062 | Kcur-30 | RESHAPE | Kcur-30{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1063 | norm-30 | RMS_NORM | Kcur-30{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 1064 | Kcur_normed-30 | MUL | norm-30{128, 8, 5, 1} , blk.30.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1065 | Kcur-30 | ROPE | Kcur_normed-30{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1066 | Kcur-30 (view) | VIEW | Kcur-30{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1067 | cache_k_l30 (view) | SET_ROWS | Kcur-30 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1068 | Vcur-30 (view) | VIEW | Vcur-30{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1069 | cache_v_l30 (view) | SET_ROWS | Vcur-30 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1070 | Qcur-30 (view) | VIEW | Qcur-30{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1071 | Qcur-30 (view) (permuted) | PERMUTE | Qcur-30 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 1072 | cache_k_l30 (view) | VIEW | cache_k_l30{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1073 | cache_k_l30 (view) (permuted) | PERMUTE | cache_k_l30 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1074 | cache_v_l30 (view) | VIEW | cache_v_l30{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1075 | cache_v_l30 (view) (permuted) | PERMUTE | cache_v_l30 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1076 | node_1075 | FLASH_ATTN_EXT | Qcur-30 (view) (permuted){128, 5, 32, 1} , cache_k_l30 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 1077 | kqv_out-30 | RESHAPE | node_1075{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 1078 | node_1077 | MUL_MAT | blk.30.attn_output.weight{4096, 4096, 1, 1} , kqv_out-30{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1079 | ffn_inp-30 | ADD | node_1077{4096, 5, 1, 1} , l_out-29{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1080 | norm-30 | RMS_NORM | ffn_inp-30{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1081 | ffn_norm-30 | MUL | norm-30{4096, 5, 1, 1} , blk.30.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1082 | ffn_gate-30 | MUL_MAT | blk.30.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-30{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1083 | ffn_up-30 | MUL_MAT | blk.30.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-30{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1084 | ffn_swiglu-30 | SWIGLU | ffn_gate-30{12288, 5, 1, 1} , ffn_up-30{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1085 | ffn_out-30 | MUL_MAT | blk.30.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-30{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1086 | l_out-30 | ADD | ffn_out-30{4096, 5, 1, 1} , ffn_inp-30{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1087 | norm-31 | RMS_NORM | l_out-30{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1088 | attn_norm-31 | MUL | norm-31{4096, 5, 1, 1} , blk.31.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1089 | Qcur-31 | MUL_MAT | blk.31.attn_q.weight{4096, 4096, 1, 1} , attn_norm-31{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1090 | Qcur-31 | RESHAPE | Qcur-31{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 1091 | norm-31 | RMS_NORM | Qcur-31{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1092 | Qcur_normed-31 | MUL | norm-31{128, 32, 5, 1} , blk.31.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1093 | Qcur-31 | ROPE | Qcur_normed-31{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1094 | Vcur-31 | MUL_MAT | blk.31.attn_v.weight{4096, 1024, 1, 1} , attn_norm-31{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1095 | Vcur-31 | RESHAPE | Vcur-31{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1096 | Kcur-31 | MUL_MAT | blk.31.attn_k.weight{4096, 1024, 1, 1} , attn_norm-31{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1097 | Kcur-31 | RESHAPE | Kcur-31{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1098 | norm-31 | RMS_NORM | Kcur-31{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 1099 | Kcur_normed-31 | MUL | norm-31{128, 8, 5, 1} , blk.31.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1100 | Kcur-31 | ROPE | Kcur_normed-31{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1101 | Kcur-31 (view) | VIEW | Kcur-31{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1102 | cache_k_l31 (view) | SET_ROWS | Kcur-31 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1103 | Vcur-31 (view) | VIEW | Vcur-31{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1104 | cache_v_l31 (view) | SET_ROWS | Vcur-31 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1105 | Qcur-31 (view) | VIEW | Qcur-31{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1106 | Qcur-31 (view) (permuted) | PERMUTE | Qcur-31 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 1107 | cache_k_l31 (view) | VIEW | cache_k_l31{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1108 | cache_k_l31 (view) (permuted) | PERMUTE | cache_k_l31 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1109 | cache_v_l31 (view) | VIEW | cache_v_l31{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1110 | cache_v_l31 (view) (permuted) | PERMUTE | cache_v_l31 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1111 | node_1110 | FLASH_ATTN_EXT | Qcur-31 (view) (permuted){128, 5, 32, 1} , cache_k_l31 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 1112 | kqv_out-31 | RESHAPE | node_1110{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 1113 | node_1112 | MUL_MAT | blk.31.attn_output.weight{4096, 4096, 1, 1} , kqv_out-31{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1114 | ffn_inp-31 | ADD | node_1112{4096, 5, 1, 1} , l_out-30{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1115 | norm-31 | RMS_NORM | ffn_inp-31{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1116 | ffn_norm-31 | MUL | norm-31{4096, 5, 1, 1} , blk.31.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1117 | ffn_gate-31 | MUL_MAT | blk.31.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-31{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1118 | ffn_up-31 | MUL_MAT | blk.31.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-31{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1119 | ffn_swiglu-31 | SWIGLU | ffn_gate-31{12288, 5, 1, 1} , ffn_up-31{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1120 | ffn_out-31 | MUL_MAT | blk.31.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-31{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1121 | l_out-31 | ADD | ffn_out-31{4096, 5, 1, 1} , ffn_inp-31{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1122 | norm-32 | RMS_NORM | l_out-31{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1123 | attn_norm-32 | MUL | norm-32{4096, 5, 1, 1} , blk.32.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1124 | Qcur-32 | MUL_MAT | blk.32.attn_q.weight{4096, 4096, 1, 1} , attn_norm-32{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1125 | Qcur-32 | RESHAPE | Qcur-32{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 1126 | norm-32 | RMS_NORM | Qcur-32{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1127 | Qcur_normed-32 | MUL | norm-32{128, 32, 5, 1} , blk.32.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1128 | Qcur-32 | ROPE | Qcur_normed-32{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1129 | Vcur-32 | MUL_MAT | blk.32.attn_v.weight{4096, 1024, 1, 1} , attn_norm-32{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1130 | Vcur-32 | RESHAPE | Vcur-32{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1131 | Kcur-32 | MUL_MAT | blk.32.attn_k.weight{4096, 1024, 1, 1} , attn_norm-32{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1132 | Kcur-32 | RESHAPE | Kcur-32{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1133 | norm-32 | RMS_NORM | Kcur-32{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 1134 | Kcur_normed-32 | MUL | norm-32{128, 8, 5, 1} , blk.32.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1135 | Kcur-32 | ROPE | Kcur_normed-32{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1136 | Kcur-32 (view) | VIEW | Kcur-32{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1137 | cache_k_l32 (view) | SET_ROWS | Kcur-32 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1138 | Vcur-32 (view) | VIEW | Vcur-32{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1139 | cache_v_l32 (view) | SET_ROWS | Vcur-32 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1140 | Qcur-32 (view) | VIEW | Qcur-32{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1141 | Qcur-32 (view) (permuted) | PERMUTE | Qcur-32 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 1142 | cache_k_l32 (view) | VIEW | cache_k_l32{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1143 | cache_k_l32 (view) (permuted) | PERMUTE | cache_k_l32 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1144 | cache_v_l32 (view) | VIEW | cache_v_l32{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1145 | cache_v_l32 (view) (permuted) | PERMUTE | cache_v_l32 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1146 | node_1145 | FLASH_ATTN_EXT | Qcur-32 (view) (permuted){128, 5, 32, 1} , cache_k_l32 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 1147 | kqv_out-32 | RESHAPE | node_1145{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 1148 | node_1147 | MUL_MAT | blk.32.attn_output.weight{4096, 4096, 1, 1} , kqv_out-32{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1149 | ffn_inp-32 | ADD | node_1147{4096, 5, 1, 1} , l_out-31{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1150 | norm-32 | RMS_NORM | ffn_inp-32{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1151 | ffn_norm-32 | MUL | norm-32{4096, 5, 1, 1} , blk.32.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1152 | ffn_gate-32 | MUL_MAT | blk.32.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-32{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1153 | ffn_up-32 | MUL_MAT | blk.32.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-32{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1154 | ffn_swiglu-32 | SWIGLU | ffn_gate-32{12288, 5, 1, 1} , ffn_up-32{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1155 | ffn_out-32 | MUL_MAT | blk.32.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-32{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1156 | l_out-32 | ADD | ffn_out-32{4096, 5, 1, 1} , ffn_inp-32{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1157 | norm-33 | RMS_NORM | l_out-32{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1158 | attn_norm-33 | MUL | norm-33{4096, 5, 1, 1} , blk.33.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1159 | Qcur-33 | MUL_MAT | blk.33.attn_q.weight{4096, 4096, 1, 1} , attn_norm-33{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1160 | Qcur-33 | RESHAPE | Qcur-33{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 1161 | norm-33 | RMS_NORM | Qcur-33{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1162 | Qcur_normed-33 | MUL | norm-33{128, 32, 5, 1} , blk.33.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1163 | Qcur-33 | ROPE | Qcur_normed-33{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1164 | Vcur-33 | MUL_MAT | blk.33.attn_v.weight{4096, 1024, 1, 1} , attn_norm-33{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1165 | Vcur-33 | RESHAPE | Vcur-33{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1166 | Kcur-33 | MUL_MAT | blk.33.attn_k.weight{4096, 1024, 1, 1} , attn_norm-33{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1167 | Kcur-33 | RESHAPE | Kcur-33{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1168 | norm-33 | RMS_NORM | Kcur-33{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 1169 | Kcur_normed-33 | MUL | norm-33{128, 8, 5, 1} , blk.33.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1170 | Kcur-33 | ROPE | Kcur_normed-33{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1171 | Kcur-33 (view) | VIEW | Kcur-33{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1172 | cache_k_l33 (view) | SET_ROWS | Kcur-33 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1173 | Vcur-33 (view) | VIEW | Vcur-33{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1174 | cache_v_l33 (view) | SET_ROWS | Vcur-33 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1175 | Qcur-33 (view) | VIEW | Qcur-33{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1176 | Qcur-33 (view) (permuted) | PERMUTE | Qcur-33 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 1177 | cache_k_l33 (view) | VIEW | cache_k_l33{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1178 | cache_k_l33 (view) (permuted) | PERMUTE | cache_k_l33 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1179 | cache_v_l33 (view) | VIEW | cache_v_l33{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1180 | cache_v_l33 (view) (permuted) | PERMUTE | cache_v_l33 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1181 | node_1180 | FLASH_ATTN_EXT | Qcur-33 (view) (permuted){128, 5, 32, 1} , cache_k_l33 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 1182 | kqv_out-33 | RESHAPE | node_1180{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 1183 | node_1182 | MUL_MAT | blk.33.attn_output.weight{4096, 4096, 1, 1} , kqv_out-33{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1184 | ffn_inp-33 | ADD | node_1182{4096, 5, 1, 1} , l_out-32{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1185 | norm-33 | RMS_NORM | ffn_inp-33{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1186 | ffn_norm-33 | MUL | norm-33{4096, 5, 1, 1} , blk.33.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1187 | ffn_gate-33 | MUL_MAT | blk.33.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-33{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1188 | ffn_up-33 | MUL_MAT | blk.33.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-33{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1189 | ffn_swiglu-33 | SWIGLU | ffn_gate-33{12288, 5, 1, 1} , ffn_up-33{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1190 | ffn_out-33 | MUL_MAT | blk.33.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-33{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1191 | l_out-33 | ADD | ffn_out-33{4096, 5, 1, 1} , ffn_inp-33{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1192 | norm-34 | RMS_NORM | l_out-33{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1193 | attn_norm-34 | MUL | norm-34{4096, 5, 1, 1} , blk.34.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1194 | Qcur-34 | MUL_MAT | blk.34.attn_q.weight{4096, 4096, 1, 1} , attn_norm-34{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1195 | Qcur-34 | RESHAPE | Qcur-34{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 1196 | norm-34 | RMS_NORM | Qcur-34{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1197 | Qcur_normed-34 | MUL | norm-34{128, 32, 5, 1} , blk.34.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1198 | Qcur-34 | ROPE | Qcur_normed-34{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1199 | Vcur-34 | MUL_MAT | blk.34.attn_v.weight{4096, 1024, 1, 1} , attn_norm-34{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1200 | Vcur-34 | RESHAPE | Vcur-34{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1201 | Kcur-34 | MUL_MAT | blk.34.attn_k.weight{4096, 1024, 1, 1} , attn_norm-34{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1202 | Kcur-34 | RESHAPE | Kcur-34{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1203 | norm-34 | RMS_NORM | Kcur-34{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 1204 | Kcur_normed-34 | MUL | norm-34{128, 8, 5, 1} , blk.34.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1205 | Kcur-34 | ROPE | Kcur_normed-34{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1206 | Kcur-34 (view) | VIEW | Kcur-34{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1207 | cache_k_l34 (view) | SET_ROWS | Kcur-34 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1208 | Vcur-34 (view) | VIEW | Vcur-34{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1209 | cache_v_l34 (view) | SET_ROWS | Vcur-34 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1210 | Qcur-34 (view) | VIEW | Qcur-34{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1211 | Qcur-34 (view) (permuted) | PERMUTE | Qcur-34 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 1212 | cache_k_l34 (view) | VIEW | cache_k_l34{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1213 | cache_k_l34 (view) (permuted) | PERMUTE | cache_k_l34 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1214 | cache_v_l34 (view) | VIEW | cache_v_l34{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1215 | cache_v_l34 (view) (permuted) | PERMUTE | cache_v_l34 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1216 | node_1215 | FLASH_ATTN_EXT | Qcur-34 (view) (permuted){128, 5, 32, 1} , cache_k_l34 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 1217 | kqv_out-34 | RESHAPE | node_1215{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 1218 | node_1217 | MUL_MAT | blk.34.attn_output.weight{4096, 4096, 1, 1} , kqv_out-34{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1219 | ffn_inp-34 | ADD | node_1217{4096, 5, 1, 1} , l_out-33{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1220 | norm-34 | RMS_NORM | ffn_inp-34{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1221 | ffn_norm-34 | MUL | norm-34{4096, 5, 1, 1} , blk.34.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1222 | ffn_gate-34 | MUL_MAT | blk.34.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-34{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1223 | ffn_up-34 | MUL_MAT | blk.34.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-34{4096, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1224 | ffn_swiglu-34 | SWIGLU | ffn_gate-34{12288, 5, 1, 1} , ffn_up-34{12288, 5, 1, 1} | → | {12288, 5, 1, 1} |
| 1225 | ffn_out-34 | MUL_MAT | blk.34.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-34{12288, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1226 | l_out-34 | ADD | ffn_out-34{4096, 5, 1, 1} , ffn_inp-34{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1227 | norm-35 | RMS_NORM | l_out-34{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1228 | attn_norm-35 | MUL | norm-35{4096, 5, 1, 1} , blk.35.attn_norm.weight{4096, 1, 1, 1} | → | {4096, 5, 1, 1} |
| 1229 | Qcur-35 | MUL_MAT | blk.35.attn_q.weight{4096, 4096, 1, 1} , attn_norm-35{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1230 | Qcur-35 | RESHAPE | Qcur-35{4096, 5, 1, 1} | → | {128, 32, 5, 1} |
| 1231 | norm-35 | RMS_NORM | Qcur-35{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1232 | Qcur_normed-35 | MUL | norm-35{128, 32, 5, 1} , blk.35.attn_q_norm.weight{128, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1233 | Qcur-35 | ROPE | Qcur_normed-35{128, 32, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 32, 5, 1} |
| 1234 | Vcur-35 | MUL_MAT | blk.35.attn_v.weight{4096, 1024, 1, 1} , attn_norm-35{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1235 | Vcur-35 | RESHAPE | Vcur-35{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1236 | Kcur-35 | MUL_MAT | blk.35.attn_k.weight{4096, 1024, 1, 1} , attn_norm-35{4096, 5, 1, 1} | → | {1024, 5, 1, 1} |
| 1237 | Kcur-35 | RESHAPE | Kcur-35{1024, 5, 1, 1} | → | {128, 8, 5, 1} |
| 1238 | norm-35 | RMS_NORM | Kcur-35{128, 8, 5, 1} | → | {128, 8, 5, 1} |
| 1239 | Kcur_normed-35 | MUL | norm-35{128, 8, 5, 1} , blk.35.attn_k_norm.weight{128, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1240 | Kcur-35 | ROPE | Kcur_normed-35{128, 8, 5, 1} , leaf_6{5, 1, 1, 1} | → | {128, 8, 5, 1} |
| 1241 | Kcur-35 (view) | VIEW | Kcur-35{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1242 | cache_k_l35 (view) | SET_ROWS | Kcur-35 (view){1024, 5, 1, 1} , leaf_10{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1243 | Vcur-35 (view) | VIEW | Vcur-35{128, 8, 5, 1} | → | {1024, 5, 1, 1} |
| 1244 | cache_v_l35 (view) | SET_ROWS | Vcur-35 (view){1024, 5, 1, 1} , leaf_12{5, 1, 1, 1} | → | {1024, 89856, 1, 1} |
| 1245 | Qcur-35 (view) | VIEW | Qcur-35{128, 32, 5, 1} | → | {128, 32, 5, 1} |
| 1246 | Qcur-35 (view) (permuted) | PERMUTE | Qcur-35 (view){128, 32, 5, 1} | → | {128, 5, 32, 1} |
| 1247 | cache_k_l35 (view) | VIEW | cache_k_l35{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1248 | cache_k_l35 (view) (permuted) | PERMUTE | cache_k_l35 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1249 | cache_v_l35 (view) | VIEW | cache_v_l35{1024, 89856, 1, 1} | → | {128, 8, 256, 1} |
| 1250 | cache_v_l35 (view) (permuted) | PERMUTE | cache_v_l35 (view){128, 8, 256, 1} | → | {128, 256, 8, 1} |
| 1251 | node_1250 | FLASH_ATTN_EXT | Qcur-35 (view) (permuted){128, 5, 32, 1} , cache_k_l35 (view) (permuted){128, 256, 8, 1} | → | {128, 32, 5, 1} |
| 1252 | kqv_out-35 | RESHAPE | node_1250{128, 32, 5, 1} | → | {4096, 5, 1, 1} |
| 1253 | node_1252 | MUL_MAT | blk.35.attn_output.weight{4096, 4096, 1, 1} , kqv_out-35{4096, 5, 1, 1} | → | {4096, 5, 1, 1} |
| 1254 | node_1253 | GET_ROWS | node_1252{4096, 5, 1, 1} , leaf_474{1, 1, 1, 1} | → | {4096, 1, 1, 1} |
| 1255 | node_1254 | GET_ROWS | l_out-34{4096, 5, 1, 1} , leaf_474{1, 1, 1, 1} | → | {4096, 1, 1, 1} |
| 1256 | ffn_inp-35 | ADD | node_1253{4096, 1, 1, 1} , node_1254{4096, 1, 1, 1} | → | {4096, 1, 1, 1} |
| 1257 | norm-35 | RMS_NORM | ffn_inp-35{4096, 1, 1, 1} | → | {4096, 1, 1, 1} |
| 1258 | ffn_norm-35 | MUL | norm-35{4096, 1, 1, 1} , blk.35.ffn_norm.weight{4096, 1, 1, 1} | → | {4096, 1, 1, 1} |
| 1259 | ffn_gate-35 | MUL_MAT | blk.35.ffn_gate.weight{4096, 12288, 1, 1} , ffn_norm-35{4096, 1, 1, 1} | → | {12288, 1, 1, 1} |
| 1260 | ffn_up-35 | MUL_MAT | blk.35.ffn_up.weight{4096, 12288, 1, 1} , ffn_norm-35{4096, 1, 1, 1} | → | {12288, 1, 1, 1} |
| 1261 | ffn_swiglu-35 | SWIGLU | ffn_gate-35{12288, 1, 1, 1} , ffn_up-35{12288, 1, 1, 1} | → | {12288, 1, 1, 1} |
| 1262 | ffn_out-35 | MUL_MAT | blk.35.ffn_down.weight{12288, 4096, 1, 1} , ffn_swiglu-35{12288, 1, 1, 1} | → | {4096, 1, 1, 1} |
| 1263 | l_out-35 | ADD | ffn_out-35{4096, 1, 1, 1} , ffn_inp-35{4096, 1, 1, 1} | → | {4096, 1, 1, 1} |
| 1264 | norm | RMS_NORM | l_out-35{4096, 1, 1, 1} | → | {4096, 1, 1, 1} |
| 1265 | result_norm | MUL | norm{4096, 1, 1, 1} , output_norm.weight{4096, 1, 1, 1} | → | {4096, 1, 1, 1} |
| 1266 | result_output | MUL_MAT | output.weight{4096, 151936, 1, 1} , result_norm{4096, 1, 1, 1} | → | {151936, 1, 1, 1} |
llama-eval-callback's debug output, built and run natively via WSL. GET_ROWS is an embedding lookup, MUL_MAT a matrix multiply, FLASH_ATTN_EXT llama.cpp's fused attention kernel (query, key-cache and value-cache in one op, rather than separate matmul-softmax-matmul steps).