Skip to content

Commit 0583d95

Browse files
gabe-l-hartgabe-l-hart
committed
fix: Remove unnecessary prefix on tensor constants
 Signed-off-by: Gabe Goodhart <ghart@us.ibm.com> Co-authored-by: Sigbjørn Skjæret <sigbjorn.skjaeret@scala.com>
1 parent f8b81c0 commit 0583d95

File tree

1 file changed

+13
-13
lines changed

1 file changed

+13
-13
lines changed

‎src/llama-model.cpp

Lines changed: 13 additions & 13 deletions
Original file line numberDiff line numberDiff line change
@@ -3430,10 +3430,10 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
34303430
// output
34313431
{
34323432
output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
3433-
output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
3433+
output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, TENSOR_NOT_REQUIRED);
34343434
// if output is NULL, init from the input tok embed, duplicated to allow offloading
34353435
if (output == NULL) {
3436-
output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
3436+
output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, TENSOR_DUPLICATED);
34373437
}
34383438
}
34393439

@@ -3446,10 +3446,10 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
34463446
if (hparams.is_recurrent(i)) {
34473447
// ssm layers
34483448
layer.ssm_in = create_tensor(tn(LLM_TENSOR_SSM_IN, "weight", i), {n_embd, d_in_proj}, 0);
3449-
layer.ssm_in_b = create_tensor(tn(LLM_TENSOR_SSM_IN, "bias", i), {n_embd, d_in_proj}, llama_model_loader::TENSOR_NOT_REQUIRED);
3449+
layer.ssm_in_b = create_tensor(tn(LLM_TENSOR_SSM_IN, "bias", i), {n_embd, d_in_proj}, TENSOR_NOT_REQUIRED);
34503450

34513451
layer.ssm_conv1d = create_tensor(tn(LLM_TENSOR_SSM_CONV1D, "weight", i), {d_conv, d_inner + 2*n_group*d_state}, 0);
3452-
layer.ssm_conv1d_b = create_tensor(tn(LLM_TENSOR_SSM_CONV1D, "bias", i), {d_inner + 2*n_group*d_state}, llama_model_loader::TENSOR_NOT_REQUIRED);
3452+
layer.ssm_conv1d_b = create_tensor(tn(LLM_TENSOR_SSM_CONV1D, "bias", i), {d_inner + 2*n_group*d_state}, TENSOR_NOT_REQUIRED);
34533453

34543454
layer.ssm_dt_b = create_tensor(tn(LLM_TENSOR_SSM_DT, "bias", i), {n_ssm_head}, 0);
34553455

@@ -3470,17 +3470,17 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
34703470
layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa_i}, 0);
34713471
layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa_i}, 0);
34723472
layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head_i, n_embd}, 0);
3473-
layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
3474-
layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_k_gqa_i}, llama_model_loader::TENSOR_NOT_REQUIRED);
3475-
layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_v_gqa_i}, llama_model_loader::TENSOR_NOT_REQUIRED);
3476-
layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
3473+
layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
3474+
layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_k_gqa_i}, TENSOR_NOT_REQUIRED);
3475+
layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_v_gqa_i}, TENSOR_NOT_REQUIRED);
3476+
layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
34773477
}
34783478

34793479
// feed forward (w/ optional biases)
34803480
if (n_expert > 0) {
34813481
// MoE FFN
34823482
layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
3483-
layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0));
3483+
layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
34843484
layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
34853485
layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff, n_expert}, TENSOR_NOT_REQUIRED);
34863486
layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), { n_ff, n_embd, n_expert}, 0);
@@ -3494,13 +3494,13 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
34943494
}
34953495
} else {
34963496
layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
3497-
layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, llama_model_loader::TENSOR_NOT_REQUIRED | (i != 0 ? llama_model_loader::TENSOR_DUPLICATED : 0));
3497+
layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
34983498
layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);
34993499
layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0);
35003500
layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0);
3501-
layer.ffn_gate_b = create_tensor(tn(LLM_TENSOR_FFN_GATE, "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED);
3502-
layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
3503-
layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED);
3501+
layer.ffn_gate_b = create_tensor(tn(LLM_TENSOR_FFN_GATE, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);
3502+
layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
3503+
layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);
35043504
}
35053505
}
35063506
} break;

0 commit comments

Comments
 (0)