Skip to content

Commit 4b93a2a

Browse files
gabe-l-hartCISC
gabe-l-hart
and
CISC
authored
fix: Remove unnecessary prefix on tensor constants
 Signed-off-by: Gabe Goodhart <[email protected]> Co-authored-by: Sigbjørn Skjæret <[email protected]> Co-authored-by: Sigbjørn Skjæret <[email protected]>
1 parent 0b84bd5 commit 4b93a2a

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
@@ -3399,10 +3399,10 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
33993399
// output
34003400
{
34013401
output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
3402-
output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_NOT_REQUIRED);
3402+
output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, TENSOR_NOT_REQUIRED);
34033403
// if output is NULL, init from the input tok embed, duplicated to allow offloading
34043404
if (output == NULL) {
3405-
output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, llama_model_loader::TENSOR_DUPLICATED);
3405+
output = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, TENSOR_DUPLICATED);
34063406
}
34073407
}
34083408

@@ -3415,10 +3415,10 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
34153415
if (hparams.is_recurrent(i)) {
34163416
// ssm layers
34173417
layer.ssm_in = create_tensor(tn(LLM_TENSOR_SSM_IN, "weight", i), {n_embd, d_in_proj}, 0);
3418-
layer.ssm_in_b = create_tensor(tn(LLM_TENSOR_SSM_IN, "bias", i), {n_embd, d_in_proj}, llama_model_loader::TENSOR_NOT_REQUIRED);
3418+
layer.ssm_in_b = create_tensor(tn(LLM_TENSOR_SSM_IN, "bias", i), {n_embd, d_in_proj}, TENSOR_NOT_REQUIRED);
34193419

34203420
layer.ssm_conv1d = create_tensor(tn(LLM_TENSOR_SSM_CONV1D, "weight", i), {d_conv, d_inner + 2*n_group*d_state}, 0);
3421-
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);
3421+
layer.ssm_conv1d_b = create_tensor(tn(LLM_TENSOR_SSM_CONV1D, "bias", i), {d_inner + 2*n_group*d_state}, TENSOR_NOT_REQUIRED);
34223422

34233423
layer.ssm_dt_b = create_tensor(tn(LLM_TENSOR_SSM_DT, "bias", i), {n_ssm_head}, 0);
34243424

@@ -3439,17 +3439,17 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
34393439
layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa_i}, 0);
34403440
layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa_i}, 0);
34413441
layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd_head_k * n_head_i, n_embd}, 0);
3442-
layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
3443-
layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_k_gqa_i}, llama_model_loader::TENSOR_NOT_REQUIRED);
3444-
layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_v_gqa_i}, llama_model_loader::TENSOR_NOT_REQUIRED);
3445-
layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
3442+
layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
3443+
layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_k_gqa_i}, TENSOR_NOT_REQUIRED);
3444+
layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_v_gqa_i}, TENSOR_NOT_REQUIRED);
3445+
layer.bo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
34463446
}
34473447

34483448
// feed forward (w/ optional biases)
34493449
if (n_expert > 0) {
34503450
// MoE FFN
34513451
layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
3452-
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));
3452+
layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
34533453
layer.ffn_gate_inp = create_tensor(tn(LLM_TENSOR_FFN_GATE_INP, "weight", i), {n_embd, n_expert}, 0);
34543454
layer.ffn_gate_exps = create_tensor(tn(LLM_TENSOR_FFN_GATE_EXPS, "weight", i), {n_embd, n_ff, n_expert}, TENSOR_NOT_REQUIRED);
34553455
layer.ffn_down_exps = create_tensor(tn(LLM_TENSOR_FFN_DOWN_EXPS, "weight", i), { n_ff, n_embd, n_expert}, 0);
@@ -3463,13 +3463,13 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
34633463
}
34643464
} else {
34653465
layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);
3466-
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));
3466+
layer.rope_freqs = create_tensor(tn(LLM_TENSOR_ROPE_FREQS, "weight", i), {n_rot/2}, TENSOR_NOT_REQUIRED | (i != 0 ? TENSOR_DUPLICATED : 0));
34673467
layer.ffn_gate = create_tensor(tn(LLM_TENSOR_FFN_GATE, "weight", i), {n_embd, n_ff}, 0);
34683468
layer.ffn_down = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "weight", i), { n_ff, n_embd}, 0);
34693469
layer.ffn_up = create_tensor(tn(LLM_TENSOR_FFN_UP, "weight", i), {n_embd, n_ff}, 0);
3470-
layer.ffn_gate_b = create_tensor(tn(LLM_TENSOR_FFN_GATE, "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED);
3471-
layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
3472-
layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, llama_model_loader::TENSOR_NOT_REQUIRED);
3470+
layer.ffn_gate_b = create_tensor(tn(LLM_TENSOR_FFN_GATE, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);
3471+
layer.ffn_down_b = create_tensor(tn(LLM_TENSOR_FFN_DOWN, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
3472+
layer.ffn_up_b = create_tensor(tn(LLM_TENSOR_FFN_UP, "bias", i), {n_ff}, TENSOR_NOT_REQUIRED);
34733473
}
34743474
}
34753475
} break;

0 commit comments

Comments
 (0)