Skip to content

Commit 960e726

Browse files
CISCCISC
authored
chore : cleanup llama_model_loader::TENSOR_ usage (ggml-org#12492)
1 parent ea1518e commit 960e726

File tree

1 file changed

+21
-21
lines changed

1 file changed

+21
-21
lines changed

‎src/llama-model.cpp

Lines changed: 21 additions & 21 deletions
Original file line numberDiff line numberDiff line change
@@ -2329,7 +2329,7 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
23292329
layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), { n_embd }, 0);
23302330
layer.attn_norm_b = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "bias", i), { n_embd }, 0);
23312331

2332-
layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), { n_embd, n_embd + 2 * n_embd_gqa }, llama_model_loader::TENSOR_NOT_REQUIRED);
2332+
layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), { n_embd, n_embd + 2 * n_embd_gqa }, TENSOR_NOT_REQUIRED);
23332333
if (layer.wqkv == nullptr) {
23342334
layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd}, 0);
23352335
layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, 0);
@@ -3215,16 +3215,16 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
32153215
auto & layer = layers[i];
32163216

32173217
layer.attn_norm = create_tensor(tn(LLM_TENSOR_ATTN_NORM, "weight", i), {n_embd}, 0);
3218-
layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
3219-
layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
3218+
layer.wqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "weight", i), {n_embd, n_embd + 2*n_embd_gqa}, TENSOR_NOT_REQUIRED);
3219+
layer.bqkv = create_tensor(tn(LLM_TENSOR_ATTN_QKV, "bias", i), {n_embd + 2*n_embd_gqa}, TENSOR_NOT_REQUIRED);
32203220

32213221
if (layer.wqkv == nullptr) {
32223222
layer.wq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "weight", i), {n_embd, n_embd_head_k * n_head}, 0);
32233223
layer.wk = create_tensor(tn(LLM_TENSOR_ATTN_K, "weight", i), {n_embd, n_embd_k_gqa}, 0);
32243224
layer.wv = create_tensor(tn(LLM_TENSOR_ATTN_V, "weight", i), {n_embd, n_embd_v_gqa}, 0);
3225-
layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
3226-
layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
3227-
layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, llama_model_loader::TENSOR_NOT_REQUIRED);
3225+
layer.bq = create_tensor(tn(LLM_TENSOR_ATTN_Q, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
3226+
layer.bk = create_tensor(tn(LLM_TENSOR_ATTN_K, "bias", i), {n_embd_gqa}, TENSOR_NOT_REQUIRED);
3227+
layer.bv = create_tensor(tn(LLM_TENSOR_ATTN_V, "bias", i), {n_embd_gqa}, TENSOR_NOT_REQUIRED);
32283228
}
32293229

32303230
layer.wo = create_tensor(tn(LLM_TENSOR_ATTN_OUT, "weight", i), {n_embd, n_embd}, 0);
@@ -3335,12 +3335,12 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
33353335
layer.time_mix_w2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_W2, "weight", i), {time_mix_extra_dim, n_embd, 5}, 0);
33363336

33373337
layer.time_mix_lerp_x = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_X, "weight", i), {n_embd, 1, 1}, 0);
3338-
layer.time_mix_lerp_w = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_W, "weight", i), {n_embd, 1, 1}, llama_model_loader::TENSOR_NOT_REQUIRED);
3339-
layer.time_mix_lerp_k = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_K, "weight", i), {n_embd, 1, 1}, llama_model_loader::TENSOR_NOT_REQUIRED);
3340-
layer.time_mix_lerp_v = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_V, "weight", i), {n_embd, 1, 1}, llama_model_loader::TENSOR_NOT_REQUIRED);
3341-
layer.time_mix_lerp_r = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_R, "weight", i), {n_embd, 1, 1}, llama_model_loader::TENSOR_NOT_REQUIRED);
3342-
layer.time_mix_lerp_g = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_G, "weight", i), {n_embd, 1, 1}, llama_model_loader::TENSOR_NOT_REQUIRED);
3343-
layer.time_mix_lerp_fused = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_FUSED, "weight", i), {n_embd, 1, 1, 5}, llama_model_loader::TENSOR_NOT_REQUIRED);
3338+
layer.time_mix_lerp_w = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_W, "weight", i), {n_embd, 1, 1}, TENSOR_NOT_REQUIRED);
3339+
layer.time_mix_lerp_k = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_K, "weight", i), {n_embd, 1, 1}, TENSOR_NOT_REQUIRED);
3340+
layer.time_mix_lerp_v = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_V, "weight", i), {n_embd, 1, 1}, TENSOR_NOT_REQUIRED);
3341+
layer.time_mix_lerp_r = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_R, "weight", i), {n_embd, 1, 1}, TENSOR_NOT_REQUIRED);
3342+
layer.time_mix_lerp_g = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_G, "weight", i), {n_embd, 1, 1}, TENSOR_NOT_REQUIRED);
3343+
layer.time_mix_lerp_fused = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_FUSED, "weight", i), {n_embd, 1, 1, 5}, TENSOR_NOT_REQUIRED);
33443344
GGML_ASSERT(!(layer.time_mix_lerp_fused == NULL && layer.time_mix_lerp_w == NULL));
33453345

33463346
layer.time_mix_first = create_tensor(tn(LLM_TENSOR_TIME_MIX_FIRST, "weight", i), {head_size, n_embd / head_size}, 0);
@@ -3370,7 +3370,7 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
33703370
tok_embd = create_tensor(tn(LLM_TENSOR_TOKEN_EMBD, "weight"), {n_embd, n_vocab}, 0);
33713371

33723372
output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
3373-
output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
3373+
output_norm_b = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "bias"), {n_embd}, TENSOR_NOT_REQUIRED);
33743374
output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab}, 0);
33753375

33763376
const int time_mix_extra_dim = hparams.time_mix_extra_dim;
@@ -3396,7 +3396,7 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
33963396
layer.time_mix_lerp_x = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_X, "weight", i), {n_embd, 1, 1}, 0);
33973397
layer.time_mix_lerp_fused = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_FUSED, "weight", i), {n_embd, 1, 1, 5}, 0);
33983398

3399-
layer.time_mix_first = create_tensor(tn(LLM_TENSOR_TIME_MIX_FIRST, "weight", i), {head_size, n_embd / head_size}, llama_model_loader::TENSOR_NOT_REQUIRED);
3399+
layer.time_mix_first = create_tensor(tn(LLM_TENSOR_TIME_MIX_FIRST, "weight", i), {head_size, n_embd / head_size}, TENSOR_NOT_REQUIRED);
34003400
layer.time_mix_decay = create_tensor(tn(LLM_TENSOR_TIME_MIX_DECAY, "weight", i), {n_embd}, 0);
34013401
layer.time_mix_decay_w1 = create_tensor(tn(LLM_TENSOR_TIME_MIX_DECAY_W1, "weight", i), {n_embd, time_decay_extra_dim}, 0);
34023402
layer.time_mix_decay_w2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_DECAY_W2, "weight", i), {time_decay_extra_dim, attn_hidden_size}, 0);
@@ -3405,9 +3405,9 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
34053405
layer.time_mix_receptance = create_tensor(tn(LLM_TENSOR_TIME_MIX_RECEPTANCE, "weight", i), {attn_hidden_size, n_embd}, 0);
34063406
layer.time_mix_gate = create_tensor(tn(LLM_TENSOR_TIME_MIX_GATE, "weight", i), {attn_hidden_size, n_embd}, 0);
34073407
// optional bias tensors
3408-
layer.time_mix_key_b = create_tensor(tn(LLM_TENSOR_TIME_MIX_KEY, "bias", i), {attn_key_value_size}, llama_model_loader::TENSOR_NOT_REQUIRED);
3409-
layer.time_mix_value_b = create_tensor(tn(LLM_TENSOR_TIME_MIX_VALUE, "bias", i), {attn_key_value_size}, llama_model_loader::TENSOR_NOT_REQUIRED);
3410-
layer.time_mix_receptance_b = create_tensor(tn(LLM_TENSOR_TIME_MIX_RECEPTANCE, "bias", i), {attn_hidden_size}, llama_model_loader::TENSOR_NOT_REQUIRED);
3408+
layer.time_mix_key_b = create_tensor(tn(LLM_TENSOR_TIME_MIX_KEY, "bias", i), {attn_key_value_size}, TENSOR_NOT_REQUIRED);
3409+
layer.time_mix_value_b = create_tensor(tn(LLM_TENSOR_TIME_MIX_VALUE, "bias", i), {attn_key_value_size}, TENSOR_NOT_REQUIRED);
3410+
layer.time_mix_receptance_b = create_tensor(tn(LLM_TENSOR_TIME_MIX_RECEPTANCE, "bias", i), {attn_hidden_size}, TENSOR_NOT_REQUIRED);
34113411

34123412
layer.time_mix_output = create_tensor(tn(LLM_TENSOR_TIME_MIX_OUTPUT, "weight", i), {n_embd, attn_hidden_size}, 0);
34133413

@@ -3528,8 +3528,8 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
35283528
layer.time_mix_v2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_V2, "weight", i), {n_lora_value_res_mix, n_embd}, 0);
35293529
}
35303530

3531-
layer.time_mix_g1 = create_tensor(tn(LLM_TENSOR_TIME_MIX_G1, "weight", i), {n_embd, n_lora_gate}, llama_model_loader::TENSOR_NOT_REQUIRED);
3532-
layer.time_mix_g2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_G2, "weight", i), {n_lora_gate, n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
3531+
layer.time_mix_g1 = create_tensor(tn(LLM_TENSOR_TIME_MIX_G1, "weight", i), {n_embd, n_lora_gate}, TENSOR_NOT_REQUIRED);
3532+
layer.time_mix_g2 = create_tensor(tn(LLM_TENSOR_TIME_MIX_G2, "weight", i), {n_lora_gate, n_embd}, TENSOR_NOT_REQUIRED);
35333533

35343534
try {
35353535
layer.time_mix_lerp_fused = create_tensor(tn(LLM_TENSOR_TIME_MIX_LERP_FUSED, "weight", i), {n_embd, 1, 1, 6}, 0);
@@ -3546,8 +3546,8 @@ bool llama_model::load_tensors(llama_model_loader & ml) {
35463546
layer.time_mix_value = create_tensor(tn(LLM_TENSOR_TIME_MIX_VALUE, "weight", i), {attn_hidden_size, n_embd}, 0);
35473547
layer.time_mix_receptance = create_tensor(tn(LLM_TENSOR_TIME_MIX_RECEPTANCE, "weight", i), {attn_hidden_size, n_embd}, 0);
35483548

3549-
layer.time_mix_ln = create_tensor(tn(LLM_TENSOR_TIME_MIX_LN, "weight", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
3550-
layer.time_mix_ln_b = create_tensor(tn(LLM_TENSOR_TIME_MIX_LN, "bias", i), {n_embd}, llama_model_loader::TENSOR_NOT_REQUIRED);
3549+
layer.time_mix_ln = create_tensor(tn(LLM_TENSOR_TIME_MIX_LN, "weight", i), {n_embd}, TENSOR_NOT_REQUIRED);
3550+
layer.time_mix_ln_b = create_tensor(tn(LLM_TENSOR_TIME_MIX_LN, "bias", i), {n_embd}, TENSOR_NOT_REQUIRED);
35513551
layer.time_mix_output = create_tensor(tn(LLM_TENSOR_TIME_MIX_OUTPUT, "weight", i), {n_embd, attn_hidden_size}, 0);
35523552

35533553
layer.ffn_norm = create_tensor(tn(LLM_TENSOR_FFN_NORM, "weight", i), {n_embd}, 0);

0 commit comments

Comments
 (0)