model: apply .scale tensors automatically + fix correctness (#32669)

* model: apply .scale tensors automatically

* rm build_moe_ffn "_s" overload

* use llama_prec_policy

* fix ci

* build_lora_mm_id: rm w_s
This commit is contained in:
Xuan-Son Nguyen authored and GitHub committed 2026-11-21 16:30:01 +01:00
1 parent e50c254a22
commit 19bc41dbb4
39 files changed
+164 -290

No files matched your search

+2 -5
View File
@@ -97,6 +97,7 @@ void llama_model_inkling2::load_arch_tensors(llama_model_loader &) {
// output
output = create_tensor(tn(LLM_TENSOR_OUTPUT, "weight"), {n_embd, n_vocab_padded}, 0);
prec_policy.set_acc(output, GGML_PREC_F32);
output_norm = create_tensor(tn(LLM_TENSOR_OUTPUT_NORM, "weight"), {n_embd}, 0);
output_res_query = create_tensor(tn(LLM_TENSOR_OUTPUT_RES_QUERY, "weight"), {n_embd}, 0);
@@ -406,10 +407,7 @@ ggml_tensor * llama_model_inkling2::graph::build_layer(int il, int ip, ggml_tens
LLAMA_EXPERT_GATING_FUNC_TYPE_SIGMOID,
il,
nullptr,
layer.ffn_gate_up_exps,
layer.ffn_up_exps_s,
layer.ffn_gate_exps_s,
layer.ffn_down_exps_s);
layer.ffn_gate_up_exps);
cur = convolve(cur, layer.ffn_post_conv, nullptr, state, history, recr, 2*n_embd_k_gqa + n_embd);
cb(cur, "ffn_out", il);
@@ -433,7 +431,6 @@ void llama_model_inkling2::graph::build_output(ggml_tensor * cur, const std::vec
// lm_head
cur = build_lora_mm(model.output, cur);
ggml_prec_set_acc(cur, GGML_PREC_F32);
if (hparams.f_final_logit_softcapping != 0.0f) {
const float cap = hparams.f_final_logit_softcapping;