mikhail/llama.cpp-mtp-turboquant
1
0
Fork 0
Code Issues Pull Requests Actions 30 Packages Projects Releases Wiki Activity

model : support NVFP4 tensors for Gemma4 (#21971)

Browse Source 
* support nvfp4 tensors for Gemma4

* add wo_s to build_attn

* add wo_s to build_attn

* fix glm4
This commit is contained in:
Sigbjørn Skjæret
2026-04-16 16:51:47 +02:00
committed by GitHub
parent b572d1ecd6
commit f772f6e434
105 changed files with 149 additions and 148 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/models/t5-dec.cpp
+2 -2
View File
@@ -51,7 +51,7 @@ llm_build_t5_dec::llm_build_t5_dec(const llama_model & model, const llm_graph_pa
ggml_tensor * kq_b = build_pos_bias(pos_bucket_dec, attn_rel_b);
cur = build_attn(inp_attn_self,
model.layers[il].wo, model.layers[il].bo,
model.layers[il].wo, model.layers[il].bo, model.layers[il].wo_s,
Qcur, Kcur, Vcur, kq_b, nullptr, nullptr, 1.0f, il);
cb(cur, "kqv_out", il);
}
@@ -82,7 +82,7 @@ llm_build_t5_dec::llm_build_t5_dec(const llama_model & model, const llm_graph_pa
Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_outputs_enc);
cur = build_attn(inp_attn_cross,
model.layers[il].wo_cross, nullptr,
model.layers[il].wo_cross, nullptr, nullptr,
Qcur, Kcur, Vcur, nullptr, nullptr, nullptr, 1.0f, il);
cb(cur, "kqv_out", il);