Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

model: use ggml_rope_set_offset() (#27382)

Browse Source 
* model: use ggml_rope_set_offset()

* partially apply to deepseek2
This commit is contained in:
Xuan-Son Nguyen
2026-08-21 18:54:29 +02:00
committed by GitHub
parent d7fa69b7de
commit 873e5d8e39
6 changed files with 71 additions and 133 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/models/dflash.cpp
+2 -10
View File
@@ -591,17 +591,9 @@ llama_model_dflash::graph_dsv4::graph_dsv4(const llama_model & model, const llm_
kv = build_norm(kv, layer.attn_kv_norm, nullptr, LLM_NORM_RMS, il);
kv = ggml_reshape_3d(ctx0, kv, n_embd_head, 1, n_tokens);
ggml_tensor * kv_nope = ggml_view_3d(ctx0, kv, n_embd_head_nope, 1, n_tokens,
ggml_row_size(kv->type, n_embd_head),
ggml_row_size(kv->type, n_embd_head),
0);
ggml_tensor * kv_pe = ggml_view_3d(ctx0, kv, n_embd_head_rope, 1, n_tokens,
ggml_row_size(kv->type, n_embd_head),
ggml_row_size(kv->type, n_embd_head),
ggml_row_size(kv->type, n_embd_head_nope));
kv_pe = ggml_rope_ext(ctx0, kv_pe, inp_pos, nullptr, n_embd_head_rope, rope_type, 0,
kv = ggml_rope_ext(ctx0, kv, inp_pos, nullptr, n_embd_head_rope, rope_type, 0,
freq_base, 1.0f, 0.0f, 1.0f, 0.0f, 0.0f);
kv = ggml_concat(ctx0, kv_nope, kv_pe, 0);
kv = ggml_rope_set_offset(kv, n_embd_head_nope);
cb(kv, "kv_injected", il);
if (inp_attn->self_k_rot_swa) {