llama: fix kv shifting, add tests, add per-model build_rope (#31683)

* llama: fix kv shifting and add tests

* llama: add graph build_rope (#31693)

* llama: add graph build_rope

* improve

* migrate new models

* update review skill

* fix ci

* add --kv-shift to CI
This commit is contained in:
Xuan-Son Nguyen authored and GitHub committed 2026-11-05 21:10:36 +01:00
1 parent b1c902f026
commit 6023692b1e
144 files changed
+1154 -1847

No files matched your search

+2 -11
View File
@@ -162,17 +162,8 @@ llama_model_ricursive::graph::graph(const llm_graph_params & params) : llm_graph
cb(Kcur, "Kcur_normed", il);
// global layers use no RoPE (NoPE): the reference rotates only sliding_attention layers (modeling_ricursive.py:127-129,203-205)
if (hparams.has_rope(il)) {
const float freq_base_l = model.get_rope_freq_base (cparams, il);
const float freq_scale_l = model.get_rope_freq_scale(cparams, il);
Qcur = ggml_rope_ext(ctx0, Qcur, inp_pos, nullptr,
n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
ext_factor, attn_factor, beta_fast, beta_slow);
Kcur = ggml_rope_ext(ctx0, Kcur, inp_pos, nullptr,
n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
ext_factor, attn_factor, beta_fast, beta_slow);
}
Qcur = build_rope(Qcur, inp_pos, il);
Kcur = build_rope(Kcur, inp_pos, il);
cb(Qcur, "Qcur", il);
cb(Kcur, "Kcur", il);
cb(Vcur, "Vcur", il);