bybrooklyn/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2026-04-18 09:03:35 -04:00
Code Issues Packages Projects Releases Wiki Activity

Revert "Update vendored llama.cpp to b7847" (#14061)

Browse Source
This commit is contained in:
Jeffrey Morgan
2026-02-03 18:39:36 -08:00
committed by GitHub
parent a6355329bf
commit b1fccabb34
240 changed files with 5050 additions and 21247 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
llama/llama.cpp/src/models/smallthinker.cpp vendored
View File

@@ -26,16 +26,10 @@ llm_build_smallthinker<iswa>::llm_build_smallthinker(const llama_model & model,
ggml_tensor * inp_out_ids = build_inp_out_ids();
for (int il = 0; il < n_layer; ++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);
ggml_tensor * inpSA = inpL;
ggml_tensor * probs = nullptr;
// This overlaps with SWA layers in current models, so get_rope_freq_base/scale may be superfluous
const bool use_rope = hparams.n_no_rope_layer_step == n_layer ||
il % hparams.n_no_rope_layer_step != 0;
ggml_tensor * probs = build_lora_mm(model.layers[il].ffn_gate_inp, inpL); // [n_expert, n_tokens]
probs = build_lora_mm(model.layers[il].ffn_gate_inp, inpL); // [n_expert, n_tokens]
cb(probs, "ffn_moe_logits", il);
// norm
@@ -58,11 +52,11 @@ llm_build_smallthinker<iswa>::llm_build_smallthinker(const llama_model & model,
Kcur = ggml_reshape_3d(ctx0, Kcur, n_embd_head, n_head_kv, n_tokens);
Vcur = ggml_reshape_3d(ctx0, Vcur, n_embd_head, n_head_kv, n_tokens);
if (use_rope) {
Qcur = ggml_rope_ext(ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base_l, freq_scale_l,
if (hparams.n_no_rope_layer_step == n_layer || il % hparams.n_no_rope_layer_step != 0) {
Qcur = ggml_rope_ext(ctx0, Qcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
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,
Kcur = ggml_rope_ext(ctx0, Kcur, inp_pos, nullptr, n_rot, rope_type, n_ctx_orig, freq_base, freq_scale,
ext_factor, attn_factor, beta_fast, beta_slow);
}
cb(Qcur, "Qcur", il);