tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 03:55:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama : fix KV shift for qwen2vl (#13870)

Browse Source 
* llama : fix KV shift for qwen2vl

* add ref to the PR
This commit is contained in:
Xuan-Son Nguyen
2025-05-28 22:35:31 +02:00
committed by GitHub
parent 10961339b2
commit 763d06edb7
2 changed files with 11 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/llama-graph.cpp
View File

@ -455,7 +455,7 @@ llm_graph_context::llm_graph_context(const llm_graph_params & params) :
}
int64_t llm_graph_context::n_pos_per_embd() const {
return arch == LLM_ARCH_QWEN2VL ? 4 : 1;
return hparams.rope_type == LLAMA_ROPE_TYPE_MROPE ? 4 : 1;
}
void llm_graph_context::cb(ggml_tensor * cur, const char * name, int il) const {

12
src/llama-kv-cache.cpp
View File

@ -757,11 +757,19 @@ ggml_tensor * llama_kv_cache_unified::build_rope_shift(
const auto & yarn_beta_slow = cparams.yarn_beta_slow;
const auto & n_rot = hparams.n_rot;
const auto & rope_type = hparams.rope_type;
const auto & rope_type = hparams.rope_type == LLAMA_ROPE_TYPE_MROPE
// @ngxson : this is a workaround
// for M-RoPE, we want to rotate the whole vector when doing KV shift
// a normal RoPE should work, we just need to use the correct ordering
// ref: https://github.com/ggml-org/llama.cpp/pull/13870
? LLAMA_ROPE_TYPE_NEOX
: hparams.rope_type;
// See llm_build_deepseek2() for why attn_factor has to be scaled for YaRN RoPE to work correctly.
// See https://github.com/ggerganov/llama.cpp/discussions/7416 for detailed explanation.
const float yarn_attn_factor = model.arch == LLM_ARCH_DEEPSEEK2 ? 1.0f / (1.0f + 0.1f * logf(1.0f / freq_scale)) : cparams.yarn_attn_factor;
const float yarn_attn_factor = model.arch == LLM_ARCH_DEEPSEEK2
? 1.0f / (1.0f + 0.1f * logf(1.0f / freq_scale))
: cparams.yarn_attn_factor;
ggml_tensor * tmp;