Mirrors/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/llama.cpp.git synced 2024-12-26 14:20:31 +01:00
Code Issues Packages Projects Releases Wiki Activity

llama : fix Gemma-2 Query scaling factors (#8473)

Browse Source 
* 9B - query_pre_attn_scalar = 256 not 224

See 03e657582d

Gemma 9b should use 256 and not 224 (self.config.hidden_size // self.config.num_attention_heads)

* llama : fix Gemma-2 Query scaling factor

ggml-ci

---------

Co-authored-by: Daniel Han <danielhanchen@gmail.com>
This commit is contained in:
Georgi Gerganov 2024-07-14 14:05:09 +03:00 committed by GitHub
parent e236528e76
commit 73cf442e7b
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
convert_hf_to_gguf.py
View File

@ -2504,11 +2504,6 @@ class Gemma2Model(Model):
) )
self.gguf_writer.add_sliding_window(self.hparams["sliding_window"]) self.gguf_writer.add_sliding_window(self.hparams["sliding_window"])
# sanity check
attn_scalar = self.hparams["query_pre_attn_scalar"]
if attn_scalar != hparams["hidden_size"] / hparams["num_attention_heads"]:
raise ValueError("query_pre_attn_scalar must be equal to n_embd / n_head")
def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]: def modify_tensors(self, data_torch: Tensor, name: str, bid: int | None) -> Iterable[tuple[str, Tensor]]:
del bid # unused del bid # unused

7
src/llama.cpp
View File

@ -11680,7 +11680,12 @@ struct llm_build_context {
ext_factor, attn_factor, beta_fast, beta_slow); ext_factor, attn_factor, beta_fast, beta_slow);
cb(Qcur, "Qcur", il); cb(Qcur, "Qcur", il);
Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd / n_head))); // ref: https://github.com/google/gemma_pytorch/commit/03e657582d17cb5a8617ebf333c1c16f3694670e
switch (model.type) {
case e_model::MODEL_9B: Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd_head_k))); break;
case e_model::MODEL_27B: Qcur = ggml_scale(ctx0, Qcur, 1.0f / sqrtf(float(n_embd / n_head))); break;
default: GGML_ASSERT(false);
};
cb(Qcur, "Qcur_scaled", il); cb(Qcur, "Qcur_scaled", il);
Kcur = ggml_rope_ext( Kcur = ggml_rope_ext(