vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : update WavTokenizer to non-causal attn

Browse source
This commit is contained in:
Georgi Gerganov 2024-12-17 10:25:17 +02:00
parent 2033fb7eef
commit 824fa750d4
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
2 changed files with 3 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

2
convert_hf_to_gguf.py
View file

@ -2069,6 +2069,8 @@ class WavTokenizerDecModel(Model):
self.gguf_writer.add_convnext_embedding_length(self.hparams["convnext"]["n_embd"])
self.gguf_writer.add_convnext_block_count (self.hparams["convnext"]["n_layer"])
self.gguf_writer.add_causal_attention(False)
@Model.register("Qwen2MoeForCausalLM")
class Qwen2MoeModel(Model):

1
src/llama.cpp
View file

@ -6393,6 +6393,7 @@ static void llm_load_hparams(
ml.get_key(LLM_KV_ATTENTION_LAYERNORM_EPS, hparams.f_norm_eps);
ml.get_key(LLM_KV_ATTENTION_GROUPNORM_EPS, hparams.f_norm_group_eps);
ml.get_key(LLM_KV_ATTENTION_GROUPNORM_GROUPS, hparams.n_norm_groups);
ml.get_key(LLM_KV_ATTENTION_CAUSAL, hparams.causal_attn);
} break;
default: (void)0;
}