xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-06-11 19:02:16 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Model] support bitsandbytes quantization with minicpm model (#10842)

Browse Source 
Signed-off-by: Ubuntu <zixuanzhang@bytedance.com>
This commit is contained in:
zixuanzhang226 2024-12-02 19:06:41 -08:00 committed by GitHub
parent 4433195ab7
commit d746268e92
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 10 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/model_executor/models/minicpm.py
View File

@ -534,6 +534,16 @@ class MiniCPMForCausalLM(nn.Module, SupportsLoRA, SupportsPP):
} }
embedding_padding_modules = ["lm_head"] embedding_padding_modules = ["lm_head"]
# BitandBytes specific attributes
bitsandbytes_stacked_params_mapping = {
# shard_name, weight_name, index
"q_proj": ("qkv_proj", 0),
"k_proj": ("qkv_proj", 1),
"v_proj": ("qkv_proj", 2),
"gate_proj": ("gate_up_proj", 0),
"up_proj": ("gate_up_proj", 1),
}
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""): def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__() super().__init__()
config = vllm_config.model_config.hf_config config = vllm_config.model_config.hf_config