ayo/ipex-llm
1
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 1

LLM: Update qwen readme (#10245)

Browse source
This commit is contained in:
Wang, Jian4 2024-02-26 17:03:09 +08:00 committed by GitHub
parent 15ad2fd72e
commit 6c74b99a28
2 changed files with 2 additions and 2 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
python/llm/example/CPU/Speculative-Decoding/qwen/README.md
View file

@ -92,7 +92,7 @@ First token latency x.xxxxs
```
### 4. Accelerate with BIGDL_OPT_IPEX
`Qwen/Qwen-72B-Chat` is not supported using ipex now.
To accelerate speculative decoding on CPU, you can install our validated version of [IPEX 2.3.0+git004cd72d](https://github.com/intel/intel-extension-for-pytorch/tree/004cd72db60e87bb0712d42e3120bac9854bd77e) by following steps: (Other versions of IPEX may have some conflicts and can not accelerate speculative decoding correctly.)
#### 4.1 Download IPEX installation script

2
python/llm/src/bigdl/llm/transformers/speculative.py
View file

@ -475,7 +475,7 @@ def speculative_generate(self,
("qwen" in self.config.model_type) or
("chatglm" in self.config.model_type)):
invalidInputError(False, "BigDL Speculative Decoding with IPEX BF16 only supports \
Llama, Baichuan2, Mistral and ChatGLM and Qwen models currently.")
Llama, Baichuan2, Mistral, ChatGLM and Qwen models currently.")
if "chatglm" in self.config.model_type:
global query_group_size
query_group_size = draft_model.config.num_attention_heads // \