remove vllm patch

the-seeds · Jan 23, 2024 · 95a30f3 · 95a30f3
1 parent cced589
commit 95a30f3
Show file tree

Hide file tree

Showing 4 changed files with 2 additions and 76 deletions.
diff --git a/requirements.txt b/requirements.txt
@@ -1,4 +1,4 @@
 numpy
 sse-starlette
-transformers
-vllm==0.2.6
+transformers>=4.34.0
+vllm>=0.2.6
diff --git a/src/imitater/config/config.py b/src/imitater/config/config.py
@@ -14,5 +14,3 @@ class Config:
     embed_model_path: str
     embed_model_device: List[int]
     embed_batch_size: int
-
-    enable_attn_bias: bool
diff --git a/src/imitater/model/chat_model.py b/src/imitater/model/chat_model.py
@@ -4,7 +4,6 @@
 from vllm import AsyncEngineArgs, AsyncLLMEngine, SamplingParams
 
 from ..agent import get_agent
-from ..utils.vllm_monkey_patch import llama_attn_bias_monkey_patch
 
 
 if TYPE_CHECKING:
@@ -22,9 +21,6 @@ def __init__(self, config: "Config") -> None:
         self._load_generation_config()
 
     def _init_vllm_engine(self) -> None:
-        if self._config.enable_attn_bias:
-            llama_attn_bias_monkey_patch()
-
         engine_args = AsyncEngineArgs(model=self._config.chat_model_path, trust_remote_code=True)
         engine_args.tensor_parallel_size = len(self._config.chat_model_device)
         self._engine = AsyncLLMEngine.from_engine_args(engine_args)

diff --git a/src/imitater/utils/vllm_monkey_patch.py b/src/imitater/utils/vllm_monkey_patch.py