add openai_harmony to requirements (#4006)

* add openai_harmony to requirements

* remove try-except

Author: irexyc

lmdeploy/serve/openai/harmony_utils.py

@@ -3,15 +3,11 @@
 from typing import List
 
 import shortuuid
+from openai_harmony import HarmonyEncodingName, Role, StreamableParser, load_harmony_encoding
 
 from lmdeploy.serve.openai.protocol import (ChatMessage, DeltaFunctionCall, DeltaMessage, DeltaToolCall, FunctionCall,
                                             ToolCall)
 
-try:
-    from openai_harmony import HarmonyEncodingName, Role, StreamableParser, load_harmony_encoding
-except (ImportError, ModuleNotFoundError):
-    pass
-
 _harmony_encoding = None
 
 

lmdeploy/tokenizer.py

@@ -384,14 +384,10 @@ class GptOssTokenizer(HuggingFaceTokenizer):
 
     def __init__(self, model_dir: str):
         super(GptOssTokenizer, self).__init__(model_dir)
-        try:
-            import openai_harmony  # noqa: F401
-            from openai_harmony import HarmonyEncodingName, Role, StreamableParser, load_harmony_encoding
-            encoding = load_harmony_encoding(HarmonyEncodingName.HARMONY_GPT_OSS)
-            self.role = Role.ASSISTANT
-            self.parser = partial(StreamableParser, encoding, role=Role.ASSISTANT)
-        except ImportError:
-            raise ImportError('Please install openai_harmony by `pip install openai_harmony`')
+        from openai_harmony import HarmonyEncodingName, Role, StreamableParser, load_harmony_encoding
+        encoding = load_harmony_encoding(HarmonyEncodingName.HARMONY_GPT_OSS)
+        self.role = Role.ASSISTANT
+        self.parser = partial(StreamableParser, encoding, role=Role.ASSISTANT)
 
     def detokenize_incrementally(self,
                                  all_input_ids: Sequence[int],

requirements/runtime_ascend.txt

@@ -6,6 +6,7 @@ fire
 mmengine-lite
 numpy
 openai
+openai_harmony
 outlines<0.1.0
 partial_json_parser
 peft<=0.11.1

requirements/runtime_camb.txt

@@ -5,6 +5,7 @@ fire
 mmengine-lite
 numpy
 openai
+openai_harmony
 outlines<0.1.0
 partial_json_parser
 peft<=0.11.1

requirements/runtime_cuda.txt

@@ -6,6 +6,7 @@ fire
 mmengine-lite
 numpy
 openai
+openai_harmony
 outlines
 partial_json_parser
 peft<=0.14.0

requirements/runtime_maca.txt

@@ -5,6 +5,7 @@ fire
 mmengine-lite
 numpy
 openai
+openai_harmony
 outlines<0.1.0
 partial_json_parser
 peft<=0.11.1

requirements/runtime_rocm.txt

@@ -5,6 +5,7 @@ fire
 mmengine-lite
 numpy
 openai
+openai_harmony
 outlines
 partial_json_parser
 peft<=0.14.0