vision support

refact_webgui/webgui/selfhost_fastapi_completions.py

@@ -71,7 +71,7 @@ class NlpCompletion(NlpSamplingParams):
 
 class ChatMessage(BaseModel):
     role: str
-    content: str
+    content: Union[str, List[Dict]]
     # TODO: validate using pydantic
     tool_calls: Optional[List[Dict[str, Any]]] = None
     tool_call_id: Optional[str] = None
@@ -520,7 +520,7 @@ async def litellm_streamer():
                     max_tokens=min(model_dict.get('T_out', post.max_tokens), post.max_tokens),
                     tools=post.tools,
                     tool_choice=post.tool_choice,
-                    stop=post.stop,
+                    stop=post.stop if post.stop else None,
                     n=post.n,
                 )
                 finish_reason = None
@@ -559,7 +559,7 @@ async def litellm_non_streamer():
                     max_tokens=min(model_dict.get('T_out', post.max_tokens), post.max_tokens),
                     tools=post.tools,
                     tool_choice=post.tool_choice,
-                    stop=post.stop,
+                    stop=post.stop if post.stop else None,
                     n=post.n,
                 )
                 finish_reason = None

setup.py

@@ -35,7 +35,7 @@ class PyPackage:
     "refact_webgui": PyPackage(
         requires=["aiohttp", "aiofiles", "cryptography", "fastapi==0.100.0", "giturlparse", "pydantic>=2",
                   "starlette==0.27.0", "uvicorn", "uvloop", "termcolor", "python-multipart", "more_itertools",
-                  "scyllapy==1.3.0", "pandas>=2.0.3", "litellm>=1.44.24"],
+                  "scyllapy==1.3.0", "pandas>=2.0.3", "litellm>=1.49.5"],
         requires_packages=["refact_known_models", "refact_utils"],
         data=["webgui/static/*", "webgui/static/components/modals/*",
               "webgui/static/dashboards/*", "webgui/static/assets/*", "webgui/static/utils/*",]),