switched to HF text-generation-inference

inference/README.md

@@ -86,7 +86,7 @@ For the worker, you'll also want to have the text-generation-inference server
 running:
 
 ```bash
-docker run --rm -it -p 8001:80 -e MODEL_ID=distilgpt2 ykilcher/text-generation-inference
+docker run --rm -it -p 8001:80 -e MODEL_ID=distilgpt2 ghcr.io/huggingface/text-generation-inference
 ```
 
 Run the client:

inference/full-dev-setup.sh

@@ -5,7 +5,7 @@
 tmux new-session -d -s "inference-dev-setup"
 tmux send-keys "docker run --rm -it -p 6379:6379 redis" C-m
 tmux split-window -h
-tmux send-keys "docker run --rm -it -p 8001:80 -e MODEL_ID=distilgpt2 ykilcher/text-generation-inference" C-m
+tmux send-keys "docker run --rm -it -p 8001:80 -e MODEL_ID=distilgpt2 ghcr.io/huggingface/text-generation-inference" C-m
 tmux split-window -h
 tmux send-keys "cd server" C-m
 tmux send-keys "uvicorn main:app --reload" C-m

inference/server/main.py

@@ -57,7 +57,7 @@ class MessageRequest(pydantic.BaseModel):
 
 
 class TokenResponseEvent(pydantic.BaseModel):
-    token: str
+    token: inference.TokenResponse
 
 
 class MessageRequestState(str, enum.Enum):
@@ -143,7 +143,7 @@ async def create_message(id: str, message_request: MessageRequest, fastapi_reque
 
         chat.conversation.messages.append(
             protocol.ConversationMessage(
-                text="".join([d.token for d in result_data[:-1]]),
+                text=response_packet.generated_text.text,
                 is_assistant=True,
             )
         )

inference/text-client/__main__.py

@@ -32,7 +32,7 @@ def main(backend_url: str = "http://127.0.0.1:8000"):
         print("Assistant: ", end="", flush=True)
         for event in client.events():
             data = json.loads(event.data)
-            print(data["token"], end="", flush=True)
+            print(data["token"]["text"], end="", flush=True)
         print()
 
 

inference/worker/__main__.py

@@ -54,24 +54,48 @@ def main(
                     "top_p": work_request.top_p,
                     "temperature": work_request.temperature,
                     "seed": work_request.seed,
+                    # "stop": ["User:", "Assistant:"], # TODO: this doesn't work... why?
                 },
             },
             stream=True,
             headers={"Accept": "text/event-stream"},
         )
-        response.raise_for_status()
+        try:
+            response.raise_for_status()
+        except requests.HTTPError:
+            logger.exception("Failed to get response from inference server")
+            return
 
         client = sseclient.SSEClient(response)
         for event in client.events():
+            logger.debug(f"Received event: {event}")
             data = json.loads(event.data)
-            if data["is_end"]:
+            if data["generated_text"]:
                 break
-            intermediate = data["event"]
-            ws.send(inference.WorkResponsePacket(token=intermediate["token"]).json())
-        ws.send(inference.WorkResponsePacket(is_end=True).json())
+            token = data["token"]
+            ws.send(
+                inference.WorkResponsePacket(
+                    token=inference.TokenResponse(
+                        text=token["text"],
+                        log_prob=token["logprob"],
+                        token_id=token["id"],
+                    )
+                ).json()
+            )
+        ws.send(
+            inference.WorkResponsePacket(
+                is_end=True,
+                generated_text=inference.GeneratedTextResponse(
+                    text=data["generated_text"],
+                ),
+            ).json()
+        )
 
     def on_error(ws: websocket.WebSocket, error: Exception):
-        logger.error(f"Connection error: {error}")
+        try:
+            raise error
+        except Exception:
+            logger.exception("Error in websocket")
 
     def on_close(ws: websocket.WebSocket, close_status_code: int, close_msg: str):
         logger.warning(f"Connection closed: {close_status_code=} {close_msg=}")

oasst-shared/oasst_shared/schemas/inference.py

@@ -20,6 +20,17 @@ class WorkRequest(pydantic.BaseModel):
     temperature: float = 1.0
 
 
+class TokenResponse(pydantic.BaseModel):
+    text: str
+    log_prob: float
+    token_id: int
+
+
+class GeneratedTextResponse(pydantic.BaseModel):
+    text: str
+
+
 class WorkResponsePacket(pydantic.BaseModel):
-    token: str | None = None
+    token: TokenResponse | None = None
+    generated_text: GeneratedTextResponse | None = None
     is_end: bool = False