Raise error in chat completion when unprocessable (#2257)

* Fix chat completion should throw error if not unprocessable

* add docstring

* forgot cassette

src/huggingface_hub/inference/_client.py

@@ -728,21 +728,23 @@ def chat_completion(
                     ),
                     stream=stream,
                 )
-            except HTTPError:
-                # Let's consider the server is not a chat completion server.
-                # Then we call again `chat_completion` which will render the chat template client side.
-                # (can be HTTP 500, HTTP 400, HTTP 404 depending on the server)
-                _set_as_non_chat_completion_server(model)
-                return self.chat_completion(
-                    messages=messages,
-                    model=model,
-                    stream=stream,
-                    max_tokens=max_tokens,
-                    seed=seed,
-                    stop=stop,
-                    temperature=temperature,
-                    top_p=top_p,
-                )
+            except HTTPError as e:
+                if e.response.status_code in (400, 404, 500):
+                    # Let's consider the server is not a chat completion server.
+                    # Then we call again `chat_completion` which will render the chat template client side.
+                    # (can be HTTP 500, HTTP 400, HTTP 404 depending on the server)
+                    _set_as_non_chat_completion_server(model)
+                    return self.chat_completion(
+                        messages=messages,
+                        model=model,
+                        stream=stream,
+                        max_tokens=max_tokens,
+                        seed=seed,
+                        stop=stop,
+                        temperature=temperature,
+                        top_p=top_p,
+                    )
+                raise
 
             if stream:
                 return _stream_chat_completion_response_from_bytes(data)  # type: ignore[arg-type]

src/huggingface_hub/inference/_generated/_async_client.py

@@ -728,21 +728,23 @@ async def chat_completion(
                     ),
                     stream=stream,
                 )
-            except _import_aiohttp().ClientResponseError:
-                # Let's consider the server is not a chat completion server.
-                # Then we call again `chat_completion` which will render the chat template client side.
-                # (can be HTTP 500, HTTP 400, HTTP 404 depending on the server)
-                _set_as_non_chat_completion_server(model)
-                return await self.chat_completion(
-                    messages=messages,
-                    model=model,
-                    stream=stream,
-                    max_tokens=max_tokens,
-                    seed=seed,
-                    stop=stop,
-                    temperature=temperature,
-                    top_p=top_p,
-                )
+            except _import_aiohttp().ClientResponseError as e:
+                if e.status in (400, 404, 500):
+                    # Let's consider the server is not a chat completion server.
+                    # Then we call again `chat_completion` which will render the chat template client side.
+                    # (can be HTTP 500, HTTP 400, HTTP 404 depending on the server)
+                    _set_as_non_chat_completion_server(model)
+                    return await self.chat_completion(
+                        messages=messages,
+                        model=model,
+                        stream=stream,
+                        max_tokens=max_tokens,
+                        seed=seed,
+                        stop=stop,
+                        temperature=temperature,
+                        top_p=top_p,
+                    )
+                raise
 
             if stream:
                 return _async_stream_chat_completion_response_from_bytes(data)  # type: ignore[arg-type]

tests/cassettes/InferenceClientVCRTest.test_chat_completion_unprocessable_entity.yaml

@@ -0,0 +1,53 @@
+interactions:
+- request:
+    body: '{"model": "tgi", "messages": "please output ''Observation''", "frequency_penalty":
+      null, "logit_bias": null, "logprobs": null, "max_tokens": 200, "n": null, "presence_penalty":
+      null, "seed": null, "stop": ["Observation", "Final Answer"], "temperature":
+      null, "tool_choice": null, "tool_prompt": null, "tools": null, "top_logprobs":
+      null, "top_p": null, "stream": false}'
+    headers:
+      Accept:
+      - '*/*'
+      Accept-Encoding:
+      - gzip, deflate, br
+      Connection:
+      - keep-alive
+      Content-Length:
+      - '367'
+      Content-Type:
+      - application/json
+      X-Amzn-Trace-Id:
+      - 52f7edcd-57c0-4258-8ff1-e3c520a43ef7
+      user-agent:
+      - unknown/None; hf_hub/0.23.0.dev0; python/3.10.12; torch/2.2.1; tensorflow/2.15.0.post1;
+        fastcore/1.5.23
+    method: POST
+    uri: https://api-inference.huggingface.co/models/meta-llama/Meta-Llama-3-70B-Instruct/v1/chat/completions
+  response:
+    body:
+      string: 'Failed to deserialize the JSON body into the target type: messages:
+        invalid type: string "please output ''Observation''", expected a sequence
+        at line 1 column 58'
+    headers:
+      Connection:
+      - keep-alive
+      Content-Type:
+      - text/plain; charset=utf-8
+      Date:
+      - Mon, 29 Apr 2024 16:18:50 GMT
+      Transfer-Encoding:
+      - chunked
+      access-control-allow-credentials:
+      - 'true'
+      access-control-allow-origin:
+      - '*'
+      vary:
+      - origin, Origin, Access-Control-Request-Method, Access-Control-Request-Headers
+      x-request-id:
+      - yECOxGeXLet5-Ae5rw808
+      x-sha:
+      - e8cf5276ae3e97cfde8a058e64a636f2cde47820
+    status:
+      code: 422
+      message: Unprocessable Entity
+version: 1

tests/test_inference_client.py

@@ -326,6 +326,19 @@ def test_chat_completion_with_tool(self) -> None:
             "location": "San Francisco, CA",
         }
 
+    def test_chat_completion_unprocessable_entity(self) -> None:
+        """Regression test for #2225.
+
+        See https://github.com/huggingface/huggingface_hub/issues/2225.
+        """
+        with self.assertRaises(HfHubHTTPError):
+            self.client.chat_completion(
+                "please output 'Observation'",  # Not a list of messages
+                stop=["Observation", "Final Answer"],
+                max_tokens=200,
+                model="meta-llama/Meta-Llama-3-70B-Instruct",
+            )
+
     @expect_deprecation("InferenceClient.conversational")
     def test_conversational(self) -> None:
         output = self.client.conversational("Hi, who are you?")

utils/generate_async_inference_client.py

@@ -335,8 +335,10 @@ def _adapt_text_generation_to_async(code: str) -> str:
 def _adapt_chat_completion_to_async(code: str) -> str:
     # Catch `aiohttp` error instead of `requests` error
     code = code.replace(
-        "except HTTPError:",
-        "except _import_aiohttp().ClientResponseError:",
+        """            except HTTPError as e:
+                if e.response.status_code in (400, 404, 500):""",
+        """            except _import_aiohttp().ClientResponseError as e:
+                if e.status in (400, 404, 500):""",
     )
 
     # Await text-generation call