Support for vLLM chat completion endpoint

[noobHappylife]

Currently, i noticed the module HFClientVLLM only calls the completion endpoints. As vLLM now support applying chat templates. Would it be possible to add support for chat completion endpoints?

the following is a naive approach to add a chat completion support, edited the _generate method for HFClientVLLM

    def _generate(self, prompt, **kwargs):
        kwargs = {**self.kwargs, **kwargs}

        model_type = kwargs.get("model_type",None)
        url = self.urls.pop(0)
        self.urls.append(url)

        system_prompt = kwargs.get("system_prompt",None)
        if model_type == "chat":
            messages = [{"role": "user", "content": prompt}]
            if system_prompt:
                messages.insert(0, {"role": "system", "content": system_prompt})
            payload = {
                "model": self.kwargs["model"],
                "messages": messages,
                **kwargs,
            }
            response = send_hfvllm_chat_request_v00(
                f"{url}/v1/chat/completions",
                json=payload,
                headers=self.headers,
            )

            try:
                json_response = response.json()
                completions = json_response["choices"]
                response = {
                    "prompt": prompt,
                    "choices": [{"text": c["message"]['content']} for c in completions],
                }
                return response

            except Exception:
                print("Failed to parse JSON response:", response.text)
                raise Exception("Received invalid JSON response from server")
        else:
            payload = {
                "model": self.kwargs["model"],
                "prompt": prompt,
                **kwargs,
            }

            response = send_hfvllm_request_v00(
                f"{url}/v1/completions",
                json=payload,
                headers=self.headers,
            )

            try:
                json_response = response.json()
                completions = json_response["choices"]
                response = {
                    "prompt": prompt,
                    "choices": [{"text": c["text"]} for c in completions],
                }
                return response

            except Exception:
                print("Failed to parse JSON response:", response.text)
                raise Exception("Received invalid JSON response from server")


@CacheMemory.cache
def send_hfvllm_request_v00(arg, **kwargs):
    return requests.post(arg, **kwargs)

@CacheMemory.cache
def send_hfvllm_chat_request_v00(arg, **kwargs):
    return requests.post(arg, **kwargs)

[arnavsinghvi11]

Thanks @noobHappylife ! Happy to merge a PR for this change.