Method call_llm

eval_heldout/rewoo/nodes/LLMNode.py:134–240

(self, prompt, stop)

        return completion

    def call_llm(self, prompt, stop):
        if self.model_name in OPENAI_COMPLETION_MODELS:
            response = openai.Completion.create(
                model=self.model_name,
                prompt=prompt,
                temperature=OPENAI_CONFIG["temperature"],
                max_tokens=OPENAI_CONFIG["max_tokens"],
                top_p=OPENAI_CONFIG["top_p"],
                frequency_penalty=OPENAI_CONFIG["frequency_penalty"],
                presence_penalty=OPENAI_CONFIG["presence_penalty"],
                stop=stop
            )
            return {"input": prompt,
                    "output": response["choices"][0]["text"],
                    "prompt_tokens": response["usage"]["prompt_tokens"],
                    "completion_tokens": response["usage"]["completion_tokens"]}
        elif self.model_name in OPENAI_CHAT_MODELS:
            print('*****GPT-4*****')
            messages = [{"role": "user", "content": prompt}]
            while True:
                try:
                    response = openai.ChatCompletion.create(
                        model=self.model_name,
                        messages=messages,
                        # temperature=OPENAI_CONFIG["temperature"],
                        temperature=0.0,
                        max_tokens=OPENAI_CONFIG["max_tokens"],
                        # top_p=OPENAI_CONFIG["top_p"],
                        # frequency_penalty=OPENAI_CONFIG["frequency_penalty"],
                        # presence_penalty=OPENAI_CONFIG["presence_penalty"],
                        # stop=stop
                    )
                    break
                except:
                    continue

            traj.append({
                'id': f'{ID}_{len(traj)}',
                'conversations': [
                    {
                        'from': 'human',
                        'value': prompt,
                    },
                    {
                        'from': 'gpt',
                        'value': response["choices"][0]["message"]["content"],
                    },
                ]
            })
            json.dump(traj, open(LOG, 'w'))

            return {"input": prompt,
                    "output": response["choices"][0]["message"]["content"],
                    "prompt_tokens": response["usage"]["prompt_tokens"],
                    "completion_tokens": response["usage"]["completion_tokens"]}
        elif self.model_name in LLAMA_WEIGHTS:
            instruction, input = prompt[0], prompt[1]
            output, prompt = self.al.lora_generate(instruction, input)