An LLM-powered knowledge curation system that researches a topic and generates a full-length report with citations.
Python agentic-ragdeep-researchemnlp2024knowledge-curationlarge-language-modelsnaaclnlpreport-generationretrieval-augmented-generation
Added GroqOpenAIModel
ndehoucheffbd82041597
diff --git a/README.md b/README.mdindex 742ab4a..669c084 100644--- a/README.md+++ b/README.md@@ -100,7 +100,7 @@ runner = STORMWikiRunner(engine_args, lm_configs, rm)```Currently, our package support:-- `OpenAIModel`, `AzureOpenAIModel`, `ClaudeModel`, `VLLMClient`, `TGIClient`, `TogetherClient`, `OllamaClient`, `GoogleModel`, `DeepSeekModel` as language model components+- `OpenAIModel`, `AzureOpenAIModel`, `GroqOpenAIModel`, `ClaudeModel`, `VLLMClient`, `TGIClient`, `TogetherClient`, `OllamaClient`, `GoogleModel`, `DeepSeekModel` as language model components- `YouRM`, `BingSearch`, `VectorRM`, `SerperRM`, `BraveRM` as retrieval module components:star2: **PRs for integrating more language models into [knowledge_storm/lm.py](knowledge_storm/lm.py) and search engines/retrievers into [knowledge_storm/rm.py](knowledge_storm/rm.py) are highly appreciated!**diff --git a/knowledge_storm/lm.py b/knowledge_storm/lm.pyindex 79d9ba5..6925248 100644--- a/knowledge_storm/lm.py+++ b/knowledge_storm/lm.py@@ -232,6 +232,113 @@ class AzureOpenAIModel(dspy.AzureOpenAI):return usage+class GroqOpenAIModel(dspy.OpenAI):+ """A wrapper class for Groq API, compatible with dspy.OpenAI."""++ def __init__(+ self,+ model: str = "mixtral-8x7b-32768",+ api_key: Optional[str] = None,+ api_base: str = "https://api.groq.com/openai/v1",+ **kwargs+ ):+ super().__init__(model=model, api_key=api_key, api_base=api_base, **kwargs)+ self._token_usage_lock = threading.Lock()+ self.prompt_tokens = 0+ self.completion_tokens = 0+ self.model = model+ self.api_key = api_key or os.getenv("GROQ_API_KEY")+ self.api_base = api_base+ if not self.api_key:+ raise ValueError("Groq API key must be provided either as an argument or as an environment variable GROQ_API_KEY")++ def log_usage(self, response):+ """Log the total tokens from the Groq API response."""+ usage_data = response.get('usage')+ if usage_data:+ with self._token_usage_lock:+ self.prompt_tokens += usage_data.get('prompt_tokens', 0)+ self.completion_tokens += usage_data.get('completion_tokens', 0)++ def get_usage_and_reset(self):+ """Get the total tokens used and reset the token usage."""+ usage = {+ self.model:+ {'prompt_tokens': self.prompt_tokens, 'completion_tokens': self.completion_tokens}+ }+ self.prompt_tokens = 0+ self.completion_tokens = 0+ return usage++ @backoff.on_exception(+ backoff.expo,+ ERRORS,+ max_time=1000,+ on_backoff=backoff_hdlr,+ giveup=giveup_hdlr,+ )+ def _create_completion(self, prompt: str, **kwargs):+ """Create a completion using the Groq API."""+ headers = {+ "Content-Type": "application/json",+ "Authorization": f"Bearer {self.api_key}"+ }++ # Remove unsupported fields+ kwargs.pop('logprobs', None)+ kwargs.pop('logit_bias', None)+ kwargs.pop('top_logprobs', None)++ # Ensure N is 1 if supplied+ if 'n' in kwargs and kwargs['n'] != 1:+ raise ValueError("Groq API only supports N=1")++ # Adjust temperature if it's 0+ if kwargs.get('temperature', 1) == 0:+ kwargs['temperature'] = 1e-8++ data = {+ "model": self.model,+ "messages": [{"role": "user", "content": prompt}],+ **kwargs+ }++ # Remove 'name' field from messages if present+ for message in data['messages']:+ message.pop('name', None)++ response = requests.post(f"{self.api_base}/chat/completions", headers=headers, json=data)+ response.raise_for_status()+ return response.json()++ def __call__(+ self,+ prompt: str,+ only_completed: bool = True,+ return_sorted: bool = False,+ **kwargs,+ ) -> List[Dict[str, Any]]:+ """Call the Groq API to generate completions."""+ assert only_completed, "for now"+ assert return_sorted is False, "for now"++ response = self._create_completion(prompt, **kwargs)++ # Log the token usage from the Groq API response.+ self.log_usage(response)++ choices = response["choices"]+ completions = [choice["message"]["content"] for choice in choices]++ history = {+ "prompt": prompt,+ "response": response,+ "kwargs": kwargs,+ }+ self.history.append(history)++ return completions+class ClaudeModel(dspy.dsp.modules.lm.LM):"""Copied from dspy/dsp/modules/anthropic.py with the addition of tracking token usage."""