Merge pull request #721 from arc53/feature/anthropic

anthropic LLM

application/api/answer/routes.py

@@ -29,6 +29,8 @@ answer = Blueprint('answer', __name__)
 
 if settings.LLM_NAME == "gpt4":
     gpt_model = 'gpt-4'
+elif settings.LLM_NAME == "anthropic":
+    gpt_model = 'claude-2'
 else:
     gpt_model = 'gpt-3.5-turbo'
 

application/llm/anthropic.py

@@ -0,0 +1,40 @@
+from application.llm.base import BaseLLM
+from application.core.settings import settings
+
+class AnthropicLLM(BaseLLM):
+
+    def __init__(self, api_key=None):
+        from anthropic import Anthropic, HUMAN_PROMPT, AI_PROMPT
+        self.api_key = api_key or settings.ANTHROPIC_API_KEY  # If not provided, use a default from settings
+        self.anthropic = Anthropic(api_key=self.api_key)
+        self.HUMAN_PROMPT = HUMAN_PROMPT
+        self.AI_PROMPT = AI_PROMPT
+
+    def gen(self, model, messages, engine=None, max_tokens=300, stream=False, **kwargs):
+        context = messages[0]['content']
+        user_question = messages[-1]['content']
+        prompt = f"### Context \n {context} \n ### Question \n {user_question}"
+        if stream:
+            return self.gen_stream(model, prompt, max_tokens, **kwargs)
+
+        completion = self.anthropic.completions.create(
+            model=model,
+            max_tokens_to_sample=max_tokens,
+            stream=stream,
+            prompt=f"{self.HUMAN_PROMPT} {prompt}{self.AI_PROMPT}",
+        )
+        return completion.completion
+
+    def gen_stream(self, model, messages, engine=None, max_tokens=300, **kwargs):
+        context = messages[0]['content']
+        user_question = messages[-1]['content']
+        prompt = f"### Context \n {context} \n ### Question \n {user_question}"
+        stream_response = self.anthropic.completions.create(
+            model=model,
+            prompt=f"{self.HUMAN_PROMPT} {prompt}{self.AI_PROMPT}",
+            max_tokens_to_sample=max_tokens,
+            stream=True,
+        )
+
+        for completion in stream_response:
+            yield completion.completion

application/llm/llm_creator.py

@@ -2,6 +2,7 @@ from application.llm.openai import OpenAILLM, AzureOpenAILLM
 from application.llm.sagemaker import SagemakerAPILLM
 from application.llm.huggingface import HuggingFaceLLM
 from application.llm.llama_cpp import LlamaCpp
+from application.llm.anthropic import AnthropicLLM
 
 
 
@@ -11,7 +12,8 @@ class LLMCreator:
         'azure_openai': AzureOpenAILLM,
         'sagemaker': SagemakerAPILLM,
         'huggingface': HuggingFaceLLM,
-        'llama.cpp': LlamaCpp
+        'llama.cpp': LlamaCpp,
+        'anthropic': AnthropicLLM
     }
 
     @classmethod

application/requirements.txt

@@ -4,6 +4,7 @@ aiohttp-retry==2.8.3
 aiosignal==1.3.1
 aleph-alpha-client==2.16.1
 amqp==5.1.1
+anthropic==0.5.0
 async-timeout==4.0.2
 attrs==22.2.0
 billiard==3.6.4.0