ParamManagerScripts/services/llm/claude_service.py

104 lines
3.9 KiB
Python

# services/llm/claude_service.py
"""
Claude (Anthropic) service implementation
"""
import anthropic
from typing import Dict, List
import json
from .base import LLMService
from config.api_keys import APIKeyManager
from utils.logger import setup_logger
class ClaudeService(LLMService):
def __init__(
self,
model: str = "claude-3-7-sonnet-20250219", # Debemos usar el modelo claude-3-7-sonnet-20250219
temperature: float = 0.3,
max_tokens: int = 16000,
):
api_key = APIKeyManager.get_claude_key()
if not api_key:
raise ValueError(
"Claude API key not found. Please set the CLAUDE_API_KEY environment variable."
)
self.client = anthropic.Anthropic(api_key=api_key)
self.model = model
self.temperature = temperature
self.max_tokens = max_tokens
self.logger = setup_logger("claude")
def generate_text(self, prompt: str) -> str:
self.logger.info(f"--- PROMPT ---\n{prompt}")
try:
message = self.client.messages.create(
model=self.model,
max_tokens=self.max_tokens,
temperature=self.temperature,
messages=[
{
"role": "user",
"content": prompt,
}
],
)
response_content = message.content[0].text
self.logger.info(f"--- RESPONSE ---\n{response_content}")
return response_content
except Exception as e:
self.logger.error(f"Error in Claude API call: {e}")
print(f"Error in Claude API call: {e}")
return None
def get_similarity_scores(self, texts_pairs: Dict[str, List[str]]) -> List[float]:
# Claude's API doesn't have a dedicated similarity or JSON mode endpoint as straightforward as others.
# We will instruct it to return JSON.
system_prompt = (
"You are an expert in semantic analysis. Evaluate the semantic similarity between the pairs of texts provided. "
"Return your response ONLY as a JSON object containing a single key 'similarity_scores' with a list of floats from 0.0 to 1.0. "
"Do not include any other text, explanation, or markdown formatting. The output must be a valid JSON."
)
request_payload = json.dumps(texts_pairs)
try:
message = self.client.messages.create(
model=self.model,
max_tokens=self.max_tokens,
temperature=self.temperature,
system=system_prompt,
messages=[
{
"role": "user",
"content": request_payload,
}
],
)
response_content = message.content[0].text
try:
# Find the JSON part of the response
json_start = response_content.find("{")
json_end = response_content.rfind("}") + 1
if json_start == -1 or json_end == 0:
raise ValueError("No JSON object found in the response.")
json_str = response_content[json_start:json_end]
scores_data = json.loads(json_str)
if isinstance(scores_data, dict) and "similarity_scores" in scores_data:
return scores_data["similarity_scores"]
else:
raise ValueError("Unexpected JSON format from Claude.")
except (json.JSONDecodeError, ValueError) as e:
print(f"Error decoding Claude JSON response: {e}")
raise ValueError(
"Could not decode or parse similarity scores from Claude response."
)
except Exception as e:
print(f"Error in Claude similarity calculation: {e}")
return None