Add File
This commit is contained in:
692
src/landppt/ai/providers.py
Normal file
692
src/landppt/ai/providers.py
Normal file
@@ -0,0 +1,692 @@
|
|||||||
|
"""
|
||||||
|
AI provider implementations
|
||||||
|
"""
|
||||||
|
|
||||||
|
import asyncio
|
||||||
|
import json
|
||||||
|
import logging
|
||||||
|
from typing import List, Dict, Any, Optional, AsyncGenerator, Union, Tuple
|
||||||
|
|
||||||
|
from .base import AIProvider, AIMessage, AIResponse, MessageRole, TextContent, ImageContent, MessageContentType
|
||||||
|
from ..core.config import ai_config
|
||||||
|
|
||||||
|
logger = logging.getLogger(__name__)
|
||||||
|
|
||||||
|
class OpenAIProvider(AIProvider):
|
||||||
|
"""OpenAI API provider"""
|
||||||
|
|
||||||
|
def __init__(self, config: Dict[str, Any]):
|
||||||
|
super().__init__(config)
|
||||||
|
try:
|
||||||
|
import openai
|
||||||
|
self.client = openai.AsyncOpenAI(
|
||||||
|
api_key=config.get("api_key"),
|
||||||
|
base_url=config.get("base_url")
|
||||||
|
)
|
||||||
|
except ImportError:
|
||||||
|
logger.warning("OpenAI library not installed. Install with: pip install openai")
|
||||||
|
self.client = None
|
||||||
|
|
||||||
|
def _convert_message_to_openai(self, message: AIMessage) -> Dict[str, Any]:
|
||||||
|
"""Convert AIMessage to OpenAI format, supporting multimodal content"""
|
||||||
|
openai_message = {"role": message.role.value}
|
||||||
|
|
||||||
|
if isinstance(message.content, str):
|
||||||
|
# Simple text message
|
||||||
|
openai_message["content"] = message.content
|
||||||
|
elif isinstance(message.content, list):
|
||||||
|
# Multimodal message
|
||||||
|
content_parts = []
|
||||||
|
for part in message.content:
|
||||||
|
if isinstance(part, TextContent):
|
||||||
|
content_parts.append({
|
||||||
|
"type": "text",
|
||||||
|
"text": part.text
|
||||||
|
})
|
||||||
|
elif isinstance(part, ImageContent):
|
||||||
|
content_parts.append({
|
||||||
|
"type": "image_url",
|
||||||
|
"image_url": part.image_url
|
||||||
|
})
|
||||||
|
openai_message["content"] = content_parts
|
||||||
|
else:
|
||||||
|
# Fallback to string representation
|
||||||
|
openai_message["content"] = str(message.content)
|
||||||
|
|
||||||
|
if message.name:
|
||||||
|
openai_message["name"] = message.name
|
||||||
|
|
||||||
|
return openai_message
|
||||||
|
|
||||||
|
async def chat_completion(self, messages: List[AIMessage], **kwargs) -> AIResponse:
|
||||||
|
"""Generate chat completion using OpenAI"""
|
||||||
|
if not self.client:
|
||||||
|
raise RuntimeError("OpenAI client not available")
|
||||||
|
|
||||||
|
config = self._merge_config(**kwargs)
|
||||||
|
|
||||||
|
# Convert messages to OpenAI format with multimodal support
|
||||||
|
openai_messages = [
|
||||||
|
self._convert_message_to_openai(msg)
|
||||||
|
for msg in messages
|
||||||
|
]
|
||||||
|
|
||||||
|
try:
|
||||||
|
response = await self.client.chat.completions.create(
|
||||||
|
model=config.get("model", self.model),
|
||||||
|
messages=openai_messages,
|
||||||
|
# max_tokens=config.get("max_tokens", 2000),
|
||||||
|
temperature=config.get("temperature", 0.7),
|
||||||
|
top_p=config.get("top_p", 1.0)
|
||||||
|
)
|
||||||
|
|
||||||
|
choice = response.choices[0]
|
||||||
|
|
||||||
|
return AIResponse(
|
||||||
|
content=choice.message.content,
|
||||||
|
model=response.model,
|
||||||
|
usage={
|
||||||
|
"prompt_tokens": response.usage.prompt_tokens,
|
||||||
|
"completion_tokens": response.usage.completion_tokens,
|
||||||
|
"total_tokens": response.usage.total_tokens
|
||||||
|
},
|
||||||
|
finish_reason=choice.finish_reason,
|
||||||
|
metadata={"provider": "openai"}
|
||||||
|
)
|
||||||
|
|
||||||
|
except Exception as e:
|
||||||
|
# 提供更详细的错误信息
|
||||||
|
error_msg = str(e)
|
||||||
|
if "Expecting value" in error_msg:
|
||||||
|
logger.error(f"OpenAI API JSON parsing error: {error_msg}. This usually indicates the API returned malformed JSON.")
|
||||||
|
elif "timeout" in error_msg.lower():
|
||||||
|
logger.error(f"OpenAI API timeout error: {error_msg}")
|
||||||
|
elif "rate limit" in error_msg.lower():
|
||||||
|
logger.error(f"OpenAI API rate limit error: {error_msg}")
|
||||||
|
else:
|
||||||
|
logger.error(f"OpenAI API error: {error_msg}")
|
||||||
|
raise
|
||||||
|
|
||||||
|
async def text_completion(self, prompt: str, **kwargs) -> AIResponse:
|
||||||
|
"""Generate text completion using OpenAI chat format"""
|
||||||
|
messages = [AIMessage(role=MessageRole.USER, content=prompt)]
|
||||||
|
return await self.chat_completion(messages, **kwargs)
|
||||||
|
|
||||||
|
async def stream_chat_completion(self, messages: List[AIMessage], **kwargs) -> AsyncGenerator[str, None]:
|
||||||
|
"""Stream chat completion using OpenAI"""
|
||||||
|
if not self.client:
|
||||||
|
raise RuntimeError("OpenAI client not available")
|
||||||
|
|
||||||
|
config = self._merge_config(**kwargs)
|
||||||
|
|
||||||
|
# Convert messages to OpenAI format with multimodal support
|
||||||
|
openai_messages = [
|
||||||
|
self._convert_message_to_openai(msg)
|
||||||
|
for msg in messages
|
||||||
|
]
|
||||||
|
|
||||||
|
try:
|
||||||
|
stream = await self.client.chat.completions.create(
|
||||||
|
model=config.get("model", self.model),
|
||||||
|
messages=openai_messages,
|
||||||
|
# max_tokens=config.get("max_tokens", 2000),
|
||||||
|
temperature=config.get("temperature", 0.7),
|
||||||
|
top_p=config.get("top_p", 1.0),
|
||||||
|
stream=True
|
||||||
|
)
|
||||||
|
|
||||||
|
async for chunk in stream:
|
||||||
|
if chunk.choices and chunk.choices[0].delta.content:
|
||||||
|
yield chunk.choices[0].delta.content
|
||||||
|
|
||||||
|
except Exception as e:
|
||||||
|
logger.error(f"OpenAI streaming error: {e}")
|
||||||
|
raise
|
||||||
|
|
||||||
|
async def stream_text_completion(self, prompt: str, **kwargs) -> AsyncGenerator[str, None]:
|
||||||
|
"""Stream text completion using OpenAI chat format"""
|
||||||
|
messages = [AIMessage(role=MessageRole.USER, content=prompt)]
|
||||||
|
async for chunk in self.stream_chat_completion(messages, **kwargs):
|
||||||
|
yield chunk
|
||||||
|
|
||||||
|
class AnthropicProvider(AIProvider):
|
||||||
|
"""Anthropic Claude API provider"""
|
||||||
|
|
||||||
|
def __init__(self, config: Dict[str, Any]):
|
||||||
|
super().__init__(config)
|
||||||
|
try:
|
||||||
|
import anthropic
|
||||||
|
self.client = anthropic.AsyncAnthropic(
|
||||||
|
api_key=config.get("api_key")
|
||||||
|
)
|
||||||
|
except ImportError:
|
||||||
|
logger.warning("Anthropic library not installed. Install with: pip install anthropic")
|
||||||
|
self.client = None
|
||||||
|
|
||||||
|
def _convert_message_to_anthropic(self, message: AIMessage) -> Dict[str, Any]:
|
||||||
|
"""Convert AIMessage to Anthropic format, supporting multimodal content"""
|
||||||
|
anthropic_message = {"role": message.role.value}
|
||||||
|
|
||||||
|
if isinstance(message.content, str):
|
||||||
|
# Simple text message
|
||||||
|
anthropic_message["content"] = message.content
|
||||||
|
elif isinstance(message.content, list):
|
||||||
|
# Multimodal message
|
||||||
|
content_parts = []
|
||||||
|
for part in message.content:
|
||||||
|
if isinstance(part, TextContent):
|
||||||
|
content_parts.append({
|
||||||
|
"type": "text",
|
||||||
|
"text": part.text
|
||||||
|
})
|
||||||
|
elif isinstance(part, ImageContent):
|
||||||
|
# Anthropic expects base64 data without the data URL prefix
|
||||||
|
image_url = part.image_url.get("url", "")
|
||||||
|
if image_url.startswith("data:image/"):
|
||||||
|
# Extract base64 data and media type
|
||||||
|
header, base64_data = image_url.split(",", 1)
|
||||||
|
media_type = header.split(":")[1].split(";")[0]
|
||||||
|
content_parts.append({
|
||||||
|
"type": "image",
|
||||||
|
"source": {
|
||||||
|
"type": "base64",
|
||||||
|
"media_type": media_type,
|
||||||
|
"data": base64_data
|
||||||
|
}
|
||||||
|
})
|
||||||
|
else:
|
||||||
|
# For URL-based images, we'd need to fetch and convert to base64
|
||||||
|
# For now, skip or convert to text description
|
||||||
|
content_parts.append({
|
||||||
|
"type": "text",
|
||||||
|
"text": f"[Image: {image_url}]"
|
||||||
|
})
|
||||||
|
anthropic_message["content"] = content_parts
|
||||||
|
else:
|
||||||
|
# Fallback to string representation
|
||||||
|
anthropic_message["content"] = str(message.content)
|
||||||
|
|
||||||
|
return anthropic_message
|
||||||
|
|
||||||
|
async def chat_completion(self, messages: List[AIMessage], **kwargs) -> AIResponse:
|
||||||
|
"""Generate chat completion using Anthropic Claude"""
|
||||||
|
if not self.client:
|
||||||
|
raise RuntimeError("Anthropic client not available")
|
||||||
|
|
||||||
|
config = self._merge_config(**kwargs)
|
||||||
|
|
||||||
|
# Convert messages to Anthropic format
|
||||||
|
system_message = None
|
||||||
|
claude_messages = []
|
||||||
|
|
||||||
|
for msg in messages:
|
||||||
|
if msg.role == MessageRole.SYSTEM:
|
||||||
|
# System messages should be simple text for Anthropic
|
||||||
|
system_message = msg.content if isinstance(msg.content, str) else str(msg.content)
|
||||||
|
else:
|
||||||
|
claude_messages.append(self._convert_message_to_anthropic(msg))
|
||||||
|
|
||||||
|
try:
|
||||||
|
response = await self.client.messages.create(
|
||||||
|
model=config.get("model", self.model),
|
||||||
|
# max_tokens=config.get("max_tokens", 2000),
|
||||||
|
temperature=config.get("temperature", 0.7),
|
||||||
|
system=system_message,
|
||||||
|
messages=claude_messages
|
||||||
|
)
|
||||||
|
|
||||||
|
content = response.content[0].text if response.content else ""
|
||||||
|
|
||||||
|
return AIResponse(
|
||||||
|
content=content,
|
||||||
|
model=response.model,
|
||||||
|
usage={
|
||||||
|
"prompt_tokens": response.usage.input_tokens,
|
||||||
|
"completion_tokens": response.usage.output_tokens,
|
||||||
|
"total_tokens": response.usage.input_tokens + response.usage.output_tokens
|
||||||
|
},
|
||||||
|
finish_reason=response.stop_reason,
|
||||||
|
metadata={"provider": "anthropic"}
|
||||||
|
)
|
||||||
|
|
||||||
|
except Exception as e:
|
||||||
|
logger.error(f"Anthropic API error: {e}")
|
||||||
|
raise
|
||||||
|
|
||||||
|
async def text_completion(self, prompt: str, **kwargs) -> AIResponse:
|
||||||
|
"""Generate text completion using Anthropic chat format"""
|
||||||
|
messages = [AIMessage(role=MessageRole.USER, content=prompt)]
|
||||||
|
return await self.chat_completion(messages, **kwargs)
|
||||||
|
|
||||||
|
class GoogleProvider(AIProvider):
|
||||||
|
"""Google Gemini API provider"""
|
||||||
|
|
||||||
|
def __init__(self, config: Dict[str, Any]):
|
||||||
|
super().__init__(config)
|
||||||
|
try:
|
||||||
|
import google.generativeai as genai
|
||||||
|
|
||||||
|
# Configure the API key
|
||||||
|
genai.configure(api_key=config.get("api_key"))
|
||||||
|
|
||||||
|
# Store base_url for potential future use or proxy configurations
|
||||||
|
self.base_url = config.get("base_url", "https://generativelanguage.googleapis.com")
|
||||||
|
|
||||||
|
self.client = genai
|
||||||
|
self.model_instance = genai.GenerativeModel(config.get("model", "gemini-1.5-flash"))
|
||||||
|
except ImportError:
|
||||||
|
logger.warning("Google Generative AI library not installed. Install with: pip install google-generativeai")
|
||||||
|
self.client = None
|
||||||
|
self.model_instance = None
|
||||||
|
|
||||||
|
def _convert_messages_to_gemini(self, messages: List[AIMessage]):
|
||||||
|
"""Convert AIMessage list to Gemini format, supporting multimodal content"""
|
||||||
|
import google.generativeai as genai
|
||||||
|
import base64
|
||||||
|
|
||||||
|
# Try to import genai types for proper image handling
|
||||||
|
try:
|
||||||
|
from google.genai import types
|
||||||
|
GENAI_TYPES_AVAILABLE = True
|
||||||
|
except ImportError:
|
||||||
|
try:
|
||||||
|
# Fallback to older API structure
|
||||||
|
from google.generativeai import types
|
||||||
|
GENAI_TYPES_AVAILABLE = True
|
||||||
|
except ImportError:
|
||||||
|
logger.warning("Google GenAI types not available for proper image processing")
|
||||||
|
GENAI_TYPES_AVAILABLE = False
|
||||||
|
|
||||||
|
# Check if we have any images
|
||||||
|
has_images = any(
|
||||||
|
isinstance(msg.content, list) and
|
||||||
|
any(isinstance(part, ImageContent) for part in msg.content)
|
||||||
|
for msg in messages
|
||||||
|
)
|
||||||
|
|
||||||
|
if not has_images:
|
||||||
|
# Text-only mode - return string
|
||||||
|
parts = []
|
||||||
|
for msg in messages:
|
||||||
|
role_prefix = f"[{msg.role.value.upper()}]: "
|
||||||
|
if isinstance(msg.content, str):
|
||||||
|
parts.append(role_prefix + msg.content)
|
||||||
|
elif isinstance(msg.content, list):
|
||||||
|
message_parts = [role_prefix]
|
||||||
|
for part in msg.content:
|
||||||
|
if isinstance(part, TextContent):
|
||||||
|
message_parts.append(part.text)
|
||||||
|
parts.append(" ".join(message_parts))
|
||||||
|
else:
|
||||||
|
parts.append(role_prefix + str(msg.content))
|
||||||
|
return "\n\n".join(parts)
|
||||||
|
else:
|
||||||
|
# Multimodal mode - return list of parts for Gemini
|
||||||
|
content_parts = []
|
||||||
|
|
||||||
|
for msg in messages:
|
||||||
|
role_prefix = f"[{msg.role.value.upper()}]: "
|
||||||
|
|
||||||
|
if isinstance(msg.content, str):
|
||||||
|
content_parts.append(role_prefix + msg.content)
|
||||||
|
elif isinstance(msg.content, list):
|
||||||
|
text_parts = [role_prefix]
|
||||||
|
|
||||||
|
for part in msg.content:
|
||||||
|
if isinstance(part, TextContent):
|
||||||
|
text_parts.append(part.text)
|
||||||
|
elif isinstance(part, ImageContent):
|
||||||
|
# Add accumulated text first
|
||||||
|
if len(text_parts) > 1 or text_parts[0]:
|
||||||
|
content_parts.append(" ".join(text_parts))
|
||||||
|
text_parts = []
|
||||||
|
|
||||||
|
# Process image for Gemini
|
||||||
|
image_url = part.image_url.get("url", "")
|
||||||
|
if image_url.startswith("data:image/") and GENAI_TYPES_AVAILABLE:
|
||||||
|
try:
|
||||||
|
# Extract base64 data and mime type
|
||||||
|
header, base64_data = image_url.split(",", 1)
|
||||||
|
mime_type = header.split(":")[1].split(";")[0] # Extract mime type like 'image/jpeg'
|
||||||
|
image_data = base64.b64decode(base64_data)
|
||||||
|
|
||||||
|
# Create Gemini-compatible part from base64 image data
|
||||||
|
image_part = None
|
||||||
|
if GENAI_TYPES_AVAILABLE:
|
||||||
|
if hasattr(types, 'Part') and hasattr(types.Part, 'from_bytes'):
|
||||||
|
image_part = types.Part.from_bytes(
|
||||||
|
data=image_data,
|
||||||
|
mime_type=mime_type
|
||||||
|
)
|
||||||
|
elif hasattr(types, 'to_part'):
|
||||||
|
image_part = types.to_part({
|
||||||
|
'inline_data': {
|
||||||
|
'mime_type': mime_type,
|
||||||
|
'data': image_data
|
||||||
|
}
|
||||||
|
})
|
||||||
|
if image_part is None:
|
||||||
|
image_part = {
|
||||||
|
'inline_data': {
|
||||||
|
'mime_type': mime_type,
|
||||||
|
'data': image_data
|
||||||
|
}
|
||||||
|
}
|
||||||
|
content_parts.append(image_part)
|
||||||
|
logger.info(f"Successfully processed image for Gemini: {mime_type}, {len(image_data)} bytes")
|
||||||
|
except Exception as e:
|
||||||
|
logger.error(f"Failed to process image for Gemini: {e}")
|
||||||
|
content_parts.append("请参考上传的图片进行设计。图片包含了重要的设计参考信息，请根据图片的风格、色彩、布局等元素来生成模板。")
|
||||||
|
else:
|
||||||
|
# Fallback when genai types not available or not base64 image
|
||||||
|
if image_url.startswith("data:image/"):
|
||||||
|
content_parts.append("请参考上传的图片进行设计。图片包含了重要的设计参考信息，请根据图片的风格、色彩、布局等元素来生成模板。")
|
||||||
|
else:
|
||||||
|
content_parts.append(f"请参考图片 {image_url} 进行设计")
|
||||||
|
|
||||||
|
# Add remaining text
|
||||||
|
if len(text_parts) > 1 or (len(text_parts) == 1 and text_parts[0]):
|
||||||
|
content_parts.append(" ".join(text_parts))
|
||||||
|
else:
|
||||||
|
content_parts.append(role_prefix + str(msg.content))
|
||||||
|
|
||||||
|
return content_parts
|
||||||
|
|
||||||
|
async def chat_completion(self, messages: List[AIMessage], **kwargs) -> AIResponse:
|
||||||
|
"""Generate chat completion using Google Gemini"""
|
||||||
|
if not self.client or not self.model_instance:
|
||||||
|
raise RuntimeError("Google Gemini client not available")
|
||||||
|
|
||||||
|
config = self._merge_config(**kwargs)
|
||||||
|
|
||||||
|
# Convert messages to Gemini format with multimodal support
|
||||||
|
prompt = self._convert_messages_to_gemini(messages)
|
||||||
|
|
||||||
|
try:
|
||||||
|
# Configure generation parameters
|
||||||
|
# 确保max_tokens不会太小，至少1000个token用于生成内容
|
||||||
|
max_tokens = max(config.get("max_tokens", 16384), 1000)
|
||||||
|
generation_config = {
|
||||||
|
"temperature": config.get("temperature", 0.7),
|
||||||
|
"top_p": config.get("top_p", 1.0),
|
||||||
|
# "max_output_tokens": max_tokens,
|
||||||
|
}
|
||||||
|
|
||||||
|
# 配置安全设置 - 设置为较宽松的安全级别以减少误拦截
|
||||||
|
safety_settings = [
|
||||||
|
{
|
||||||
|
"category": "HARM_CATEGORY_HARASSMENT",
|
||||||
|
"threshold": "BLOCK_ONLY_HIGH"
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"category": "HARM_CATEGORY_HATE_SPEECH",
|
||||||
|
"threshold": "BLOCK_ONLY_HIGH"
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"category": "HARM_CATEGORY_SEXUALLY_EXPLICIT",
|
||||||
|
"threshold": "BLOCK_ONLY_HIGH"
|
||||||
|
},
|
||||||
|
{
|
||||||
|
"category": "HARM_CATEGORY_DANGEROUS_CONTENT",
|
||||||
|
"threshold": "BLOCK_ONLY_HIGH"
|
||||||
|
}
|
||||||
|
]
|
||||||
|
|
||||||
|
|
||||||
|
response = await self._generate_async(prompt, generation_config, safety_settings)
|
||||||
|
logger.debug(f"Google Gemini API response: {response}")
|
||||||
|
|
||||||
|
# 检查响应状态和安全过滤
|
||||||
|
finish_reason = "stop"
|
||||||
|
content = ""
|
||||||
|
|
||||||
|
if response.candidates:
|
||||||
|
candidate = response.candidates[0]
|
||||||
|
finish_reason = candidate.finish_reason.name if hasattr(candidate.finish_reason, 'name') else str(candidate.finish_reason)
|
||||||
|
|
||||||
|
# 检查是否被安全过滤器阻止或其他问题
|
||||||
|
if finish_reason == "SAFETY":
|
||||||
|
logger.warning("Content was blocked by safety filters")
|
||||||
|
content = "[内容被安全过滤器阻止]"
|
||||||
|
elif finish_reason == "RECITATION":
|
||||||
|
logger.warning("Content was blocked due to recitation")
|
||||||
|
content = "[内容因重复而被阻止]"
|
||||||
|
elif finish_reason == "MAX_TOKENS":
|
||||||
|
logger.warning("Response was truncated due to max tokens limit")
|
||||||
|
# 尝试获取部分内容
|
||||||
|
try:
|
||||||
|
if hasattr(candidate, 'content') and candidate.content and hasattr(candidate.content, 'parts') and candidate.content.parts:
|
||||||
|
content = candidate.content.parts[0].text if candidate.content.parts[0].text else "[响应因token限制被截断，无内容]"
|
||||||
|
else:
|
||||||
|
content = "[响应因token限制被截断，无内容]"
|
||||||
|
except Exception as text_error:
|
||||||
|
logger.warning(f"Failed to get truncated response text: {text_error}")
|
||||||
|
content = "[响应因token限制被截断，无法获取内容]"
|
||||||
|
elif finish_reason == "OTHER":
|
||||||
|
logger.warning("Content was blocked for other reasons")
|
||||||
|
content = "[内容被其他原因阻止]"
|
||||||
|
else:
|
||||||
|
# 正常情况下获取文本
|
||||||
|
try:
|
||||||
|
if hasattr(candidate, 'content') and candidate.content and hasattr(candidate.content, 'parts') and candidate.content.parts:
|
||||||
|
content = candidate.content.parts[0].text if candidate.content.parts[0].text else ""
|
||||||
|
else:
|
||||||
|
# 回退到response.text
|
||||||
|
content = response.text if hasattr(response, 'text') and response.text else ""
|
||||||
|
except Exception as text_error:
|
||||||
|
logger.warning(f"Failed to get response text: {text_error}")
|
||||||
|
content = "[无法获取响应内容]"
|
||||||
|
else:
|
||||||
|
logger.warning("No candidates in response")
|
||||||
|
content = "[响应中没有候选内容]"
|
||||||
|
|
||||||
|
return AIResponse(
|
||||||
|
content=content,
|
||||||
|
model=self.model,
|
||||||
|
usage={
|
||||||
|
"prompt_tokens": response.usage_metadata.prompt_token_count if hasattr(response, 'usage_metadata') else 0,
|
||||||
|
"completion_tokens": response.usage_metadata.candidates_token_count if hasattr(response, 'usage_metadata') else 0,
|
||||||
|
"total_tokens": response.usage_metadata.total_token_count if hasattr(response, 'usage_metadata') else 0
|
||||||
|
},
|
||||||
|
finish_reason=finish_reason,
|
||||||
|
metadata={"provider": "google"}
|
||||||
|
)
|
||||||
|
|
||||||
|
except Exception as e:
|
||||||
|
logger.error(f"Google Gemini API error: {e}")
|
||||||
|
raise
|
||||||
|
|
||||||
|
async def _generate_async(self, prompt, generation_config: Dict[str, Any], safety_settings=None):
|
||||||
|
"""Async wrapper for Gemini generation - supports both text and multimodal content"""
|
||||||
|
import asyncio
|
||||||
|
loop = asyncio.get_event_loop()
|
||||||
|
|
||||||
|
def _generate_sync():
|
||||||
|
kwargs = {
|
||||||
|
"generation_config": generation_config
|
||||||
|
}
|
||||||
|
if safety_settings:
|
||||||
|
kwargs["safety_settings"] = safety_settings
|
||||||
|
|
||||||
|
return self.model_instance.generate_content(
|
||||||
|
prompt, # Can be string or list of parts
|
||||||
|
**kwargs
|
||||||
|
)
|
||||||
|
|
||||||
|
return await loop.run_in_executor(None, _generate_sync)
|
||||||
|
|
||||||
|
async def text_completion(self, prompt: str, **kwargs) -> AIResponse:
|
||||||
|
"""Generate text completion using Google Gemini"""
|
||||||
|
messages = [AIMessage(role=MessageRole.USER, content=prompt)]
|
||||||
|
return await self.chat_completion(messages, **kwargs)
|
||||||
|
|
||||||
|
class OllamaProvider(AIProvider):
|
||||||
|
"""Ollama local model provider"""
|
||||||
|
|
||||||
|
def __init__(self, config: Dict[str, Any]):
|
||||||
|
super().__init__(config)
|
||||||
|
try:
|
||||||
|
import ollama
|
||||||
|
self.client = ollama.AsyncClient(host=config.get("base_url", "http://localhost:11434"))
|
||||||
|
except ImportError:
|
||||||
|
logger.warning("Ollama library not installed. Install with: pip install ollama")
|
||||||
|
self.client = None
|
||||||
|
|
||||||
|
async def chat_completion(self, messages: List[AIMessage], **kwargs) -> AIResponse:
|
||||||
|
"""Generate chat completion using Ollama"""
|
||||||
|
if not self.client:
|
||||||
|
raise RuntimeError("Ollama client not available")
|
||||||
|
|
||||||
|
config = self._merge_config(**kwargs)
|
||||||
|
|
||||||
|
# Convert messages to Ollama format with multimodal support
|
||||||
|
ollama_messages = []
|
||||||
|
for msg in messages:
|
||||||
|
if isinstance(msg.content, str):
|
||||||
|
# Simple text message
|
||||||
|
ollama_messages.append({"role": msg.role.value, "content": msg.content})
|
||||||
|
elif isinstance(msg.content, list):
|
||||||
|
# Multimodal message - convert to text description for Ollama
|
||||||
|
content_parts = []
|
||||||
|
for part in msg.content:
|
||||||
|
if isinstance(part, TextContent):
|
||||||
|
content_parts.append(part.text)
|
||||||
|
elif isinstance(part, ImageContent):
|
||||||
|
# Ollama doesn't support images directly, add text description
|
||||||
|
image_url = part.image_url.get("url", "")
|
||||||
|
if image_url.startswith("data:image/"):
|
||||||
|
content_parts.append("[Image provided - base64 data]")
|
||||||
|
else:
|
||||||
|
content_parts.append(f"[Image: {image_url}]")
|
||||||
|
ollama_messages.append({
|
||||||
|
"role": msg.role.value,
|
||||||
|
"content": " ".join(content_parts)
|
||||||
|
})
|
||||||
|
else:
|
||||||
|
# Fallback to string representation
|
||||||
|
ollama_messages.append({"role": msg.role.value, "content": str(msg.content)})
|
||||||
|
|
||||||
|
try:
|
||||||
|
response = await self.client.chat(
|
||||||
|
model=config.get("model", self.model),
|
||||||
|
messages=ollama_messages,
|
||||||
|
options={
|
||||||
|
"temperature": config.get("temperature", 0.7),
|
||||||
|
"top_p": config.get("top_p", 1.0),
|
||||||
|
# "num_predict": config.get("max_tokens", 2000)
|
||||||
|
}
|
||||||
|
)
|
||||||
|
|
||||||
|
content = response.get("message", {}).get("content", "")
|
||||||
|
|
||||||
|
return AIResponse(
|
||||||
|
content=content,
|
||||||
|
model=config.get("model", self.model),
|
||||||
|
usage=self._calculate_usage(
|
||||||
|
" ".join([msg.content for msg in messages]),
|
||||||
|
content
|
||||||
|
),
|
||||||
|
finish_reason="stop",
|
||||||
|
metadata={"provider": "ollama"}
|
||||||
|
)
|
||||||
|
|
||||||
|
except Exception as e:
|
||||||
|
logger.error(f"Ollama API error: {e}")
|
||||||
|
raise
|
||||||
|
|
||||||
|
async def text_completion(self, prompt: str, **kwargs) -> AIResponse:
|
||||||
|
"""Generate text completion using Ollama"""
|
||||||
|
messages = [AIMessage(role=MessageRole.USER, content=prompt)]
|
||||||
|
return await self.chat_completion(messages, **kwargs)
|
||||||
|
|
||||||
|
class AIProviderFactory:
|
||||||
|
"""Factory for creating AI providers"""
|
||||||
|
|
||||||
|
_providers = {
|
||||||
|
"openai": OpenAIProvider,
|
||||||
|
"anthropic": AnthropicProvider,
|
||||||
|
"google": GoogleProvider,
|
||||||
|
"gemini": GoogleProvider, # Alias for google
|
||||||
|
"ollama": OllamaProvider,
|
||||||
|
"302ai": OpenAIProvider # 302.AI uses OpenAI-compatible API
|
||||||
|
}
|
||||||
|
|
||||||
|
@classmethod
|
||||||
|
def create_provider(cls, provider_name: str, config: Optional[Dict[str, Any]] = None) -> AIProvider:
|
||||||
|
"""Create an AI provider instance"""
|
||||||
|
if config is None:
|
||||||
|
config = ai_config.get_provider_config(provider_name)
|
||||||
|
|
||||||
|
# Built-in providers
|
||||||
|
if provider_name not in cls._providers:
|
||||||
|
raise ValueError(f"Unknown provider: {provider_name}")
|
||||||
|
|
||||||
|
provider_class = cls._providers[provider_name]
|
||||||
|
return provider_class(config)
|
||||||
|
|
||||||
|
@classmethod
|
||||||
|
def get_available_providers(cls) -> List[str]:
|
||||||
|
"""Get list of available providers"""
|
||||||
|
return list(cls._providers.keys())
|
||||||
|
|
||||||
|
class AIProviderManager:
|
||||||
|
"""Manager for AI provider instances with caching and reloading"""
|
||||||
|
|
||||||
|
def __init__(self):
|
||||||
|
self._provider_cache = {}
|
||||||
|
self._config_cache = {}
|
||||||
|
|
||||||
|
def get_provider(self, provider_name: Optional[str] = None) -> AIProvider:
|
||||||
|
"""Get AI provider instance with caching"""
|
||||||
|
if provider_name is None:
|
||||||
|
provider_name = ai_config.default_ai_provider
|
||||||
|
|
||||||
|
# Get current config for the provider
|
||||||
|
current_config = ai_config.get_provider_config(provider_name)
|
||||||
|
|
||||||
|
# Check if we have a cached provider and if config has changed
|
||||||
|
cache_key = provider_name
|
||||||
|
if (cache_key in self._provider_cache and
|
||||||
|
cache_key in self._config_cache and
|
||||||
|
self._config_cache[cache_key] == current_config):
|
||||||
|
return self._provider_cache[cache_key]
|
||||||
|
|
||||||
|
# Create new provider instance
|
||||||
|
provider = AIProviderFactory.create_provider(provider_name, current_config)
|
||||||
|
|
||||||
|
# Cache the provider and config
|
||||||
|
self._provider_cache[cache_key] = provider
|
||||||
|
self._config_cache[cache_key] = current_config
|
||||||
|
|
||||||
|
return provider
|
||||||
|
|
||||||
|
def clear_cache(self):
|
||||||
|
"""Clear provider cache to force reload"""
|
||||||
|
self._provider_cache.clear()
|
||||||
|
self._config_cache.clear()
|
||||||
|
|
||||||
|
def reload_provider(self, provider_name: str):
|
||||||
|
"""Reload a specific provider"""
|
||||||
|
cache_key = provider_name
|
||||||
|
if cache_key in self._provider_cache:
|
||||||
|
del self._provider_cache[cache_key]
|
||||||
|
if cache_key in self._config_cache:
|
||||||
|
del self._config_cache[cache_key]
|
||||||
|
|
||||||
|
# Global provider manager
|
||||||
|
_provider_manager = AIProviderManager()
|
||||||
|
|
||||||
|
def get_ai_provider(provider_name: Optional[str] = None) -> AIProvider:
|
||||||
|
"""Get AI provider instance"""
|
||||||
|
return _provider_manager.get_provider(provider_name)
|
||||||
|
|
||||||
|
|
||||||
|
def get_role_provider(role: str, provider_override: Optional[str] = None) -> Tuple[AIProvider, Dict[str, Optional[str]]]:
|
||||||
|
"""Get provider and settings for a specific task role"""
|
||||||
|
settings = ai_config.get_model_config_for_role(role, provider_override=provider_override)
|
||||||
|
provider = get_ai_provider(settings["provider"])
|
||||||
|
return provider, settings
|
||||||
|
|
||||||
|
def reload_ai_providers():
|
||||||
|
"""Reload all AI providers (clear cache)"""
|
||||||
|
_provider_manager.clear_cache()
|
||||||
Reference in New Issue
Block a user