Update services.py
Browse files- services.py +104 -47
services.py
CHANGED
@@ -3,9 +3,12 @@
|
|
3 |
"""
|
4 |
Manages interactions with external services like LLM providers and web search APIs.
|
5 |
This module has been refactored to support multiple LLM providers:
|
6 |
-
- Hugging Face
|
7 |
-
- Groq
|
8 |
- Fireworks AI
|
|
|
|
|
|
|
9 |
"""
|
10 |
import os
|
11 |
import logging
|
@@ -18,16 +21,23 @@ from huggingface_hub import InferenceClient
|
|
18 |
from tavily import TavilyClient
|
19 |
from groq import Groq
|
20 |
import fireworks.client as Fireworks
|
|
|
|
|
|
|
21 |
|
22 |
# --- Setup Logging & Environment ---
|
23 |
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
24 |
load_dotenv()
|
25 |
|
26 |
-
# --- API Keys ---
|
27 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
28 |
TAVILY_API_KEY = os.getenv("TAVILY_API_KEY")
|
29 |
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
|
30 |
FIREWORKS_API_KEY = os.getenv("FIREWORKS_API_KEY")
|
|
|
|
|
|
|
|
|
31 |
|
32 |
# --- Type Definitions ---
|
33 |
Messages = List[Dict[str, Any]]
|
@@ -36,78 +46,125 @@ class LLMService:
|
|
36 |
"""A multi-provider wrapper for LLM Inference APIs."""
|
37 |
|
38 |
def __init__(self):
|
39 |
-
# Initialize clients if their API keys are available
|
40 |
self.hf_client = InferenceClient(token=HF_TOKEN) if HF_TOKEN else None
|
41 |
self.groq_client = Groq(api_key=GROQ_API_KEY) if GROQ_API_KEY else None
|
42 |
-
self.
|
43 |
-
if
|
44 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
45 |
|
46 |
def generate_code_stream(
|
47 |
-
self, model_id: str, messages: Messages, max_tokens: int =
|
48 |
) -> Generator[str, None, None]:
|
49 |
"""
|
50 |
Streams code generation, dispatching to the correct provider based on model_id.
|
51 |
-
The model_id format is 'provider/model-name' or a full HF model_id.
|
52 |
"""
|
53 |
-
provider =
|
54 |
-
model_name = model_id
|
55 |
-
|
56 |
-
if '/' in model_id:
|
57 |
-
parts = model_id.split('/', 1)
|
58 |
-
if parts[0] in ['groq', 'fireworks', 'huggingface']:
|
59 |
-
provider = parts[0]
|
60 |
-
model_name = parts[1]
|
61 |
-
|
62 |
logging.info(f"Dispatching to provider: {provider} for model: {model_name}")
|
63 |
|
64 |
try:
|
65 |
-
# --- Groq
|
66 |
-
if provider
|
67 |
-
|
68 |
-
|
69 |
-
|
70 |
-
|
71 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
72 |
for chunk in stream:
|
73 |
-
if chunk.choices[0].delta.content:
|
74 |
yield chunk.choices[0].delta.content
|
75 |
|
76 |
-
# ---
|
77 |
-
elif provider == '
|
78 |
-
if not self.
|
79 |
-
raise ValueError("
|
80 |
-
|
81 |
-
|
82 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
83 |
for chunk in stream:
|
84 |
if chunk.choices[0].delta.content:
|
85 |
yield chunk.choices[0].delta.content
|
86 |
|
87 |
-
# --- Hugging Face Provider (Default) ---
|
88 |
else:
|
89 |
-
|
90 |
-
raise ValueError("Hugging Face API token is not configured.")
|
91 |
-
# For HF, the model_name is the full original model_id
|
92 |
-
stream = self.hf_client.chat_completion(
|
93 |
-
model=model_name, messages=messages, stream=True, max_tokens=max_tokens
|
94 |
-
)
|
95 |
-
for chunk in stream:
|
96 |
-
yield chunk.choices[0].delta.content
|
97 |
|
98 |
except Exception as e:
|
99 |
logging.error(f"LLM API Error with provider {provider}: {e}")
|
100 |
yield f"Error from {provider.capitalize()}: {str(e)}"
|
101 |
|
102 |
-
|
103 |
class SearchService:
|
104 |
-
|
105 |
def __init__(self, api_key: str = TAVILY_API_KEY):
|
106 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
107 |
def is_available(self) -> bool:
|
108 |
-
|
|
|
|
|
109 |
def search(self, query: str, max_results: int = 5) -> str:
|
110 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
111 |
|
112 |
# --- Singleton Instances ---
|
113 |
llm_service = LLMService()
|
|
|
3 |
"""
|
4 |
Manages interactions with external services like LLM providers and web search APIs.
|
5 |
This module has been refactored to support multiple LLM providers:
|
6 |
+
- Hugging Face
|
7 |
+
- Groq
|
8 |
- Fireworks AI
|
9 |
+
- OpenAI
|
10 |
+
- Google Gemini
|
11 |
+
- DeepSeek (Direct API)
|
12 |
"""
|
13 |
import os
|
14 |
import logging
|
|
|
21 |
from tavily import TavilyClient
|
22 |
from groq import Groq
|
23 |
import fireworks.client as Fireworks
|
24 |
+
import openai
|
25 |
+
import google.generativeai as genai
|
26 |
+
from deepseek import OpenaiClient as DeepSeekClient
|
27 |
|
28 |
# --- Setup Logging & Environment ---
|
29 |
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
|
30 |
load_dotenv()
|
31 |
|
32 |
+
# --- API Keys from .env ---
|
33 |
HF_TOKEN = os.getenv("HF_TOKEN")
|
34 |
TAVILY_API_KEY = os.getenv("TAVILY_API_KEY")
|
35 |
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
|
36 |
FIREWORKS_API_KEY = os.getenv("FIREWORKS_API_KEY")
|
37 |
+
OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
|
38 |
+
GEMINI_API_KEY = os.getenv("GEMINI_API_KEY")
|
39 |
+
DEEPSEEK_API_KEY = os.getenv("DEEPSEEK_API_KEY")
|
40 |
+
|
41 |
|
42 |
# --- Type Definitions ---
|
43 |
Messages = List[Dict[str, Any]]
|
|
|
46 |
"""A multi-provider wrapper for LLM Inference APIs."""
|
47 |
|
48 |
def __init__(self):
|
49 |
+
# Initialize clients only if their API keys are available
|
50 |
self.hf_client = InferenceClient(token=HF_TOKEN) if HF_TOKEN else None
|
51 |
self.groq_client = Groq(api_key=GROQ_API_KEY) if GROQ_API_KEY else None
|
52 |
+
self.openai_client = openai.OpenAI(api_key=OPENAI_API_KEY) if OPENAI_API_KEY else None
|
53 |
+
self.deepseek_client = DeepSeekClient(api_key=DEEPSEEK_API_KEY) if DEEPSEEK_API_KEY else None
|
54 |
+
|
55 |
+
if FIREWORKS_API_KEY:
|
56 |
+
Fireworks.api_key = FIREWORKS_API_KEY
|
57 |
+
self.fireworks_client = Fireworks
|
58 |
+
else:
|
59 |
+
self.fireworks_client = None
|
60 |
+
|
61 |
+
if GEMINI_API_KEY:
|
62 |
+
genai.configure(api_key=GEMINI_API_KEY)
|
63 |
+
self.gemini_model = genai.GenerativeModel('gemini-1.5-pro-latest')
|
64 |
+
else:
|
65 |
+
self.gemini_model = None
|
66 |
+
|
67 |
+
def _prepare_messages_for_gemini(self, messages: Messages) -> List[Dict[str, Any]]:
|
68 |
+
"""Gemini requires a slightly different message format."""
|
69 |
+
gemini_messages = []
|
70 |
+
for msg in messages:
|
71 |
+
# Gemini uses 'model' for assistant role
|
72 |
+
role = 'model' if msg['role'] == 'assistant' else 'user'
|
73 |
+
gemini_messages.append({'role': role, 'parts': [msg['content']]})
|
74 |
+
return gemini_messages
|
75 |
|
76 |
def generate_code_stream(
|
77 |
+
self, model_id: str, messages: Messages, max_tokens: int = 8192
|
78 |
) -> Generator[str, None, None]:
|
79 |
"""
|
80 |
Streams code generation, dispatching to the correct provider based on model_id.
|
|
|
81 |
"""
|
82 |
+
provider, model_name = model_id.split('/', 1)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
83 |
logging.info(f"Dispatching to provider: {provider} for model: {model_name}")
|
84 |
|
85 |
try:
|
86 |
+
# --- OpenAI, Groq, DeepSeek, Fireworks (OpenAI-compatible) ---
|
87 |
+
if provider in ['openai', 'groq', 'deepseek', 'fireworks']:
|
88 |
+
client_map = {
|
89 |
+
'openai': self.openai_client,
|
90 |
+
'groq': self.groq_client,
|
91 |
+
'deepseek': self.deepseek_client,
|
92 |
+
'fireworks': self.fireworks_client.ChatCompletion if self.fireworks_client else None,
|
93 |
+
}
|
94 |
+
client = client_map.get(provider)
|
95 |
+
if not client:
|
96 |
+
raise ValueError(f"{provider.capitalize()} API key not configured.")
|
97 |
+
|
98 |
+
# Fireworks has a slightly different call signature
|
99 |
+
if provider == 'fireworks':
|
100 |
+
stream = client.create(model=model_name, messages=messages, stream=True, max_tokens=max_tokens)
|
101 |
+
else:
|
102 |
+
stream = client.chat.completions.create(model=model_name, messages=messages, stream=True, max_tokens=max_tokens)
|
103 |
+
|
104 |
for chunk in stream:
|
105 |
+
if chunk.choices and chunk.choices[0].delta and chunk.choices[0].delta.content:
|
106 |
yield chunk.choices[0].delta.content
|
107 |
|
108 |
+
# --- Google Gemini ---
|
109 |
+
elif provider == 'gemini':
|
110 |
+
if not self.gemini_model:
|
111 |
+
raise ValueError("Gemini API key not configured.")
|
112 |
+
gemini_messages = self._prepare_messages_for_gemini(messages)
|
113 |
+
stream = self.gemini_model.generate_content(gemini_messages, stream=True)
|
114 |
+
for chunk in stream:
|
115 |
+
yield chunk.text
|
116 |
+
|
117 |
+
# --- Hugging Face ---
|
118 |
+
elif provider == 'huggingface':
|
119 |
+
if not self.hf_client:
|
120 |
+
raise ValueError("Hugging Face API token not configured.")
|
121 |
+
# For HF, model_name is the rest of the ID, e.g., baidu/ERNIE...
|
122 |
+
hf_model_id = model_id.split('/', 1)[1]
|
123 |
+
stream = self.hf_client.chat_completion(model=hf_model_id, messages=messages, stream=True, max_tokens=max_tokens)
|
124 |
for chunk in stream:
|
125 |
if chunk.choices[0].delta.content:
|
126 |
yield chunk.choices[0].delta.content
|
127 |
|
|
|
128 |
else:
|
129 |
+
raise ValueError(f"Unknown provider: {provider}")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
130 |
|
131 |
except Exception as e:
|
132 |
logging.error(f"LLM API Error with provider {provider}: {e}")
|
133 |
yield f"Error from {provider.capitalize()}: {str(e)}"
|
134 |
|
|
|
135 |
class SearchService:
|
136 |
+
"""A wrapper for the Tavily Search API."""
|
137 |
def __init__(self, api_key: str = TAVILY_API_KEY):
|
138 |
+
if not api_key:
|
139 |
+
logging.warning("TAVILY_API_KEY not set. Web search will be disabled.")
|
140 |
+
self.client = None
|
141 |
+
else:
|
142 |
+
try:
|
143 |
+
self.client = TavilyClient(api_key=api_key)
|
144 |
+
except Exception as e:
|
145 |
+
logging.error(f"Failed to initialize Tavily client: {e}")
|
146 |
+
self.client = None
|
147 |
+
|
148 |
def is_available(self) -> bool:
|
149 |
+
"""Checks if the search service is configured and available."""
|
150 |
+
return self.client is not None
|
151 |
+
|
152 |
def search(self, query: str, max_results: int = 5) -> str:
|
153 |
+
"""Performs a web search and returns a formatted string of results."""
|
154 |
+
if not self.is_available():
|
155 |
+
return "Web search is not available."
|
156 |
+
try:
|
157 |
+
response = self.client.search(
|
158 |
+
query, search_depth="advanced", max_results=min(max(1, max_results), 10)
|
159 |
+
)
|
160 |
+
results = [
|
161 |
+
f"Title: {res.get('title', 'N/A')}\nURL: {res.get('url', 'N/A')}\nContent: {res.get('content', 'N/A')}"
|
162 |
+
for res in response.get('results', [])
|
163 |
+
]
|
164 |
+
return "Web Search Results:\n\n" + "\n---\n".join(results) if results else "No search results found."
|
165 |
+
except Exception as e:
|
166 |
+
logging.error(f"Tavily search error: {e}")
|
167 |
+
return f"Search error: {str(e)}"
|
168 |
|
169 |
# --- Singleton Instances ---
|
170 |
llm_service = LLMService()
|