Spaces:
Running
Running
File size: 22,180 Bytes
900f476 05e73cc 77e475c 900f476 b2fc09f 900f476 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 |
# model_handler.py
import os
import requests
import json
import logging
from dotenv import load_dotenv
# Load environment variables from .env file
load_dotenv()
logging.basicConfig(
level=logging.INFO,
format="%(asctime)s - %(name)s - %(levelname)s - %(message)s"
)
logger = logging.getLogger(__name__)
# Maps provider name (uppercase) to environment variable name for API key
API_KEYS_ENV_VARS = {
"HUGGINGFACE": 'HF_TOKEN', # Note: HF_TOKEN is often used for general HF auth
"GROQ": 'GROQ_API_KEY',
"OPENROUTER": 'OPENROUTER_API_KEY',
"TOGETHERAI": 'TOGETHERAI_API_KEY',
"COHERE": 'COHERE_API_KEY',
"XAI": 'XAI_API_KEY',
"OPENAI": 'OPENAI_API_KEY',
"GOOGLE": 'GOOGLE_API_KEY', # Or GOOGLE_GEMINI_API_KEY etc.
}
API_URLS = {
"HUGGINGFACE": 'https://api-inference.huggingface.co/models/',
"GROQ": 'https://api.groq.com/openai/v1/chat/completions',
"OPENROUTER": 'https://openrouter.ai/api/v1/chat/completions',
"TOGETHERAI": 'https://api.together.ai/v1/chat/completions',
"COHERE": 'https://api.cohere.ai/v1/chat', # v1 is common for chat, was v2 in ai-learn
"XAI": 'https://api.x.ai/v1/chat/completions',
"OPENAI": 'https://api.openai.com/v1/chat/completions',
"GOOGLE": 'https://generativelanguage.googleapis.com/v1beta/models/',
}
MODELS_BY_PROVIDER = json.load(open("./models.json"))
def _get_api_key(provider: str, ui_api_key_override: str = None) -> str | None:
"""
Retrieves API key for a given provider.
Priority: UI Override > Environment Variable from API_KEYS_ENV_VARS > Specific (e.g. HF_TOKEN for HuggingFace).
"""
provider_upper = provider.upper()
if ui_api_key_override and ui_api_key_override.strip():
logger.debug(f"Using UI-provided API key for {provider_upper}.")
return ui_api_key_override.strip()
env_var_name = API_KEYS_ENV_VARS.get(provider_upper)
if env_var_name:
env_key = os.getenv(env_var_name)
if env_key and env_key.strip():
logger.debug(f"Using API key from env var '{env_var_name}' for {provider_upper}.")
return env_key.strip()
# Specific fallback for HuggingFace if HF_TOKEN is set and API_KEYS_ENV_VARS['HUGGINGFACE'] wasn't specific enough
if provider_upper == 'HUGGINGFACE':
hf_token_fallback = os.getenv("HF_TOKEN")
if hf_token_fallback and hf_token_fallback.strip():
logger.debug("Using HF_TOKEN as fallback for HuggingFace provider.")
return hf_token_fallback.strip()
logger.warning(f"API Key not found for provider '{provider_upper}'. Checked UI override and environment variable '{env_var_name or 'N/A'}'.")
return None
def get_available_providers() -> list[str]:
"""Returns a sorted list of available provider names (e.g., 'groq', 'openai')."""
return sorted(list(MODELS_BY_PROVIDER.keys()))
def get_model_display_names_for_provider(provider: str) -> list[str]:
"""Returns a sorted list of model display names for a given provider."""
return sorted(list(MODELS_BY_PROVIDER.get(provider.lower(), {}).get("models", {}).keys()))
def get_default_model_display_name_for_provider(provider: str) -> str | None:
"""Gets the default model's display name for a provider."""
provider_data = MODELS_BY_PROVIDER.get(provider.lower(), {})
models_dict = provider_data.get("models", {})
default_model_id = provider_data.get("default")
if default_model_id and models_dict:
for display_name, model_id_val in models_dict.items():
if model_id_val == default_model_id:
return display_name
# Fallback to the first model in the sorted list if default not found or not set
if models_dict:
#sorted_display_names = sorted(list(models_dict.keys()))
sorted_display_names = list(models_dict.keys())
if sorted_display_names:
return sorted_display_names[0]
return None
def get_model_id_from_display_name(provider: str, display_name: str) -> str | None:
"""Gets the actual model ID from its display name for a given provider."""
models = MODELS_BY_PROVIDER.get(provider.lower(), {}).get("models", {})
return models.get(display_name)
def call_model_stream(provider: str, model_display_name: str, messages: list[dict], api_key_override: str = None, temperature: float = 0.7, max_tokens: int = None) -> iter:
"""
Calls the specified model via its provider and streams the response.
Handles provider-specific request formatting and error handling.
Yields chunks of the response text or an error string.
"""
provider_lower = provider.lower()
api_key = _get_api_key(provider_lower, api_key_override)
base_url = API_URLS.get(provider.upper())
model_id = get_model_id_from_display_name(provider_lower, model_display_name)
if not api_key:
env_var_name = API_KEYS_ENV_VARS.get(provider.upper(), 'N/A')
yield f"Error: API Key not found for {provider}. Please set it in the UI or env var '{env_var_name}'."
return
if not base_url:
yield f"Error: Unknown provider '{provider}' or missing API URL configuration."
return
if not model_id:
yield f"Error: Model ID not found for '{model_display_name}' under provider '{provider}'. Check configuration."
return
headers = {}
payload = {}
request_url = base_url
logger.info(f"Streaming from {provider}/{model_display_name} (ID: {model_id})...")
# --- Standard OpenAI-compatible providers ---
if provider_lower in ["groq", "openrouter", "togetherai", "openai", "xai"]:
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
payload = {"model": model_id, "messages": messages, "stream": True, "temperature": temperature}
if max_tokens: payload["max_tokens"] = max_tokens
if provider_lower == "openrouter":
headers["HTTP-Referer"] = os.getenv("OPENROUTER_REFERRER") or "http://localhost/gradio" # Example Referer
headers["X-Title"] = os.getenv("OPENROUTER_X_TITLE") or "Gradio AI Researcher" # Example Title
try:
response = requests.post(request_url, headers=headers, json=payload, stream=True, timeout=180)
response.raise_for_status()
# More robust SSE parsing
buffer = ""
for chunk in response.iter_content(chunk_size=None): # Process raw bytes
buffer += chunk.decode('utf-8', errors='replace')
while '\n\n' in buffer:
event_str, buffer = buffer.split('\n\n', 1)
if not event_str.strip(): continue
content_chunk = ""
for line in event_str.splitlines():
if line.startswith('data: '):
data_json = line[len('data: '):].strip()
if data_json == '[DONE]':
return # Stream finished
try:
data = json.loads(data_json)
if data.get("choices") and len(data["choices"]) > 0:
delta = data["choices"][0].get("delta", {})
if delta and delta.get("content"):
content_chunk += delta["content"]
except json.JSONDecodeError:
logger.warning(f"Failed to decode JSON from stream line: {data_json}")
if content_chunk:
yield content_chunk
# Process any remaining buffer content (less common with '\n\n' delimiter)
if buffer.strip():
logger.debug(f"Remaining buffer after OpenAI-like stream: {buffer}")
except requests.exceptions.HTTPError as e:
err_msg = f"API HTTP Error ({e.response.status_code}): {e.response.text[:500]}"
logger.error(f"{err_msg} for {provider}/{model_id}", exc_info=False)
yield f"Error: {err_msg}"
except requests.exceptions.RequestException as e:
logger.error(f"API Request Error for {provider}/{model_id}: {e}", exc_info=False)
yield f"Error: Could not connect to {provider} ({e})"
except Exception as e:
logger.exception(f"Unexpected error during {provider} stream:")
yield f"Error: An unexpected error occurred: {e}"
return
# --- Google Gemini ---
elif provider_lower == "google":
system_instruction = None
filtered_messages = []
for msg in messages:
if msg["role"] == "system": system_instruction = {"parts": [{"text": msg["content"]}]}
else:
role = "model" if msg["role"] == "assistant" else msg["role"]
filtered_messages.append({"role": role, "parts": [{"text": msg["content"]}]})
payload = {
"contents": filtered_messages,
"safetySettings": [ # Example: more permissive settings
{"category": "HARM_CATEGORY_SEXUALLY_EXPLICIT", "threshold": "BLOCK_NONE"},
{"category": "HARM_CATEGORY_HATE_SPEECH", "threshold": "BLOCK_NONE"},
{"category": "HARM_CATEGORY_HARASSMENT", "threshold": "BLOCK_NONE"},
{"category": "HARM_CATEGORY_DANGEROUS_CONTENT", "threshold": "BLOCK_NONE"},
],
"generationConfig": {"temperature": temperature}
}
if max_tokens: payload["generationConfig"]["maxOutputTokens"] = max_tokens
if system_instruction: payload["system_instruction"] = system_instruction
request_url = f"{base_url}{model_id}:streamGenerateContent?key={api_key}" # API key in query param
headers = {"Content-Type": "application/json"}
try:
response = requests.post(request_url, headers=headers, json=payload, stream=True, timeout=180)
response.raise_for_status()
# Google's stream is a bit different, often newline-delimited JSON arrays/objects
buffer = ""
for chunk in response.iter_content(chunk_size=None):
buffer += chunk.decode('utf-8', errors='replace')
# Google might send chunks that are not complete JSON objects, or multiple objects
# A common pattern is [ {obj1} , {obj2} ] where chunks split mid-array or mid-object.
# This parsing needs to be robust. A simple split by '\n' might not always work if JSON is pretty-printed.
# The previous code's `json.loads(f"[{decoded_line}]")` was an attempt to handle this.
# For now, let's assume newline delimited for simplicity, but this is a known tricky part.
while '\n' in buffer:
line, buffer = buffer.split('\n', 1)
line = line.strip()
if not line: continue
if line.startswith(','): line = line[1:] # Handle leading commas if splitting an array
try:
# Remove "data: " prefix if present (less common for Gemini direct API but good practice)
if line.startswith('data: '): line = line[len('data: '):]
# Gemini often streams an array of objects, or just one object.
# Try to parse as a single object first. If fails, try as array.
parsed_data = None
try:
parsed_data = json.loads(line)
except json.JSONDecodeError:
# If it's part of an array, it might be missing brackets.
# This heuristic is fragile. A proper SSE parser or stateful JSON parser is better.
if line.startswith('{') and line.endswith('}'): # Looks like a complete object
pass # already tried json.loads
# Try to wrap with [] if it seems like a list content without brackets
elif line.startswith('{') or line.endswith('}'):
try:
temp_parsed_list = json.loads(f"[{line}]")
if temp_parsed_list and isinstance(temp_parsed_list, list):
parsed_data = temp_parsed_list[0] # take first if it becomes a list
except json.JSONDecodeError:
logger.warning(f"Google: Still can't parse line even with array wrap: {line}")
if parsed_data:
data_to_process = [parsed_data] if isinstance(parsed_data, dict) else parsed_data # Ensure list
for event_data in data_to_process:
if not isinstance(event_data, dict): continue
if event_data.get("candidates"):
for candidate in event_data["candidates"]:
if candidate.get("content", {}).get("parts"):
for part in candidate["content"]["parts"]:
if part.get("text"):
yield part["text"]
except json.JSONDecodeError:
logger.warning(f"Google: JSONDecodeError for line: {line}")
except Exception as e_google_proc:
logger.error(f"Google: Error processing stream data: {e_google_proc}, Line: {line}")
except requests.exceptions.HTTPError as e:
err_msg = f"Google API HTTP Error ({e.response.status_code}): {e.response.text[:500]}"
logger.error(err_msg, exc_info=False)
yield f"Error: {err_msg}"
except Exception as e:
logger.exception(f"Unexpected error during Google stream:")
yield f"Error: An unexpected error occurred with Google API: {e}"
return
# --- Cohere ---
elif provider_lower == "cohere":
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json", "Accept": "application/json"}
# Cohere message format
chat_history_cohere = []
preamble_cohere = None
user_message_cohere = ""
temp_messages = list(messages) # Work with a copy
if temp_messages and temp_messages[0]["role"] == "system":
preamble_cohere = temp_messages.pop(0)["content"]
if temp_messages:
user_message_cohere = temp_messages.pop()["content"] # Last message is the current user query
for msg in temp_messages: # Remaining are history
role = "USER" if msg["role"] == "user" else "CHATBOT"
chat_history_cohere.append({"role": role, "message": msg["content"]})
if not user_message_cohere:
yield "Error: User message is empty for Cohere."
return
payload = {
"model": model_id,
"message": user_message_cohere,
"stream": True,
"temperature": temperature
}
if max_tokens: payload["max_tokens"] = max_tokens # Cohere uses max_tokens
if chat_history_cohere: payload["chat_history"] = chat_history_cohere
if preamble_cohere: payload["preamble"] = preamble_cohere
try:
response = requests.post(base_url, headers=headers, json=payload, stream=True, timeout=180)
response.raise_for_status()
# Cohere SSE format is event: type\ndata: {json}\n\n
buffer = ""
for chunk_bytes in response.iter_content(chunk_size=None):
buffer += chunk_bytes.decode('utf-8', errors='replace')
while '\n\n' in buffer:
event_str, buffer = buffer.split('\n\n', 1)
if not event_str.strip(): continue
event_type = None
data_json_str = None
for line in event_str.splitlines():
if line.startswith("event:"): event_type = line[len("event:"):].strip()
elif line.startswith("data:"): data_json_str = line[len("data:"):].strip()
if data_json_str:
try:
data = json.loads(data_json_str)
if event_type == "text-generation" and "text" in data:
yield data["text"]
elif event_type == "stream-end":
logger.debug(f"Cohere stream ended. Finish reason: {data.get('finish_reason')}")
return
except json.JSONDecodeError:
logger.warning(f"Cohere: Failed to decode JSON: {data_json_str}")
if buffer.strip():
logger.debug(f"Cohere: Remaining buffer: {buffer.strip()}")
except requests.exceptions.HTTPError as e:
err_msg = f"Cohere API HTTP Error ({e.response.status_code}): {e.response.text[:500]}"
logger.error(err_msg, exc_info=False)
yield f"Error: {err_msg}"
except Exception as e:
logger.exception(f"Unexpected error during Cohere stream:")
yield f"Error: An unexpected error occurred with Cohere API: {e}"
return
# --- HuggingFace Inference API (Basic TGI support) ---
# This is very basic and might not work for all models or complex scenarios.
# Assumes model is deployed with Text Generation Inference (TGI) and supports streaming.
elif provider_lower == "huggingface":
headers = {"Authorization": f"Bearer {api_key}", "Content-Type": "application/json"}
# Construct prompt string for TGI (often needs specific formatting)
# This is a generic attempt, specific models might need <|user|>, <|assistant|> etc.
prompt_parts = []
for msg in messages:
role_prefix = ""
if msg['role'] == 'system': role_prefix = "System: " # Or might be ignored/handled differently
elif msg['role'] == 'user': role_prefix = "User: "
elif msg['role'] == 'assistant': role_prefix = "Assistant: "
prompt_parts.append(f"{role_prefix}{msg['content']}")
# TGI typically expects a final "Assistant: " to start generating from
tgi_prompt = "\n".join(prompt_parts) + "\nAssistant: "
payload = {
"inputs": tgi_prompt,
"parameters": {
"temperature": temperature if temperature > 0 else 0.01, # TGI needs temp > 0 for sampling
"max_new_tokens": max_tokens or 1024, # Default TGI max_new_tokens
"return_full_text": False, # We only want generated part
"do_sample": True if temperature > 0 else False,
},
"stream": True
}
request_url = f"{base_url}{model_id}" # Model ID is part of URL path for HF
try:
response = requests.post(request_url, headers=headers, json=payload, stream=True, timeout=180)
response.raise_for_status()
# TGI SSE stream: data: {"token": {"id": ..., "text": "...", "logprob": ..., "special": ...}}
# Or sometimes just data: "text_chunk" for simpler models/configs
buffer = ""
for chunk_bytes in response.iter_content(chunk_size=None):
buffer += chunk_bytes.decode('utf-8', errors='replace')
while '\n' in buffer: # TGI often uses single newline
line, buffer = buffer.split('\n', 1)
line = line.strip()
if not line: continue
if line.startswith('data:'):
data_json_str = line[len('data:'):].strip()
try:
data = json.loads(data_json_str)
if "token" in data and "text" in data["token"]:
yield data["token"]["text"]
elif "generated_text" in data and data.get("details") is None: # Sometimes a final non-streaming like object might appear
# This case is tricky, if it's the *only* thing then it's not really streaming
pass # For now, ignore if it's not a token object
# Some TGI might send raw text if not fully SSE compliant for stream
# elif isinstance(data, str): yield data
except json.JSONDecodeError:
# If it's not JSON, it might be a raw string (less common for TGI stream=True)
# For safety, only yield if it's a clear text string
if not data_json_str.startswith('{') and not data_json_str.startswith('['):
yield data_json_str
else:
logger.warning(f"HF: Failed to decode JSON and not raw string: {data_json_str}")
if buffer.strip():
logger.debug(f"HF: Remaining buffer: {buffer.strip()}")
except requests.exceptions.HTTPError as e:
err_msg = f"HF API HTTP Error ({e.response.status_code}): {e.response.text[:500]}"
logger.error(err_msg, exc_info=False)
yield f"Error: {err_msg}"
except Exception as e:
logger.exception(f"Unexpected error during HF stream:")
yield f"Error: An unexpected error occurred with HF API: {e}"
return
else:
yield f"Error: Provider '{provider}' is not configured for streaming in this handler."
return |