webscout 8.2.7__py3-none-any.whl → 8.2.9__py3-none-any.whl

webscout/Provider/HuggingFaceChat.py

@@ -0,0 +1,469 @@
+from curl_cffi.requests import Session
+from curl_cffi import CurlError
+import uuid
+import json
+import time
+import random
+import re
+from typing import Any, Dict, List, Optional, Union, Generator
+
+from webscout.AIutel import Conversation
+from webscout.AIbase import Provider
+from webscout import exceptions
+from webscout.litagent import LitAgent
+
+class HuggingFaceChat(Provider):
+    """
+    A class to interact with the Hugging Face Chat API.
+    Uses cookies for authentication and supports streaming responses.
+    """
+    
+    # Available models (default models - will be updated dynamically)
+    AVAILABLE_MODELS = ['meta-llama/Llama-3.3-70B-Instruct', 'Qwen/Qwen3-235B-A22B', 'Qwen/Qwen2.5-72B-Instruct', 'CohereForAI/c4ai-command-r-plus-08-2024', 'deepseek-ai/DeepSeek-R1-Distill-Qwen-32B', 'nvidia/Llama-3.1-Nemotron-70B-Instruct-HF', 'Qwen/QwQ-32B', 'google/gemma-3-27b-it', 'mistralai/Mistral-Small-3.1-24B-Instruct-2503', 'Qwen/Qwen2.5-VL-32B-Instruct', 'microsoft/Phi-4', 'NousResearch/Hermes-3-Llama-3.1-8B', 'internal/task']
+    
+    def __init__(
+        self,
+        is_conversation: bool = True,
+        max_tokens: int = 2000, # Note: max_tokens is not used by this API
+        timeout: int = 60,
+        filepath: str = None,
+        update_file: bool = True,
+        proxies: dict = {},
+        model: str = "Qwen/QwQ-32B",
+        cookie_path: str = "cookies.json",
+        assistantId: str = None, # Note: assistantId is not used by this API
+        system_prompt: str = "You are a helpful assistant. Please answer the following question.",
+    ):
+        """Initialize the HuggingFaceChat client."""
+        self.url = "https://huggingface.co/chat"
+        self.cookie_path = cookie_path
+        # Initialize curl_cffi Session
+        self.session = Session()
+        # Set up headers for all requests (remove those handled by impersonate)
+        self.headers = {
+            "Content-Type": "application/json", # Keep Content-Type for JSON posts
+            "Accept": "*/*", # Keep Accept
+            "Accept-Language": "en-US,en;q=0.9", # Keep Accept-Language
+            "Origin": "https://huggingface.co", # Keep Origin
+            "Referer": "https://huggingface.co/chat", # Keep Referer (will be updated)
+            "Sec-Fetch-Dest": "empty", # Keep Sec-Fetch-*
+            "Sec-Fetch-Mode": "cors",
+            "Sec-Fetch-Site": "same-origin",
+            "DNT": "1", # Keep DNT
+            "Priority": "u=1, i" # Keep Priority
+        }
+        
+        # Update curl_cffi session headers and proxies
+        self.session.headers.update(self.headers)
+        self.session.proxies = proxies # Assign proxies directly
+        self.system_prompt = system_prompt
+        self.assistantId = assistantId or None # Generate a new UUID if not provided
+        # Load cookies for authentication
+        self.cookies = self.load_cookies()
+        # Apply cookies to curl_cffi session
+        if self.cookies:
+            for name, value in self.cookies.items():
+                # Set cookies on the session object
+                self.session.cookies.set(name, value, domain="huggingface.co") # Specify domain if needed
+
+        # Update available models
+        self.update_available_models()
+        
+        # Set default model if none provided
+        self.model = model
+        
+        # Provider settings
+        self.is_conversation = is_conversation
+        self.max_tokens_to_sample = max_tokens
+        self.timeout = timeout
+        self.last_response = {}
+
+        # Initialize a simplified conversation history for file saving only
+        self.conversation = Conversation(is_conversation, max_tokens, filepath, update_file)
+        
+        # Store conversation data for different models
+        self._conversation_data = {}
+
+    def update_available_models(self):
+        """Update the available models list from HuggingFace"""
+        try:
+            models = self.get_models()
+            if models and len(models) > 0:
+                self.AVAILABLE_MODELS = models
+        except Exception:
+            # Fallback to default models list if fetching fails
+            pass
+    
+    @classmethod
+    def get_models(cls):
+        """Fetch available models from HuggingFace."""
+        try:
+            # Use a temporary curl_cffi session for this class method
+            temp_session = Session()
+            response = temp_session.get(
+                "https://huggingface.co/chat", 
+                impersonate="chrome110" # Use impersonate for fetching
+            )
+            response.raise_for_status()
+            text = response.text
+            models_match = re.search(r'models:(\[.+?\]),oldModels:', text)
+            
+            if not models_match:
+                return cls.AVAILABLE_MODELS
+                
+            models_text = models_match.group(1)
+            models_text = re.sub(r',parameters:{[^}]+?}', '', models_text)
+            models_text = models_text.replace('void 0', 'null')
+            
+            def add_quotation_mark(match):
+                return f'{match.group(1)}"{match.group(2)}":'
+                
+            models_text = re.sub(r'([{,])([A-Za-z0-9_]+?):', add_quotation_mark, models_text)
+            
+            models_data = json.loads(models_text)
+            # print([model["id"] for model in models_data])
+            return [model["id"] for model in models_data]
+        except (CurlError, Exception): # Catch CurlError and other exceptions
+            return cls.AVAILABLE_MODELS
+
+    def load_cookies(self):
+        """Load cookies from a JSON file"""
+        try:
+            with open(self.cookie_path, 'r') as f:
+                cookies_data = json.load(f)
+                
+            # Convert the cookie list to a dictionary format for requests
+            cookies = {}
+            for cookie in cookies_data:
+                # Only include cookies that are not expired and have a name and value
+                if 'name' in cookie and 'value' in cookie:
+                    # Check if the cookie hasn't expired
+                    if 'expirationDate' not in cookie or cookie['expirationDate'] > time.time():
+                        cookies[cookie['name']] = cookie['value']
+            
+            return cookies
+        except Exception:
+            return {}
+
+    def create_conversation(self, model: str):
+        """Create a new conversation with the specified model."""
+        url = "https://huggingface.co/chat/conversation"
+        payload = {"model": model, "assistantId": self.assistantId, "preprompt": self.system_prompt}
+        
+        # Update referer for this specific request
+        headers = self.headers.copy()
+        headers["Referer"] = f"https://huggingface.co/chat/models/{model}"
+        
+        try:
+            # Use curl_cffi session post with impersonate
+            response = self.session.post(
+                url, 
+                json=payload, 
+                headers=headers, # Use updated headers with specific Referer
+                impersonate="chrome110" # Use a common impersonation profile
+            )
+            
+            if response.status_code == 401:
+                raise exceptions.AuthenticationError("Authentication failed. Please check your cookies.")
+            
+            # Handle other error codes
+            if response.status_code != 200:
+                return None
+                
+            data = response.json()
+            conversation_id = data.get("conversationId")
+            
+            # Store conversation data
+            if model not in self._conversation_data:
+                self._conversation_data[model] = {
+                    "conversationId": conversation_id,
+                    "messageId": str(uuid.uuid4())  # Initial message ID
+                }
+            
+            return conversation_id
+        except CurlError as e: # Catch CurlError
+            # Log or handle CurlError specifically if needed
+            return None
+        except Exception: # Catch other potential exceptions (like JSONDecodeError, HTTPError)
+            return None
+    
+    def fetch_message_id(self, conversation_id: str) -> str:
+        """Fetch the latest message ID for a conversation."""
+        try:
+            url = f"https://huggingface.co/chat/conversation/{conversation_id}/__data.json?x-sveltekit-invalidated=11"
+            response = self.session.get(
+                url, 
+                headers=self.headers, # Use base headers
+                impersonate="chrome110" # Use a common impersonation profile
+            )
+            response.raise_for_status()
+            
+            # Parse the JSON data from the response
+            json_data = None
+            for line in response.text.split('\n'):
+                if line.strip():
+                    try:
+                        parsed = json.loads(line)
+                        if isinstance(parsed, dict) and "nodes" in parsed:
+                            json_data = parsed
+                            break
+                    except json.JSONDecodeError:
+                        continue
+            
+            if not json_data:
+                # Fall back to a UUID if we can't parse the response
+                return str(uuid.uuid4())
+            
+            # Extract message ID using the same pattern as in the example
+            if json_data.get("nodes", []) and json_data["nodes"][-1].get("type") == "error":
+                return str(uuid.uuid4())
+                
+            data = json_data["nodes"][1]["data"]
+            keys = data[data[0]["messages"]]
+            message_keys = data[keys[-1]]
+            message_id = data[message_keys["id"]]
+            
+            return message_id
+            
+        except CurlError: # Catch CurlError
+            return str(uuid.uuid4()) # Fallback on CurlError
+        except Exception: # Catch other potential exceptions
+            # Fall back to a UUID if there's an error
+            return str(uuid.uuid4())
+    
+    def generate_boundary(self):
+        """Generate a random boundary for multipart/form-data requests"""
+        boundary_chars = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789"
+        boundary = "----WebKitFormBoundary"
+        boundary += "".join(random.choice(boundary_chars) for _ in range(16))
+        return boundary
+    
+    def process_response(self, response, prompt: str):
+        """Process streaming response and extract content."""
+        full_text = ""
+        sources = None
+        reasoning_text = ""
+        has_reasoning = False
+        
+        for line in response.iter_lines(decode_unicode=True):
+            if not line:
+                continue
+                
+            try:
+                # Parse each line as JSON
+                data = json.loads(line)
+                
+                # Handle different response types
+                if "type" not in data:
+                    continue
+                    
+                if data["type"] == "stream" and "token" in data:
+                    token = data["token"].replace("\u0000", "")
+                    full_text += token
+                    resp = {"text": token}
+                    yield resp
+                elif data["type"] == "finalAnswer":
+                    final_text = data.get("text", "")
+                    if final_text and not full_text:
+                        full_text = final_text
+                        resp = {"text": final_text}
+                        yield resp
+                elif data["type"] == "webSearch" and "sources" in data:
+                    sources = data["sources"]
+                elif data["type"] == "reasoning":
+                    has_reasoning = True
+                    if data.get("subtype") == "stream" and "token" in data:
+                        reasoning_text += data["token"]
+                    # elif data.get("subtype") == "status":
+                    #     # For status updates in reasoning, we can just append them as a comment
+                    #     if data.get("status"):
+                    #         reasoning_text += f"\n# {data['status']}"
+                    
+                    # If we have reasoning, prepend it to the next text output
+                    if reasoning_text and not full_text:
+                        resp = {"text": f"<think>\n{reasoning_text}\n</think>\n", "is_reasoning": True}
+                        yield resp
+                    
+            except json.JSONDecodeError:
+                continue
+        
+        # Update conversation history only for saving to file if needed
+        if full_text and self.conversation.file:
+            if has_reasoning:
+                full_text_with_reasoning = f"<think>\n{reasoning_text}\n</think>\n{full_text}"
+                self.last_response = {"text": full_text_with_reasoning}
+                self.conversation.update_chat_history(prompt, full_text_with_reasoning)
+            else:
+                self.last_response = {"text": full_text}
+                self.conversation.update_chat_history(prompt, full_text)
+        
+        return full_text
+    
+    def ask(
+        self,
+        prompt: str,
+        stream: bool = False, # API supports streaming
+        raw: bool = False,
+        optimizer: str = None, # Note: optimizer is not used by this API
+        conversationally: bool = False, # Note: conversationally is not used by this API
+        web_search: bool = False,
+    ) -> Union[Dict[str, Any], Generator]:
+        """Send a message to the HuggingFace Chat API"""
+        model = self.model
+        
+        # Check if we have a conversation for this model
+        if model not in self._conversation_data:
+            conversation_id = self.create_conversation(model)
+            if not conversation_id:
+                raise exceptions.FailedToGenerateResponseError(f"Failed to create conversation with model {model}")
+        else:
+            conversation_id = self._conversation_data[model]["conversationId"]
+            # Refresh message ID
+            self._conversation_data[model]["messageId"] = self.fetch_message_id(conversation_id)
+        
+        url = f"https://huggingface.co/chat/conversation/{conversation_id}"
+        message_id = self._conversation_data[model]["messageId"]
+        
+        # Data to send - use the prompt directly without generating a complete prompt
+        # since HuggingFace maintains conversation state internally
+        request_data = {
+            "inputs": prompt,
+            "id": message_id,
+            "is_retry": False,
+            "is_continue": False,
+            "web_search": web_search,
+            "tools": ["66e85bb396d054c5771bc6cb", "00000000000000000000000a"]
+        }
+        
+        # Create multipart form data
+        boundary = self.generate_boundary()
+        multipart_headers = self.headers.copy()
+        multipart_headers["Content-Type"] = f"multipart/form-data; boundary={boundary}"
+        
+        # Serialize the data to JSON
+        data_json = json.dumps(request_data, separators=(',', ':'))
+        
+        # Create the multipart form data body
+        body = f"--{boundary}\r\n"
+        body += f'Content-Disposition: form-data; name="data"\r\n'
+        body += f"Content-Type: application/json\r\n\r\n"
+        body += f"{data_json}\r\n"
+        body += f"--{boundary}--\r\n"
+        
+        multipart_headers["Content-Length"] = str(len(body))
+        
+        def for_stream():
+            try:
+                # Try with multipart/form-data first
+                response = None
+                try:
+                    # Use curl_cffi session post with impersonate
+                    response = self.session.post(
+                        url, 
+                        data=body,
+                        headers=multipart_headers, # Use multipart headers
+                        stream=True,
+                        timeout=self.timeout,
+                        impersonate="chrome110" # Use a common impersonation profile
+                    )
+                    response.raise_for_status() # Check status after potential error
+                except (CurlError, exceptions.FailedToGenerateResponseError, Exception): # Catch potential errors
+                    response = None # Ensure response is None if multipart fails
+
+                # If multipart fails or returns error, try with regular JSON
+                if not response or response.status_code != 200:
+                    # Use curl_cffi session post with impersonate
+                    response = self.session.post(
+                        url, 
+                        json=request_data, # Use JSON payload
+                        headers=self.headers, # Use class-defined headers
+                        stream=True,
+                        timeout=self.timeout,
+                        impersonate="chrome110" # Use a common impersonation profile
+                    )
+                
+                response.raise_for_status() # Check status after potential fallback
+                
+                # Process the streaming response (iter_lines works with curl_cffi)
+                yield from self.process_response(response, prompt)
+                
+            except (CurlError, exceptions.FailedToGenerateResponseError, Exception) as e: # Catch errors from both attempts
+                if isinstance(e):
+                    if hasattr(e, 'response') and e.response is not None:
+                        status_code = e.response.status_code 
+                        if status_code == 401:
+                            raise exceptions.AuthenticationError("Authentication failed. Please check your cookies.")
+                
+                # Try another model if current one fails
+                if len(self.AVAILABLE_MODELS) > 1:
+                    current_model_index = self.AVAILABLE_MODELS.index(self.model) if self.model in self.AVAILABLE_MODELS else 0
+                    next_model_index = (current_model_index + 1) % len(self.AVAILABLE_MODELS)
+                    self.model = self.AVAILABLE_MODELS[next_model_index]
+                    
+                    # Create new conversation with the alternate model
+                    conversation_id = self.create_conversation(self.model)
+                    if conversation_id:
+                        # Try again with the new model
+                        yield from self.ask(prompt, stream=True, raw=raw, optimizer=optimizer, 
+                                          conversationally=conversationally, web_search=web_search)
+                        return
+                
+                # If we get here, all models failed
+                raise exceptions.FailedToGenerateResponseError(f"Request failed after trying fallback: {str(e)}") from e
+
+
+        def for_non_stream():
+            # Aggregate the stream using the updated for_stream logic
+            response_text = ""
+            try:
+                # Ensure raw=False so for_stream yields dicts
+                for chunk_data in for_stream():
+                    if isinstance(chunk_data, dict) and "text" in chunk_data:
+                        response_text += chunk_data["text"]
+                    # Handle raw string case if raw=True was passed
+                    elif raw and isinstance(chunk_data, str):
+                         response_text += chunk_data
+            except Exception as e:
+                 # If aggregation fails but some text was received, use it. Otherwise, re-raise.
+                 if not response_text:
+                     raise exceptions.FailedToGenerateResponseError(f"Failed to get non-stream response: {str(e)}") from e
+
+            # last_response and history are updated within process_response called by for_stream
+            # Return the final aggregated response dict or raw string
+            return response_text if raw else {"text": response_text} # Return dict for consistency
+
+
+        return for_stream() if stream else for_non_stream()
+
+    def chat(
+        self,
+        prompt: str,
+        stream: bool = False,
+        optimizer: str = None, # Note: optimizer is not used by this API
+        conversationally: bool = False, # Note: conversationally is not used by this API
+        web_search: bool = False
+    ) -> Union[str, Generator]:
+        """Generate a response to a prompt"""
+        def for_stream_chat():
+            # ask() yields dicts or strings when streaming
+            gen = self.ask(
+                prompt, stream=True, raw=False, # Ensure ask yields dicts
+                optimizer=optimizer, conversationally=conversationally, web_search=web_search
+            )
+            for response_dict in gen:
+                yield self.get_message(response_dict) # get_message expects dict
+                
+        def for_non_stream_chat():
+            # ask() returns dict or str when not streaming
+            response_data = self.ask(
+                prompt, stream=False, raw=False, # Ensure ask returns dict
+                optimizer=optimizer, conversationally=conversationally, web_search=web_search
+            )
+            return self.get_message(response_data) # get_message expects dict
+            
+        return for_stream_chat() if stream else for_non_stream_chat()
+
+    def get_message(self, response: dict) -> str:
+        """Extract message text from response"""
+        assert isinstance(response, dict), "Response should be of dict data-type only"
+        return response.get("text", "")