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

webscout/Provider/OPENAI/NEMOTRON.py

@@ -0,0 +1,232 @@
+import time
+import uuid
+import requests
+import json
+import random
+import datetime
+import re
+from typing import List, Dict, Optional, Union, Generator, Any
+from webscout.Provider.OPENAI.base import OpenAICompatibleProvider, BaseChat, BaseCompletions
+from webscout.Provider.OPENAI.utils import (
+    ChatCompletionChunk, ChatCompletion, Choice, ChoiceDelta,
+    ChatCompletionMessage, CompletionUsage, format_prompt, count_tokens
+)
+try:
+    from webscout.litagent import LitAgent
+except ImportError:
+    class LitAgent:
+        def random(self) -> str:
+            return "Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/127.0.0.0 Safari/537.36"
+from webscout.AIutel import sanitize_stream
+from webscout import exceptions
+
+
+class Completions(BaseCompletions):
+    def __init__(self, client: 'NEMOTRON'):
+        self._client = client
+
+    def create(
+        self,
+        *,
+        model: str,
+        messages: List[Dict[str, str]],
+        max_tokens: Optional[int] = None,
+        stream: bool = False,
+        temperature: Optional[float] = None,
+        top_p: Optional[float] = None,
+        **kwargs: Any
+    ) -> ChatCompletion:
+        nemotron_model_name = self._client.convert_model_name(model)
+        prompt_content = format_prompt(messages, add_special_tokens=True, include_system=True, do_continue=True)
+        payload = {
+            "content": prompt_content,
+            "imageSrc": "",
+            "model": nemotron_model_name,
+            "user": self._client._get_user_data(),
+            "conversationId": kwargs.get("conversation_id", "")
+        }
+        request_id = f"chatcmpl-{uuid.uuid4()}"
+        created_time = int(time.time())
+        # Always use non-stream mode, ignore 'stream' argument
+        return self._create_non_stream(request_id, created_time, model, payload)
+
+    def _create_stream(
+        self, request_id: str, created_time: int, model_name: str, payload: Dict[str, Any]
+    ) -> Generator[ChatCompletionChunk, None, None]:
+        try:
+            response_generator = self._client._internal_make_request(payload, stream=True)
+            for text_chunk in response_generator:
+                if text_chunk:
+                    delta = ChoiceDelta(content=text_chunk, role="assistant")
+                    choice = Choice(index=0, delta=delta, finish_reason=None)
+                    chunk = ChatCompletionChunk(
+                        id=request_id,
+                        choices=[choice],
+                        created=created_time,
+                        model=model_name,
+                    )
+                    yield chunk
+            final_delta = ChoiceDelta()
+            final_choice = Choice(index=0, delta=final_delta, finish_reason="stop")
+            final_chunk = ChatCompletionChunk(
+                id=request_id,
+                choices=[final_choice],
+                created=created_time,
+                model=model_name,
+            )
+            yield final_chunk
+        except Exception as e:
+            raise IOError(f"NEMOTRON request failed: {e}") from e
+
+    def _create_non_stream(
+        self, request_id: str, created_time: int, model_name: str, payload: Dict[str, Any]
+    ) -> ChatCompletion:
+        full_response_content = ""
+        try:
+            response_generator = self._client._internal_make_request(payload, stream=False)
+            full_response_content = next(response_generator, "")
+        except Exception as e:
+            pass
+        message = ChatCompletionMessage(role="assistant", content=full_response_content)
+        choice = Choice(index=0, message=message, finish_reason="stop")
+        prompt_tokens = count_tokens(payload.get("content", ""))
+        completion_tokens = count_tokens(full_response_content)
+        usage = CompletionUsage(
+            prompt_tokens=prompt_tokens,
+            completion_tokens=completion_tokens,
+            total_tokens=prompt_tokens + completion_tokens
+        )
+        completion = ChatCompletion(
+            id=request_id,
+            choices=[choice],
+            created=created_time,
+            model=model_name,
+            usage=usage,
+        )
+        return completion
+
+class Chat(BaseChat):
+    def __init__(self, client: 'NEMOTRON'):
+        self.completions = Completions(client)
+
+class NEMOTRON(OpenAICompatibleProvider):
+    AVAILABLE_MODELS = [
+        "gpt4o",
+        "nemotron70b",
+    ]
+    
+    API_BASE_URL = "https://nemotron.one/api/chat"
+    def __init__(
+        self,
+        timeout: int = 30,
+        proxies: dict = {}
+    ):
+        self.session = requests.Session()
+        self.timeout = timeout
+        agent = LitAgent()
+        user_agent = agent.random()
+        self.base_headers = {
+            "authority": "nemotron.one",
+            "accept": "*/*",
+            "accept-language": "en-US,en;q=0.9",
+            "content-type": "application/json",
+            "origin": "https://nemotron.one",
+            "sec-ch-ua": '"Chromium";v="136", "Not.A/Brand";v="99"',
+            "sec-ch-ua-mobile": "?0",
+            "sec-ch-ua-platform": '"Windows"',
+            "user-agent": user_agent
+        }
+        self.session.headers.update(self.base_headers)
+        if proxies:
+            self.session.proxies.update(proxies)
+        self.chat = Chat(self)
+
+    def _generate_random_email(self) -> str:
+        random_letter = "abcdefghijklmnopqrstuvwxyzABCDEFGHIJKLMNOPQRSTUVWXYZ0123456789"
+        random_string = ''.join(random.choice(random_letter) for _ in range(10))
+        return f"{random_string}@gmail.com"
+
+    def _generate_random_id(self) -> str:
+        timestamp = datetime.datetime.now().strftime("%Y%m%d%H%M%S%f")
+        random_letter = "abcdefghijklmnopqrstuvwxyz0123456789"
+        random_string = ''.join(random.choice(random_letter) for _ in range(8))
+        return f"cm{random_string}{timestamp[:10]}"
+
+    def _get_user_data(self) -> Dict[str, Any]:
+        current_time = datetime.datetime.now().isoformat()
+        return {
+            "name": "user",
+            "email": self._generate_random_email(),
+            "image": "https://lh3.googleusercontent.com/a/default-user=s96-c",
+            "id": self._generate_random_id(),
+            "password": None,
+            "emailVerified": None,
+            "credits": 100000000000,
+            "isPro": False,
+            "createdAt": current_time,
+            "updatedAt": current_time
+        }
+
+    def convert_model_name(self, model_alias: str) -> str:
+        """
+        Convert model names to ones supported by NEMOTRON API.
+        
+        Args:
+            model_alias: Model name to convert
+        
+        Returns:
+            NEMOTRON model name for API payload
+        """
+        # Accept only direct model names
+        if model_alias in self.AVAILABLE_MODELS:
+            return model_alias
+        
+        # Case-insensitive matching
+        for m in self.AVAILABLE_MODELS:
+            if m.lower() == model_alias.lower():
+                return m
+        
+        # Default to gpt4o if no match
+        print(f"Warning: Unknown model '{model_alias}'. Using 'gpt4o' instead.")
+        return "gpt4o"
+
+    def _internal_make_request(
+        self,
+        payload: Dict[str, Any],
+        stream: bool = False
+    ) -> Generator[str, None, None]:
+        request_headers = self.base_headers.copy()
+        request_headers["referer"] = f"https://nemotron.one/chat/{payload['model']}"
+        try:
+            if stream:
+                with self.session.post(
+                    self.API_BASE_URL,
+                    headers=request_headers,
+                    json=payload,
+                    stream=True,
+                    timeout=self.timeout
+                ) as response:
+                    response.raise_for_status()
+                    yield from sanitize_stream(
+                        response.iter_content(chunk_size=1024),
+                        to_json=False,
+                    )
+            else:
+                response = self.session.post(
+                    self.API_BASE_URL,
+                    headers=request_headers,
+                    json=payload,
+                    timeout=self.timeout
+                )
+                response.raise_for_status()
+                yield response.text
+        except requests.exceptions.RequestException as e:
+            raise exceptions.ProviderConnectionError(f"NEMOTRON API Connection error: {str(e)}")
+        except Exception as e:
+            raise RuntimeError(f"NEMOTRON API request unexpected error: {str(e)}")
+    @property
+    def models(self):
+        class _ModelList:
+            def list(inner_self):
+                return type(self).AVAILABLE_MODELS
+        return _ModelList()

webscout/Provider/OPENAI/Qwen3.py

@@ -0,0 +1,283 @@
+import requests
+import json
+import time
+import uuid
+from typing import List, Dict, Optional, Union, Generator, Any
+
+from webscout.Provider.OPENAI.base import OpenAICompatibleProvider, BaseChat, BaseCompletions
+from webscout.Provider.OPENAI.utils import (
+    ChatCompletionChunk, ChatCompletion, Choice, ChoiceDelta,
+    ChatCompletionMessage, CompletionUsage,
+    get_last_user_message, get_system_prompt,
+    count_tokens
+)
+
+class Completions(BaseCompletions):
+    def __init__(self, client: 'Qwen3'):
+        self._client = client
+
+    def create(
+        self,
+        *,
+        model: str,
+        messages: List[Dict[str, str]],
+        max_tokens: Optional[int] = 2048,
+        stream: bool = False,
+        temperature: Optional[float] = None,
+        top_p: Optional[float] = None,
+        **kwargs: Any
+    ) -> Union[ChatCompletion, Generator[ChatCompletionChunk, None, None]]:
+        payload = {
+            "data": [
+                get_last_user_message(messages),
+                {
+                    "thinking_budget": kwargs.get("thinking_budget", 38),
+                    "model": self._client.get_model(model),
+                    "sys_prompt": get_system_prompt(messages)
+                },
+                None, None
+            ],
+            "event_data": None,
+            "fn_index": 13,
+            "trigger_id": 31,
+            "session_hash": str(uuid.uuid4()).replace('-', '')
+        }
+
+        request_id = f"chatcmpl-{uuid.uuid4()}"
+        created_time = int(time.time())
+
+        if stream:
+            return self._create_stream(request_id, created_time, model, payload)
+        else:
+            return self._create_non_stream(request_id, created_time, model, payload)
+
+    def _create_stream(
+        self, request_id: str, created_time: int, model: str, payload: Dict[str, Any]
+    ) -> Generator[ChatCompletionChunk, None, None]:
+        session = self._client.session
+        headers = self._client.headers
+        # Step 1: Join the queue
+        join_resp = session.post(self._client.api_endpoint, headers=headers, json=payload, timeout=self._client.timeout)
+        join_resp.raise_for_status()
+        event_id = join_resp.json().get('event_id')
+        session_hash = payload["session_hash"]
+
+        # Step 2: Stream data
+        params = {'session_hash': session_hash}
+        stream_resp = session.get(self._client.url + "/gradio_api/queue/data", headers=self._client.stream_headers, params=params, stream=True, timeout=self._client.timeout)
+        stream_resp.raise_for_status()
+
+        # --- New logic to yield all content, tool reasoning, and status, similar to Reasoning class ---
+        is_thinking_tag_open = False # True if <think> has been yielded and not yet </think>
+
+        for line in stream_resp.iter_lines():
+            if line:
+                decoded_line = line.decode('utf-8')
+                if decoded_line.startswith('data: '):
+                    try:
+                        json_data = json.loads(decoded_line[6:])
+                        if json_data.get('msg') == 'process_generating':
+                            if 'output' in json_data and 'data' in json_data['output'] and len(json_data['output']['data']) > 5:
+                                updates_list = json_data['output']['data'][5] # This is a list of operations
+                                for op_details in updates_list:
+                                    action = op_details[0]
+                                    path = op_details[1]
+                                    value = op_details[2]
+
+                                    content_to_yield = None
+                                    is_current_op_tool = False
+                                    is_current_op_text = False
+
+                                    # Case 1: Adding a new content block (tool or text object)
+                                    if action == "add" and isinstance(value, dict) and "type" in value:
+                                        if len(path) == 4 and path[0] == "value" and path[2] == "content":
+                                            block_type = value.get("type")
+                                            content_to_yield = value.get("content")
+                                            if block_type == "tool":
+                                                is_current_op_tool = True
+                                            elif block_type == "text":
+                                                is_current_op_text = True
+                                    
+                                    # Case 2: Appending content string to an existing block
+                                    elif action == "append" and isinstance(value, str):
+                                        if len(path) == 5 and path[0] == "value" and path[2] == "content" and path[4] == "content":
+                                            block_index = path[3] # 0 for tool's content, 1 for text's content
+                                            content_to_yield = value
+                                            if block_index == 0: # Appending to tool's content
+                                                is_current_op_tool = True
+                                            elif block_index == 1: # Appending to text's content
+                                                is_current_op_text = True
+                                    
+                                    # Case 3: Tool status update (e.g., "End of Thought")
+                                    elif action == "replace" and len(path) == 6 and \
+                                         path[0] == "value" and path[2] == "content" and \
+                                         path[3] == 0 and path[4] == "options" and path[5] == "status": # path[3]==0 ensures it's the tool block
+                                        if value == "done": # Tool block processing is complete
+                                            if is_thinking_tag_open:
+                                                delta = ChoiceDelta(content="</think>\n\n", role="assistant")
+                                                yield ChatCompletionChunk(id=request_id, choices=[Choice(index=0, delta=delta)], created=created_time, model=model)
+                                                is_thinking_tag_open = False
+                                        continue # This operation itself doesn't yield visible content
+
+                                    # Yielding logic
+                                    if is_current_op_tool and content_to_yield:
+                                        if not is_thinking_tag_open:
+                                            delta = ChoiceDelta(content="<think>", role="assistant")
+                                            yield ChatCompletionChunk(id=request_id, choices=[Choice(index=0, delta=delta)], created=created_time, model=model)
+                                            is_thinking_tag_open = True
+                                        
+                                        delta = ChoiceDelta(content=content_to_yield, role="assistant")
+                                        yield ChatCompletionChunk(id=request_id, choices=[Choice(index=0, delta=delta)], created=created_time, model=model)
+
+                                    elif is_current_op_text and content_to_yield:
+                                        if is_thinking_tag_open: # If text starts, close any open thinking tag
+                                            delta = ChoiceDelta(content="</think>", role="assistant")
+                                            yield ChatCompletionChunk(id=request_id, choices=[Choice(index=0, delta=delta)], created=created_time, model=model)
+                                            is_thinking_tag_open = False
+                                        
+                                        delta = ChoiceDelta(content=content_to_yield, role="assistant")
+                                        yield ChatCompletionChunk(id=request_id, choices=[Choice(index=0, delta=delta)], created=created_time, model=model)
+
+                        if json_data.get('msg') == 'process_completed':
+                            if is_thinking_tag_open: # Ensure </think> is yielded if process completes mid-thought
+                                delta = ChoiceDelta(content="</think>", role="assistant")
+                                yield ChatCompletionChunk(id=request_id, choices=[Choice(index=0, delta=delta)], created=created_time, model=model)
+                                is_thinking_tag_open = False
+                            break
+                    except json.JSONDecodeError:
+                        continue
+                    except Exception as e:
+                        # Log or handle other potential exceptions
+                        continue
+        
+        # After the loop, ensure the tag is closed if the stream broke for reasons other than 'process_completed'
+        if is_thinking_tag_open:
+            delta = ChoiceDelta(content="</think>", role="assistant")
+            yield ChatCompletionChunk(id=request_id, choices=[Choice(index=0, delta=delta)], created=created_time, model=model)
+
+    def _create_non_stream(
+        self, request_id: str, created_time: int, model: str, payload: Dict[str, Any]
+    ) -> ChatCompletion:
+        # For non-streaming, just call the join endpoint and parse the result
+        session = self._client.session
+        headers = self._client.headers
+        resp = session.post(self._client.api_endpoint, headers=headers, json=payload, timeout=self._client.timeout)
+        resp.raise_for_status()
+        data = resp.json()
+        # Return the full content as a single message, including all tool and text reasoning if present
+        output = ""
+        if 'output' in data and 'data' in data['output'] and len(data['output']['data']) > 5:
+            updates = data['output']['data'][5]
+            parts = []
+            for update in updates:
+                if isinstance(update, list) and len(update) > 2 and isinstance(update[2], str):
+                    parts.append(update[2])
+                elif isinstance(update, list) and isinstance(update[1], list) and len(update[1]) > 4:
+                    if update[1][4] == "content":
+                        parts.append(update[2])
+                    elif update[1][4] == "options" and update[2] != "done":
+                        parts.append(str(update[2]))
+                elif isinstance(update, dict):
+                    if update.get('type') == 'tool':
+                        parts.append(update.get('content', ''))
+                    elif update.get('type') == 'text':
+                        parts.append(update.get('content', ''))
+            output = "\n".join([str(p) for p in parts if p])
+        else:
+            output = data.get('output', {}).get('data', ["", "", "", "", "", [["", "", ""]]])[5][0][2]
+        message = ChatCompletionMessage(role="assistant", content=output)
+        choice = Choice(index=0, message=message, finish_reason="stop")
+        # Use count_tokens to compute usage
+        prompt_tokens = count_tokens([m.get('content', '') for m in payload['data'] if isinstance(m, dict) and 'content' in m or isinstance(m, str)])
+        completion_tokens = count_tokens(output)
+        usage = CompletionUsage(
+            prompt_tokens=prompt_tokens,
+            completion_tokens=completion_tokens,
+            total_tokens=prompt_tokens + completion_tokens
+        )
+        completion = ChatCompletion(
+            id=request_id,
+            choices=[choice],
+            created=created_time,
+            model=model,
+            usage=usage,
+        )
+        return completion
+
+class Chat(BaseChat):
+    def __init__(self, client: 'Qwen3'):
+        self.completions = Completions(client)
+
+class Qwen3(OpenAICompatibleProvider):
+    url = "https://qwen-qwen3-demo.hf.space"
+    api_endpoint = "https://qwen-qwen3-demo.hf.space/gradio_api/queue/join?__theme=system"
+    AVAILABLE_MODELS = [
+        "qwen3-235b-a22b",
+        "qwen3-32b",
+        "qwen3-30b-a3b",
+        "qwen3-14b",
+        "qwen3-8b",
+        "qwen3-4b",
+        "qwen3-1.7b",
+        "qwen3-0.6b",
+    ]
+    MODEL_ALIASES = {
+        "qwen-3-235b": "qwen3-235b-a22b",
+        "qwen-3-30b": "qwen3-30b-a3b",
+        "qwen-3-32b": "qwen3-32b",
+        "qwen-3-14b": "qwen3-14b",
+        "qwen-3-4b": "qwen3-4b",
+        "qwen-3-1.7b": "qwen3-1.7b",
+        "qwen-3-0.6b": "qwen3-0.6b"
+    }
+
+    def __init__(self, timeout: Optional[int] = None):
+        self.timeout = timeout
+        self.session = requests.Session()
+        self.headers = {
+            'User-Agent': 'Mozilla/5.0 (X11; Linux x86_64; rv:136.0) Gecko/20100101 Firefox/136.0',
+            'Accept': '*/*',
+            'Accept-Language': 'en-US,en;q=0.5',
+            'Accept-Encoding': 'gzip, deflate, br, zstd',
+            'Referer': f'{self.url}/?__theme=system',
+            'content-type': 'application/json',
+            'Origin': self.url,
+            'Connection': 'keep-alive',
+            'Sec-Fetch-Dest': 'empty',
+            'Sec-Fetch-Mode': 'cors',
+            'Sec-Fetch-Site': 'same-origin',
+            'Pragma': 'no-cache',
+            'Cache-Control': 'no-cache',
+        }
+        self.stream_headers = {
+            'Accept': 'text/event-stream',
+            'Accept-Language': 'en-US,en;q=0.5',
+            'Referer': f'{self.url}/?__theme=system',
+            'User-Agent': self.headers['User-Agent'],
+        }
+        self.session.headers.update(self.headers)
+        self.chat = Chat(self)
+
+    def get_model(self, model):
+        return self.MODEL_ALIASES.get(model, model)
+
+    @property
+    def models(self):
+        class _ModelList:
+            def list(inner_self):
+                return type(self).AVAILABLE_MODELS
+        return _ModelList()
+
+if __name__ == "__main__":
+    client = Qwen3()
+    from rich import print
+    resp = client.chat.completions.create(
+        model="qwen3-14b",
+        messages=[
+            {"role": "system", "content": "You are a helpful assistant."},
+            {"role": "user", "content": "Hello "}
+        ],
+        stream=True
+    )
+    for chunk in resp:
+        print(chunk, end="", flush=True)