Merge pull request xtekky#1240 from hlohaus/pi

Add auto_continue to OpenaiChat
Lin-jun-xiang · Nov 12, 2023 · df8f416 · df8f416
2 parents 9d222e8 + 745ea57
commit df8f416
Show file tree

Hide file tree

Showing 7 changed files with 192 additions and 186 deletions.
diff --git a/README.md b/README.md
diff --git a/etc/testing/_providers.py b/etc/testing/_providers.py
@@ -35,8 +35,14 @@ def main():
 def get_providers() -> list[type[BaseProvider]]:
     providers = dir(Provider)
     providers = [getattr(Provider, provider) for provider in providers if provider != "RetryProvider"]
-    providers = [provider for provider in providers if isinstance(provider, type)]
-    return [provider for provider in providers if issubclass(provider, BaseProvider)]
+    providers = [provider for provider in providers if isinstance(provider, type) and hasattr(provider, "url")]
+    return [
+        provider
+        for provider in providers
+        if issubclass(provider, BaseProvider)
+        and provider.__name__ not in dir(Provider.deprecated)
+        and provider.__name__ not in dir(Provider.unfinished)
+    ]
 
 
 def create_response(_provider: type[BaseProvider]) -> str:

diff --git a/etc/tool/readme_table.py b/etc/tool/readme_table.py
@@ -3,58 +3,32 @@
 from pathlib import Path
 from urllib.parse import urlparse
 
-sys.path.append(str(Path(__file__).parent.parent))
+sys.path.append(str(Path(__file__).parent.parent.parent))
 
 import asyncio
 from g4f import models
-from g4f.Provider.base_provider import AsyncProvider, BaseProvider
-from g4f.Provider.retry_provider import RetryProvider
-from testing._providers import get_providers
-
-logging = False
-
-
-def print_imports():
-    print("##### Providers:")
-    print("```py")
-    print("from g4f.Provider import (")
-    for _provider in get_providers():
-        if _provider.working:
-            print(f"    {_provider.__name__},")
-
-    print(")")
-    print("# Usage:")
-    print("response = g4f.ChatCompletion.create(..., provider=ProviderName)")
-    print("```")
-    print()
-    print()
-
-def print_async():
-    print("##### Async support:")
-    print("```py")
-    print("_providers = [")
-    for _provider in get_providers():
-        if _provider.working and issubclass(_provider, AsyncProvider):
-            print(f"      g4f.Provider.{_provider.__name__},")
-    print("]")
-    print("```")
-    print()
-    print()
+from g4f import ChatCompletion
+from g4f.Provider.base_provider import BaseProvider
+from etc.testing._providers import get_providers
+
+from g4f import debug
+
+debug.logging = True
 
 
 async def test_async(provider: type[BaseProvider]):
     if not provider.working:
         return False
-    model = models.gpt_35_turbo.name if provider.supports_gpt_35_turbo else models.default.name
     messages = [{"role": "user", "content": "Hello Assistant!"}]
     try:
-        if issubclass(provider, AsyncProvider):
-            response = await provider.create_async(model=model, messages=messages)
-        else:
-            response = provider.create_completion(model=model, messages=messages, stream=False)
+        response = await asyncio.wait_for(ChatCompletion.create_async(
+            model=models.default,
+            messages=messages,
+            provider=provider
+        ), 30)
         return bool(response)
     except Exception as e:
-        if logging:
+        if debug.logging:
             print(f"{provider.__name__}: {e.__class__.__name__}: {e}")
         return False
 
@@ -68,44 +42,53 @@ async def test_async_list(providers: list[type[BaseProvider]]):
 
 
 def print_providers():
-    lines = [
-        "| Website| Provider| gpt-3.5 | gpt-4 | Streaming | Asynchron | Status | Auth |",
-        "| ------ | ------- | ------- | ----- | --------- | --------- | ------ | ---- |",
-    ]
 
     providers = get_providers()
     responses = asyncio.run(test_async_list(providers))
 
-    for is_working in (True, False):
-        for idx, _provider in enumerate(providers):
-            if is_working != _provider.working:
-                continue
-            if _provider == RetryProvider:
-                continue
-
-            netloc = urlparse(_provider.url).netloc
-            website = f"[{netloc}]({_provider.url})"
-
-            provider_name = f"`g4f.Provider.{_provider.__name__}`"
-
-            has_gpt_35 = "✔️" if _provider.supports_gpt_35_turbo else "❌"
-            has_gpt_4 = "✔️" if _provider.supports_gpt_4 else "❌"
-            stream = "✔️" if _provider.supports_stream else "❌"
-            can_async = "✔️" if issubclass(_provider, AsyncProvider) else "❌"
-            if _provider.working:
-                status = '![Active](https://img.shields.io/badge/Active-brightgreen)'
-                if responses[idx]:
+    for type in ("GPT-4", "GPT-3.5", "Other"):
+        lines = [
+            "",
+            f"### {type}",
+            "",
+            "| Website | Provider | GPT-3.5 | GPT-4 | Stream | Status | Auth |",
+            "| ------  | -------  | ------- | ----- | ------ | ------ | ---- |",
+        ]
+        for is_working in (True, False):
+            for idx, _provider in enumerate(providers):
+                if is_working != _provider.working:
+                    continue
+                do_continue = False
+                if type == "GPT-4" and _provider.supports_gpt_4:
+                    do_continue = True
+                elif type == "GPT-3.5" and not _provider.supports_gpt_4 and _provider.supports_gpt_35_turbo:
+                    do_continue = True
+                elif type == "Other" and not _provider.supports_gpt_4 and not _provider.supports_gpt_35_turbo:
+                    do_continue = True
+                if not do_continue:
+                    continue
+                netloc = urlparse(_provider.url).netloc
+                website = f"[{netloc}]({_provider.url})"
+
+                provider_name = f"`g4f.Provider.{_provider.__name__}`"
+
+                has_gpt_35 = "✔️" if _provider.supports_gpt_35_turbo else "❌"
+                has_gpt_4 = "✔️" if _provider.supports_gpt_4 else "❌"
+                stream = "✔️" if _provider.supports_stream else "❌"
+                if _provider.working:
                     status = '![Active](https://img.shields.io/badge/Active-brightgreen)'
+                    if responses[idx]:
+                        status = '![Active](https://img.shields.io/badge/Active-brightgreen)'
+                    else:
+                        status = '![Unknown](https://img.shields.io/badge/Unknown-grey)'
                 else:
-                    status = '![Unknown](https://img.shields.io/badge/Unknown-grey)'
-            else:
-                status = '![Inactive](https://img.shields.io/badge/Inactive-red)'
-            auth = "✔️" if _provider.needs_auth else "❌"
-
-            lines.append(
-                f"| {website} | {provider_name} | {has_gpt_35} | {has_gpt_4} | {stream} | {can_async} | {status} | {auth} |"
-            )
-    print("\n".join(lines))
+                    status = '![Inactive](https://img.shields.io/badge/Inactive-red)'
+                auth = "✔️" if _provider.needs_auth else "❌"
+
+                lines.append(
+                    f"| {website} | {provider_name} | {has_gpt_35} | {has_gpt_4} | {stream} | {status} | {auth} |"
+                )
+        print("\n".join(lines))
 
 def print_models():
     base_provider_names = {
@@ -151,8 +134,6 @@ def get_models():
 
 
 if __name__ == "__main__":
-    print_imports()
-    print_async()
     print_providers()
     print("\n", "-" * 50, "\n")
     print_models()
diff --git a/g4f/Provider/MyShell.py b/g4f/Provider/MyShell.py
@@ -10,6 +10,7 @@
 class MyShell(AsyncGeneratorProvider):
     url = "https://app.myshell.ai/chat"
     working = True
+    supports_gpt_35_turbo = True
 
     @classmethod
     async def create_async_generator(

diff --git a/g4f/Provider/needs_auth/OpenaiChat.py b/g4f/Provider/needs_auth/OpenaiChat.py
@@ -1,6 +1,6 @@
 from __future__ import annotations
 
-import uuid, json, time, os
+import uuid, json, time, asyncio
 from py_arkose_generator.arkose import get_values_for_request
 
 from ..base_provider import AsyncGeneratorProvider
@@ -24,6 +24,7 @@ async def create_async_generator(
         proxy: str = None,
         timeout: int = 120,
         access_token: str = None,
+        auto_continue: bool = False,
         cookies: dict = None,
         **kwargs
     ) -> AsyncResult:
@@ -34,50 +35,73 @@ async def create_async_generator(
             "Accept": "text/event-stream",
             "Authorization": f"Bearer {access_token}",
         }
-        async with StreamSession(
-            proxies=proxies,
-            headers=headers,
-            impersonate="chrome107",
-            timeout=timeout
-        ) as session:
-            messages = [
-                {
-                    "id": str(uuid.uuid4()),
-                    "author": {"role": "user"},
-                    "content": {"content_type": "text", "parts": [format_prompt(messages)]},
-                },
-            ]
-            data = {
-                "action": "next",
-                "arkose_token": await get_arkose_token(proxy),
-                "messages": messages,
-                "conversation_id": None,
-                "parent_message_id": str(uuid.uuid4()),
-                "model": "text-davinci-002-render-sha",
-                "history_and_training_disabled": True,
-            }
-            async with session.post(f"{cls.url}/backend-api/conversation", json=data) as response:
-                response.raise_for_status()
-                last_message = ""
-                async for line in response.iter_lines():
-                    if line.startswith(b"data: "):
-                        line = line[6:]
-                        if line == b"[DONE]":
-                            break
-                        try:
-                            line = json.loads(line)
-                        except:
-                            continue
-                        if "message" not in line:
-                            continue
-                        if "error" in line and line["error"]:
-                            raise RuntimeError(line["error"])
-                        if "message_type" not in line["message"]["metadata"]:
-                            continue
-                        if line["message"]["metadata"]["message_type"] == "next":
-                            new_message = line["message"]["content"]["parts"][0]
-                            yield new_message[len(last_message):]
-                            last_message = new_message
+        messages = [
+            {
+                "id": str(uuid.uuid4()),
+                "author": {"role": "user"},
+                "content": {"content_type": "text", "parts": [format_prompt(messages)]},
+            },
+        ]
+        message_id = str(uuid.uuid4())
+        data = {
+            "action": "next",
+            "arkose_token": await get_arkose_token(proxy),
+            "messages": messages,
+            "conversation_id": None,
+            "parent_message_id": message_id,
+            "model": "text-davinci-002-render-sha",
+            "history_and_training_disabled": not auto_continue,
+        }
+        conversation_id = None
+        while not end_turn:
+            if not auto_continue:
+                end_turn = True
+            async with StreamSession(
+                proxies=proxies,
+                headers=headers,
+                impersonate="chrome107",
+                timeout=timeout
+            ) as session:
+                async with session.post(f"{cls.url}/backend-api/conversation", json=data) as response:
+                    try:
+                        response.raise_for_status()
+                    except:
+                        raise RuntimeError(f"Response: {await response.text()}")
+                    last_message = ""
+                    async for line in response.iter_lines():
+                        if line.startswith(b"data: "):
+                            line = line[6:]
+                            if line == b"[DONE]":
+                                break
+                            try:
+                                line = json.loads(line)
+                            except:
+                                continue
+                            if "message" not in line:
+                                continue
+                            if "error" in line and line["error"]:
+                                raise RuntimeError(line["error"])
+                            end_turn = line["message"]["end_turn"]
+                            message_id = line["message"]["id"]
+                            if line["conversation_id"]:
+                                conversation_id = line["conversation_id"]
+                            if "message_type" not in line["message"]["metadata"]:
+                                continue
+                            if line["message"]["metadata"]["message_type"] in ("next", "continue"):
+                                new_message = line["message"]["content"]["parts"][0]
+                                yield new_message[len(last_message):]
+                                last_message = new_message
+                            if end_turn:
+                                return
+                data = {
+                    "action": "continue",
+                    "arkose_token": await get_arkose_token(proxy),
+                    "conversation_id": conversation_id,
+                    "parent_message_id": message_id,
+                    "model": "text-davinci-002-render-sha",
+                    "history_and_training_disabled": False,
+                }
+                await asyncio.sleep(5)
 
     @classmethod
     async def browse_access_token(cls) -> str:

diff --git a/g4f/models.py b/g4f/models.py
@@ -82,7 +82,7 @@ def __all__() -> list[str]:
     best_provider = RetryProvider([Llama2, DeepInfra]))
 
 llama2_13b = Model(
-    name          ="meta-llama/Llama-2-13b-chat-hf",
+    name          = "meta-llama/Llama-2-13b-chat-hf",
     base_provider = 'huggingface',
     best_provider = RetryProvider([Llama2, DeepInfra]))
 

diff --git a/requirements.txt b/requirements.txt
@@ -21,4 +21,4 @@ asgiref
 fastapi
 uvicorn
 flask
-git+https://github.com/hlohaus/py-arkose-token-generator.git
+py-arkose-generator