update docs

Author: binary-husky

ajet/tuner_lib/weight_tuner/experimental/as_oai_model_client.py

@@ -149,7 +149,11 @@ async def _service_loop(self):
                                 continue
 
                             try:
-                                parsed_msg = InterchangeCompletionRequest.model_validate_json(data)
+                                try:
+                                    parsed_msg = InterchangeCompletionRequest(**json.loads(data))
+                                except Exception as e:
+                                    logger.error(f"Failed to parse SSE event data: {e}" + data)
+                                    continue
 
                                 result = await self.llm_infer(
                                     req=parsed_msg.completion_request,

ajet/tuner_lib/weight_tuner/experimental/as_oai_model_server.py

@@ -250,7 +250,7 @@ async def chat_completions(request: Request, authorization: str = Header(None)):
     ajet_remote_handler_received[key][timeline_uuid] = int_req
 
     # Wait for response (with periodic checks for client disconnect)
-    max_wait_time = 1800  # 30 minutes timeout
+    max_wait_time = 600  # 10 minutes timeout
     elapsed_time = 0
 
     try:

ajet/utils/lowlevel_hook.py

@@ -0,0 +1,47 @@
+# def patch_task_creation():
+#     pass
+
+
+import inspect
+import asyncio
+from functools import wraps
+
+def patch_task_creation():
+    # Hook asyncio.create_task
+    original_create_task = asyncio.create_task
+
+    @wraps(original_create_task)
+    def debug_create_task(coro, name=None, **kwargs):
+        if not name:
+            caller = inspect.stack()[1]
+            coro_name = getattr(coro, '__name__', str(coro))
+            name = f"DEBUG_{coro_name}_at_{caller.filename.split('/')[-1]}:{caller.lineno}"
+            print(f"🎯 asyncio.create_task: {name}")
+        return original_create_task(coro, name=name, **kwargs)
+
+    # Hook loop.create_task
+    original_loop_create_task = asyncio.AbstractEventLoop.create_task
+
+    def debug_loop_create_task(self, coro, name=None, context=None):
+        if not name:
+            caller = inspect.stack()[1]
+            coro_name = getattr(coro, '__name__', str(coro))
+            name = f"DEBUG_{coro_name}_at_{caller.filename.split('/')[-1]}:{caller.lineno}"
+            print(f"🎯 loop.create_task: {name}")
+        return original_loop_create_task(self, coro, name=name, context=context)
+
+    # Hook Task.__init__
+    original_task_init = asyncio.Task.__init__
+
+    def debug_task_init(self, coro, loop=None, name=None, context=None):
+        if not name:
+            caller = inspect.stack()[2]  # 需要往上找更远的调用者
+            coro_name = getattr(coro, '__name__', str(coro))
+            name = f"DEBUG_{coro_name}_at_{caller.filename.split('/')[-1]}:{caller.lineno}"
+            print(f"🎯 Task.__init__: {name}")
+        return original_task_init(self, coro, loop=loop, name=name, context=context)
+
+    asyncio.create_task = debug_create_task
+    asyncio.AbstractEventLoop.create_task = debug_loop_create_task
+
+patch_task_creation()

docs/en/agent_framework_support.md

@@ -1,8 +1,9 @@
 # Agent Framework Support
 
+AgentJet currently supports (and has been tested with) the following agentic frameworks.
 
 <div class="card-grid">
-    <a href="en/tune_your_first_agent/" class="feature-card">
+    <a href="../support_agentscope" class="feature-card">
         <div class="card-header"><img src="https://api.iconify.design/lucide:rocket.svg"
                 class="card-icon card-icon-agent" alt="">
             <h3>AgentScope</h3>
@@ -11,7 +12,7 @@
             Agent-Oriented Programming for Building LLM Applications.
         </p>
     </a>
-    <a href="#example-library" class="feature-card">
+    <a href="../support_langchain" class="feature-card">
         <div class="card-header"><img src="https://api.iconify.design/lucide:library.svg"
                 class="card-icon card-icon-general" alt="">
             <h3>LangChain</h3>
@@ -20,16 +21,16 @@
             LangChain provides the engineering platform and open source frameworks developers use to build, test, and deploy reliable AI agents.
         </p>
     </a>
-    <a href="https://benchmark.agent-matrix.com/" class="feature-card">
-        <div class="card-header"><img src="https://api.iconify.design/lucide:shield-check.svg" class="card-icon card-icon-tool"
+    <a href="../support_oaisdk" class="feature-card">
+        <div class="card-header"><img src="https://api.iconify.design/lucide:shield.svg" class="card-icon card-icon-tool"
                 alt="">
             <h3>OpenAI SDK</h3>
         </div>
         <p class="card-desc">
           The OpenAI Agents SDK allows you to build agentic AI applications in a lightweight and easy-to-use package with minimal abstractions. By the way, both vLLM and SGLang offer compatible services.
         </p>
     </a>
-    <a href="https://benchmark.agent-matrix.com/" class="feature-card">
+    <a href="../support_http" class="feature-card">
         <div class="card-header"><img src="https://api.iconify.design/lucide:shield-check.svg" class="card-icon card-icon-tool"
                 alt="">
             <h3>Raw HTTP</h3>
@@ -39,4 +40,15 @@
           in this AI era, you can always start from scratch and build your own "high-scrapers".
         </p>
     </a>
+    <a href="../support_http" class="feature-card">
+        <div class="card-header"><img src="https://api.iconify.design/lucide:rocket.svg" class="card-icon card-icon-tool"
+                alt="">
+            <h3>More Agentic Frameworks</h3>
+        </div>
+        <p class="card-desc">
+          We are still testing all other agentic frameworks.
+          Meanwhile, "Raw HTTP" methods can provide `base_url` and `api_key` as AgentJet endpoint,
+          which theoretically can support most agentic frameworks.
+        </p>
+    </a>
 </div>

docs/en/support_agentscope.md

@@ -23,7 +23,7 @@ This article introduce the way to convert different types of ways to convert you
     )
     ```
 
-=== "After Convertion (`as_agentscope_model`)"
+=== "After Convertion (`as_agentscope_model()`)"
 
     ```python
     model = tuner.as_agentscope_model() # 🛩️ change here
@@ -35,7 +35,7 @@ This article introduce the way to convert different types of ways to convert you
     )
     ```
 
-=== "After Convertion (`as_agentscope_model`)"
+=== "After Convertion (`as_oai_baseurl_apikey()`)"
 
     ```python
     url_and_apikey = tuner.as_oai_baseurl_apikey()

docs/en/support_http.md

@@ -0,0 +1,97 @@
+# Without Any Agentic Framework
+
+Why use the Agent SDKs and all these abstractions? If you want to take control of the foundation of LLM Agents,
+in this AI era, you can always start from scratch and build your own "high-scrapers".
+
+## Http
+
+- use `tuner.as_oai_baseurl_apikey()` to obtain baseurl + apikey arguments
+
+### Explain with examples
+
+=== "Before Convertion"
+
+    ```python
+    # tuner to api key
+    base_url = "https://openrouter.ai/api/v1"
+    api_key = "sk-1234567"
+
+    # take out query
+    query = workflow_task.task.main_query
+
+    messages = [
+        {
+            "role": "system",
+            "content": self.system_prompt
+        },
+        {
+            "role": "user",
+            "content": query
+        }
+    ]
+
+    # use raw http requests (non-streaming) to get response
+    response = requests.post(
+            f"{base_url}/chat/completions",
+            json={
+                "model": "fill_whatever_model", # Of course, this `model` field will be ignored.
+                "messages": messages,
+            },
+            headers={
+                "Authorization": f"Bearer {api_key}"
+            }
+    )
+    final_answer = response.json()['choices'][0]['message']['content']
+    ```
+
+=== "After Convertion (`as_oai_baseurl_apikey`)"
+
+    ```python
+    # tuner to api key
+    url_and_apikey = tuner.as_oai_baseurl_apikey()
+    base_url = url_and_apikey.base_url
+    api_key = url_and_apikey.api_key
+
+    # take out query
+    query = workflow_task.task.main_query
+
+    messages = [
+        {
+            "role": "system",
+            "content": self.system_prompt
+        },
+        {
+            "role": "user",
+            "content": query
+        }
+    ]
+
+    # use raw http requests (non-streaming) to get response
+    response = requests.post(
+            f"{base_url}/chat/completions",
+            json={
+                "model": "fill_whatever_model", # Of course, this `model` field will be ignored.
+                "messages": messages,
+            },
+            headers={
+                "Authorization": f"Bearer {api_key}"
+            }
+    )
+    final_answer = response.json()['choices'][0]['message']['content']
+    ```
+
+
+
+!!! warning ""
+    - when you are using the `tuner.as_oai_baseurl_apikey()` api, you must enable the following feature in the yaml configuration.
+
+    ```yaml
+
+    ajet:
+        ...
+        enable_experimental_reverse_proxy: True
+        ...
+
+    ```
+
+

docs/en/support_langchain.md

@@ -0,0 +1,88 @@
+# Supported Agent Frameworks: AgentScope
+
+This article introduce the way to convert different types of ways to convert your existing workflows into AgentJet workflows.
+
+
+## AgentScope
+
+1. use `tuner.as_oai_baseurl_apikey()` to override OpenAIChatModel's baseurl + apikey argument
+
+### Explain with examples
+
+=== "Before Convertion"
+
+    ```python
+    from langchain_openai import ChatOpenAI
+
+
+
+
+    # create openai model
+    llm = ChatOpenAI(
+        model="gpt-5",
+    )
+    agent=create_agent(
+        model=llm,
+        system_prompt=self.system_prompt,
+    )
+
+    # take out query
+    query = workflow_task.task.main_query
+
+    response = agent.invoke({
+        "messages": [
+            {
+                "role": "user",
+                "content": query
+            }
+        ],
+    })
+    ```
+
+=== "After Convertion (`as_oai_baseurl_apikey`)"
+
+    ```python
+    from langchain_openai import ChatOpenAI
+
+    url_and_apikey = tuner.as_oai_baseurl_apikey()
+    base_url = url_and_apikey.base_url
+    api_key = url_and_apikey.api_key
+
+    llm = ChatOpenAI(
+        model="whatever",
+        base_url=base_url,
+        api_key=lambda:api_key,
+    )
+    agent = create_agent(
+        model=llm,
+        system_prompt=self.system_prompt,
+    )
+
+    # take out query
+    query = workflow_task.task.main_query
+
+    response = agent.invoke({
+        "messages": [
+            {
+                "role": "user",
+                "content": query
+            }
+        ],
+    })
+    ```
+
+
+
+!!! warning ""
+    - when you are using the `tuner.as_oai_baseurl_apikey()` api, you must enable the following feature in the yaml configuration.
+
+    ```yaml
+
+    ajet:
+        ...
+        enable_experimental_reverse_proxy: True
+        ...
+
+    ```
+
+