beswarm 0.2.73__tar.gz → 0.2.76__tar.gz

{beswarm-0.2.73 → beswarm-0.2.76}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: beswarm
-Version: 0.2.73
+Version: 0.2.76
 Summary: MAS
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown

{beswarm-0.2.73 → beswarm-0.2.76}/beswarm/agents/planact.py

@@ -12,10 +12,10 @@ from typing import List, Dict, Union
 
 from ..broker import MessageBroker
 from ..aient.aient.models import chatgpt
-from ..aient.aient.models.chatgpt import ModelNotFoundError, TaskComplete, RetryFailedError
 from ..aient.aient.plugins import get_function_call_list, registry
 from ..prompt import worker_system_prompt, instruction_system_prompt
 from ..utils import extract_xml_content, get_current_screen_image_message, replace_xml_content, register_mcp_tools, setup_logger
+from ..aient.aient.models.chatgpt import ModelNotFoundError, TaskComplete, RetryFailedError, InputTokenCountExceededError, BadRequestError
 
 try:
     from importlib import metadata
@@ -154,11 +154,12 @@ class InstructionAgent(BaseAgent):
             self.logger.error("❌ Commander retry failed, retrying...")
             self.broker.publish(message, self.error_topic)
             return
-
-        if "'status_code': 413" in raw_response or \
-        "'status_code': 400" in raw_response:
+        except InputTokenCountExceededError as e:
             self.broker.publish({"status": "error", "result": "The request body is too long, please try again."}, self.status_topic)
             return
+        except BadRequestError as e:
+            self.broker.publish({"status": "error", "result": "Bad request error!"}, self.status_topic)
+            return
 
         self.broker.publish({"status": "new_message", "result": "\n🤖 指令智能体:\n" + raw_response}, self.status_topic)
 
@@ -225,6 +226,9 @@ class WorkerAgent(BaseAgent):
         except TaskComplete as e:
             self.broker.publish({"status": "finished", "result": e.completion_message}, self.status_topic)
             return
+        except InputTokenCountExceededError as e:
+            self.broker.publish({"status": "error", "result": "The request body is too long, please try again."}, self.status_topic)
+            return
         except RetryFailedError as e:
             self.logger.error("❌ Worker retry failed, retrying...")
             self.broker.publish(message, self.error_topic)

{beswarm-0.2.73 → beswarm-0.2.76}/beswarm/aient/aient/core/response.py

@@ -21,13 +21,12 @@ async def check_response(response, error_log):
         return {"error": f"{error_log} HTTP Error", "status_code": response.status_code, "details": error_json}
     return None
 
-async def gemini_json_poccess(response_str):
+async def gemini_json_poccess(response_json):
     promptTokenCount = 0
     candidatesTokenCount = 0
     totalTokenCount = 0
     image_base64 = None
 
-    response_json = await asyncio.to_thread(json.loads, response_str)
     json_data = safe_get(response_json, "candidates", 0, "content", default=None)
     finishReason = safe_get(response_json, "candidates", 0 , "finishReason", default=None)
     if finishReason:
@@ -48,7 +47,7 @@ async def gemini_json_poccess(response_str):
 
     function_call_name = safe_get(json_data, "functionCall", "name", default=None)
     function_full_response = safe_get(json_data, "functionCall", "args", default="")
-    function_full_response = json.dumps(function_full_response) if function_full_response else None
+    function_full_response = await asyncio.to_thread(json.dumps, function_full_response) if function_full_response else None
 
     blockReason = safe_get(json_data, 0, "promptFeedback", "blockReason", default=None)
 
@@ -77,7 +76,7 @@ async def fetch_gemini_response_stream(client, url, headers, payload, model, tim
                 if line.startswith("data: "):
                     parts_json = line.lstrip("data: ").strip()
                     try:
-                        await asyncio.to_thread(json.loads, parts_json)
+                        response_json = await asyncio.to_thread(json.loads, parts_json)
                     except json.JSONDecodeError:
                         logger.error(f"JSON decode error: {parts_json}")
                         continue
@@ -85,12 +84,12 @@ async def fetch_gemini_response_stream(client, url, headers, payload, model, tim
                     parts_json += line
                     parts_json = parts_json.lstrip("[,")
                     try:
-                        await asyncio.to_thread(json.loads, parts_json)
+                        response_json = await asyncio.to_thread(json.loads, parts_json)
                     except json.JSONDecodeError:
                         continue
 
                 # https://ai.google.dev/api/generate-content?hl=zh-cn#FinishReason
-                is_thinking, reasoning_content, content, image_base64, function_call_name, function_full_response, finishReason, blockReason, promptTokenCount, candidatesTokenCount, totalTokenCount = await gemini_json_poccess(parts_json)
+                is_thinking, reasoning_content, content, image_base64, function_call_name, function_full_response, finishReason, blockReason, promptTokenCount, candidatesTokenCount, totalTokenCount = await gemini_json_poccess(response_json)
 
                 if is_thinking:
                     sse_string = await generate_sse_response(timestamp, model, reasoning_content=reasoning_content)
@@ -184,7 +183,7 @@ async def fetch_vertex_claude_response_stream(client, url, headers, payload, mod
             function_call_id = function_call["id"]
             sse_string = await generate_sse_response(timestamp, model, content=None, tools_id=function_call_id, function_call_name=function_call_name)
             yield sse_string
-            function_full_response = json.dumps(function_call["input"])
+            function_full_response = await asyncio.to_thread(json.dumps, function_call["input"])
             sse_string = await generate_sse_response(timestamp, model, content=None, tools_id=function_call_id, function_call_name=None, function_call_content=function_full_response)
             yield sse_string
 
@@ -214,6 +213,9 @@ async def fetch_gpt_response_stream(client, url, headers, payload, timeout):
             while "\n" in buffer:
                 line, buffer = buffer.split("\n", 1)
                 # logger.info("line: %s", repr(line))
+                if line.startswith(": keepalive"):
+                    yield line + end_of_line
+                    continue
                 if line and not line.startswith(":") and (result:=line.lstrip("data: ").strip()):
                     if result.strip() == "[DONE]":
                         break
@@ -307,7 +309,8 @@ async def fetch_gpt_response_stream(client, url, headers, payload, timeout):
                     else:
                         if no_stream_content:
                             del line["choices"][0]["message"]
-                        yield "data: " + json.dumps(line).strip() + end_of_line
+                        json_line = await asyncio.to_thread(json.dumps, line)
+                        yield "data: " + json_line.strip() + end_of_line
     yield "data: [DONE]" + end_of_line
 
 async def fetch_azure_response_stream(client, url, headers, payload, timeout):
@@ -364,7 +367,8 @@ async def fetch_azure_response_stream(client, url, headers, payload, timeout):
                     else:
                         if no_stream_content:
                             del line["choices"][0]["message"]
-                        yield "data: " + json.dumps(line).strip() + end_of_line
+                        json_line = await asyncio.to_thread(json.dumps, line)
+                        yield "data: " + json_line.strip() + end_of_line
     yield "data: [DONE]" + end_of_line
 
 async def fetch_cloudflare_response_stream(client, url, headers, payload, model, timeout):

{beswarm-0.2.73 → beswarm-0.2.76}/beswarm/aient/aient/core/utils.py

@@ -243,7 +243,8 @@ def parse_rate_limit(limit_string):
         'h': 3600, 'hr': 3600, 'hour': 3600,
         'd': 86400, 'day': 86400,
         'mo': 2592000, 'month': 2592000,
-        'y': 31536000, 'year': 31536000
+        'y': 31536000, 'year': 31536000,
+        'tpr': -1,
     }
 
     # 处理多个限制条件
@@ -377,6 +378,32 @@ class ThreadSafeCircularList:
                     logger.warning(f"All API keys are rate limited!")
                     raise HTTPException(status_code=429, detail="Too many requests")
 
+    async def is_tpr_exceeded(self, model: str = None, tokens: int = 0) -> bool:
+        """Checks if the request exceeds the TPR (Tokens Per Request) limit."""
+        if not tokens:
+            return False
+
+        async with self.lock:
+            rate_limit = None
+            model_key = model or "default"
+            if model and model_key in self.rate_limits:
+                rate_limit = self.rate_limits[model_key]
+            else:
+                # fuzzy match
+                for limit_model in self.rate_limits:
+                    if limit_model != "default" and model and limit_model in model:
+                        rate_limit = self.rate_limits[limit_model]
+                        break
+            if rate_limit is None:
+                rate_limit = self.rate_limits.get("default", [])
+
+            for limit_count, limit_period in rate_limit:
+                if limit_period == -1:  # TPR limit
+                    if tokens > limit_count:
+                        # logger.warning(f"API provider for model {model_key} exceeds TPR limit ({tokens}/{limit_count}).")
+                        return True
+        return False
+
     async def is_all_rate_limited(self, model: str = None) -> bool:
         """检查是否所有的items都被速率限制
 

{beswarm-0.2.73 → beswarm-0.2.76}/beswarm/aient/aient/models/chatgpt.py

@@ -37,10 +37,18 @@ class RateLimitError(Exception):
     """Custom exception for rate limit (429) errors."""
     pass
 
+class BadRequestError(Exception):
+    """Custom exception for bad request (400) errors."""
+    pass
+
 class HTTPError(Exception):
     """Custom exception for HTTP 500 errors."""
     pass
 
+class InputTokenCountExceededError(Exception):
+    """Custom exception for input token count exceeding the maximum."""
+    pass
+
 class ConfigurationError(Exception):
     """Custom exception for configuration errors."""
     pass
@@ -802,6 +810,10 @@ class chatgpt(BaseLLM):
                             raise ModelNotFoundError(f"Model: {model or self.engine} not found!")
                         if "HTTP Error', 'status_code': 429" in processed_chunk:
                             raise RateLimitError(f"Rate limit exceeded for model: {model or self.engine}")
+                        if "HTTP Error', 'status_code': 413" in processed_chunk:
+                            raise InputTokenCountExceededError(processed_chunk)
+                        if "HTTP Error', 'status_code': 400" in processed_chunk:
+                            raise BadRequestError(f"Bad Request: {processed_chunk}")
                         if "HTTP Error', 'status_code': " in processed_chunk:
                             raise HTTPError(f"HTTP Error: {processed_chunk}")
                     yield processed_chunk
@@ -814,7 +826,7 @@ class chatgpt(BaseLLM):
                 return # Stop iteration
             except httpx.RemoteProtocolError:
                 continue
-            except httpx.ReadError:
+            except httpx.ReadError as e:
                 self.logger.warning(f"{e}, retrying...")
                 continue
             except APITimeoutError:
@@ -826,6 +838,12 @@ class chatgpt(BaseLLM):
             except RateLimitError as e:
                 self.logger.warning(f"{e}, retrying...")
                 continue
+            except InputTokenCountExceededError as e:
+                self.logger.error(f"The request body is too long: {e}")
+                raise
+            except BadRequestError as e:
+                self.logger.error(f"Bad request error: {e}")
+                raise
             except ValidationError as e:
                 self.logger.warning(f"Validation failed: {e}. Retrying with corrective prompt.")
                 need_done_prompt = [

{beswarm-0.2.73 → beswarm-0.2.76}/beswarm.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: beswarm
-Version: 0.2.73
+Version: 0.2.76
 Summary: MAS
 Requires-Python: >=3.11
 Description-Content-Type: text/markdown

{beswarm-0.2.73 → beswarm-0.2.76}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "beswarm"
-version = "0.2.73"
+version = "0.2.76"
 description = "MAS"
 readme = "README.md"
 requires-python = ">=3.11"