diff --git a/.github/workflows/doc.yaml b/.github/workflows/doc.yaml index 98e2fc9d..fba9b693 100644 --- a/.github/workflows/doc.yaml +++ b/.github/workflows/doc.yaml @@ -59,4 +59,4 @@ jobs: steps: - name: Deploy to GitHub Pages id: deployment - uses: actions/deploy-pages@v4 \ No newline at end of file + uses: actions/deploy-pages@v4 diff --git a/.pre-commit-config.yaml b/.pre-commit-config.yaml index 0e78eeb7..6f5736a8 100644 --- a/.pre-commit-config.yaml +++ b/.pre-commit-config.yaml @@ -11,30 +11,17 @@ repos: - id: check-merge-conflict - id: detect-private-key - - repo: https://github.com/psf/black - rev: 23.7.0 - hooks: - - id: black - language_version: python3.10 - args: [--line-length=100] - - repo: https://github.com/pycqa/isort - rev: 5.12.0 - hooks: - - id: isort - args: ["--profile", "black", "--filter-files"] - - repo: https://github.com/pycqa/flake8 - rev: 6.1.0 + - repo: https://github.com/myint/autoflake + rev: v2.2.0 hooks: - - id: flake8 - additional_dependencies: [flake8-docstrings] - args: [ - "--max-line-length=100", - "--max-complexity=20", - "--select=C,E,F,W,B,B950", - "--ignore=E203,E266,E501,W503", - ] + - id: autoflake + args: [ + --in-place, + --remove-all-unused-imports, + --ignore-init-module-imports + ] - repo: https://github.com/pre-commit/mirrors-mypy rev: v1.7.0 diff --git a/README.md b/README.md index f2b520f5..b01cff25 100644 --- a/README.md +++ b/README.md @@ -152,4 +152,4 @@ If you use AgentJet in your research, please cite:
[⭐ Star Us](https://github.com/modelscope/AgentJet) · [Report Bug](https://github.com/modelscope/AgentJet/issues) · [Request Feature](https://github.com/modelscope/AgentJet/issues) -
\ No newline at end of file + diff --git a/ajet/backbone/main_vllm.py b/ajet/backbone/main_vllm.py index 3f0a724c..686a35cd 100644 --- a/ajet/backbone/main_vllm.py +++ b/ajet/backbone/main_vllm.py @@ -1,4 +1,3 @@ -import atexit import os import sys from types import SimpleNamespace @@ -83,7 +82,7 @@ def submit_chat_completions(self, messages, sampling_params, request_id, tools=[ "content": message["content"], "tool_calls": message.get("tool_calls", None), "tokens": [ - TokenAndProbVllmDebug(t) for t in completion.choices[0].logprobs.content # type: ignore + TokenAndProbVllmDebug(t) for t in completion.choices[0].logprobs.content # type: ignore ], } ) @@ -131,7 +130,7 @@ async def submit_chat_completions_async(self, messages, sampling_params, request "content": message["content"], "tool_calls": message.get("tool_calls", None), "tokens": [ - TokenAndProbVllmDebug(t) for t in completion.choices[0].logprobs.content # type: ignore + TokenAndProbVllmDebug(t) for t in completion.choices[0].logprobs.content # type: ignore ], } ) diff --git a/ajet/backbone/trainer_trinity.py b/ajet/backbone/trainer_trinity.py index 1a75a1bc..8000a636 100644 --- a/ajet/backbone/trainer_trinity.py +++ b/ajet/backbone/trainer_trinity.py @@ -1,12 +1,12 @@ -import asyncio import os -from typing import Dict, List, Literal, Optional, cast - +import asyncio import datasets import openai import swanlab + from loguru import logger from transformers import AutoTokenizer +from typing import Dict, List, Literal, Optional, cast from trinity.buffer.reader import READER from trinity.buffer.reader.file_reader import TaskFileReader, _HFBatchReader from trinity.buffer.schema import FORMATTER @@ -19,9 +19,7 @@ from trinity.utils.monitor import MONITOR, Monitor from ajet.backbone.warm_up import warm_up_process -from ajet.context_tracker.multiagent_tracking import ( - MultiAgentContextTracker, -) +from ajet.context_tracker.multiagent_tracking import MultiAgentContextTracker from ajet.schema.trajectory import Sample from ajet.task_reader import dict_to_ajet_task from ajet.task_rollout.native_parallel_worker import DynamicRolloutManager @@ -65,7 +63,6 @@ def __init__( ) def convert_task(self, task: TrinityTask): - from ajet.schema.task import Task assert isinstance(task.raw_task, dict) return dict_to_ajet_task(task.raw_task) diff --git a/ajet/backbone/warm_up.py b/ajet/backbone/warm_up.py index f4c2973e..fcae673f 100644 --- a/ajet/backbone/warm_up.py +++ b/ajet/backbone/warm_up.py @@ -101,4 +101,4 @@ def warm_up_process(config): experiment_name = config.ajet.experiment_name init_parallel_rollout_logger(experiment_name) warm_up_task_judge_when_needed(config) - clean_up_tmp_ajet_dir(config) \ No newline at end of file + clean_up_tmp_ajet_dir(config) diff --git a/ajet/context_tracker/base_tracker.py b/ajet/context_tracker/base_tracker.py index 0ff706fa..948aee3e 100644 --- a/ajet/context_tracker/base_tracker.py +++ b/ajet/context_tracker/base_tracker.py @@ -1,5 +1,5 @@ from typing import List, Tuple, Union -from typing import List, Union, Tuple, Dict, Optional, Any +from typing import List, Union, Tuple, Dict, Optional from ajet.schema.task import WorkflowTask from ajet.schema.extended_msg import ( diff --git a/ajet/context_tracker/timeline_merging/timeline_merging.py b/ajet/context_tracker/timeline_merging/timeline_merging.py index e81475dd..4fb19baa 100644 --- a/ajet/context_tracker/timeline_merging/timeline_merging.py +++ b/ajet/context_tracker/timeline_merging/timeline_merging.py @@ -1,6 +1,5 @@ from typing import List -from beast_logger import print_listofdict from ajet.context_tracker.basic_tracker import ExtendedMessage diff --git a/ajet/schema/convertion.py b/ajet/schema/convertion.py index e2a6a2c0..408bbcdb 100644 --- a/ajet/schema/convertion.py +++ b/ajet/schema/convertion.py @@ -4,11 +4,10 @@ from openai.types.chat.chat_completion_message import ChatCompletionMessage from agentscope.model import ChatResponse as AgentScopeChatResponse from openai.types.completion_usage import CompletionUsage -from typing import Any, Callable, Dict, List, Literal, Type, Union +from typing import List, Type from agentscope.message import TextBlock, ToolUseBlock from agentscope._utils._common import _json_loads_with_repair from pydantic import BaseModel -from agentscope.model import ChatResponse def convert_llm_proxy_response_to_oai_response(llm_proxy_response): @@ -106,4 +105,3 @@ def convert_llm_proxy_response_to_agentscope_response( ) return parsed_response - diff --git a/ajet/schema/logprob.py b/ajet/schema/logprob.py index 42d2c572..dc736fb8 100644 --- a/ajet/schema/logprob.py +++ b/ajet/schema/logprob.py @@ -11,4 +11,4 @@ class TokenAndProb(BaseModel): token_id: int logprob: float - decoded_string: str \ No newline at end of file + decoded_string: str diff --git a/ajet/task_reader/__init__.py b/ajet/task_reader/__init__.py index 19a1a8e3..2d7d7322 100644 --- a/ajet/task_reader/__init__.py +++ b/ajet/task_reader/__init__.py @@ -123,4 +123,4 @@ def dict_to_ajet_task(task_dict: dict) -> Task: task_id=task_dict.get("task_id", ""), env_type=task_dict.get("env_type", ""), metadata=task_dict.get("metadata", {}), - ) \ No newline at end of file + ) diff --git a/ajet/task_rollout/async_llm_bridge.py b/ajet/task_rollout/async_llm_bridge.py index f43ba1c8..ff494844 100644 --- a/ajet/task_rollout/async_llm_bridge.py +++ b/ajet/task_rollout/async_llm_bridge.py @@ -3,14 +3,13 @@ import json import time import uuid -from typing import Any, Callable, Dict, List, Literal, Type, Union +from typing import Any, Callable, Dict, List, Literal, Union from loguru import logger from omegaconf import DictConfig from pydantic import BaseModel -from transformers.tokenization_utils import PreTrainedTokenizer from vllm.entrypoints.openai.tool_parsers.hermes_tool_parser import Hermes2ProToolParser from vllm.outputs import RequestOutput as VerlVllmRequestOutput diff --git a/ajet/task_runner/base_runner.py b/ajet/task_runner/base_runner.py index 65aa5c13..d8c15492 100644 --- a/ajet/task_runner/base_runner.py +++ b/ajet/task_runner/base_runner.py @@ -3,7 +3,6 @@ from threading import Lock from typing import Any, Callable, Union, Type from multiprocessing import Process, Queue -from unittest import result from ajet.context_tracker.basic_tracker import BaseContextTracker from ajet.schema.task import WorkflowOutput, WorkflowTask @@ -117,4 +116,3 @@ def run_user_workflow( else: raise ValueError(f"Unsupported wrapper type: {self.wrapper_type}") - diff --git a/ajet/task_runner/general_runner.py b/ajet/task_runner/general_runner.py index 2904cfae..7ea76710 100644 --- a/ajet/task_runner/general_runner.py +++ b/ajet/task_runner/general_runner.py @@ -1,7 +1,6 @@ -from venv import logger from ajet import AjetTuner -from ajet import Workflow, WorkflowOutput +from ajet import WorkflowOutput from ajet.context_tracker.multiagent_tracking import ( MultiAgentContextTracker, ) diff --git a/ajet/tuner.py b/ajet/tuner.py index 93602d05..aacc3ab9 100644 --- a/ajet/tuner.py +++ b/ajet/tuner.py @@ -1,4 +1,4 @@ -from typing import TYPE_CHECKING, Any, Literal, Callable, Union, Type +from typing import TYPE_CHECKING, Callable, Union, Type from ajet.context_tracker.multiagent_tracking import ( MultiAgentContextTracker, diff --git a/ajet/tuner_lib/weight_tuner/__init__.py b/ajet/tuner_lib/weight_tuner/__init__.py index 317e8699..abb540c1 100644 --- a/ajet/tuner_lib/weight_tuner/__init__.py +++ b/ajet/tuner_lib/weight_tuner/__init__.py @@ -1,4 +1,3 @@ from ajet.tuner_lib.weight_tuner.as_agentscope_model import AgentScopeModelTuner from ajet.tuner_lib.weight_tuner.as_oai_sdk_model import OpenaiClientModelTuner - diff --git a/ajet/tuner_lib/weight_tuner/as_agentscope_model.py b/ajet/tuner_lib/weight_tuner/as_agentscope_model.py index 4af1754c..67a5ef8b 100644 --- a/ajet/tuner_lib/weight_tuner/as_agentscope_model.py +++ b/ajet/tuner_lib/weight_tuner/as_agentscope_model.py @@ -1,7 +1,7 @@ -from typing import TYPE_CHECKING, Any, Literal, Type +from typing import Any, Literal, Type from agentscope._utils._common import _create_tool_from_base_model -from agentscope.model import ChatModelBase, ChatResponse, DashScopeChatModel +from agentscope.model import ChatResponse, DashScopeChatModel from loguru import logger from pydantic import BaseModel @@ -10,9 +10,6 @@ ) from ajet.task_rollout.async_llm_bridge import AgentScopeLlmProxyWithTracker -if TYPE_CHECKING: - from ajet import Workflow - class AgentScopeModelTuner(DashScopeChatModel): """ diff --git a/ajet/tuner_lib/weight_tuner/as_oai_baseurl_apikey.py b/ajet/tuner_lib/weight_tuner/as_oai_baseurl_apikey.py index ba3e9693..90c2cc72 100644 --- a/ajet/tuner_lib/weight_tuner/as_oai_baseurl_apikey.py +++ b/ajet/tuner_lib/weight_tuner/as_oai_baseurl_apikey.py @@ -1,22 +1,13 @@ import os -import asyncio -from typing import TYPE_CHECKING, Any, List, Callable, Literal, Type, Union -from loguru import logger +from typing import Any from pydantic import BaseModel, Field from ajet.context_tracker.multiagent_tracking import ( MultiAgentContextTracker, ) -from ajet.task_rollout.async_llm_bridge import OpenaiLlmProxyWithTracker -from ajet.utils.magic_mock import SpecialMagicMock -from openai.types.chat.chat_completion import ChatCompletion -from openai.resources.chat.chat import Chat, AsyncChat +from openai.resources.chat.chat import AsyncChat from openai.resources.completions import AsyncCompletions -from openai import OpenAI, AsyncOpenAI -from ajet.utils.networking import find_free_port from .experimental.as_oai_model_client import generate_auth_token -if TYPE_CHECKING: - from ajet import Workflow class MockAsyncCompletions(AsyncCompletions): async def create(self, *args, **kwargs) -> Any: # type: ignore diff --git a/ajet/tuner_lib/weight_tuner/as_oai_sdk_model.py b/ajet/tuner_lib/weight_tuner/as_oai_sdk_model.py index 943d5c2c..59248fee 100644 --- a/ajet/tuner_lib/weight_tuner/as_oai_sdk_model.py +++ b/ajet/tuner_lib/weight_tuner/as_oai_sdk_model.py @@ -1,19 +1,12 @@ -import asyncio -from typing import TYPE_CHECKING, Any, List, Callable, Literal, Type, Union -from loguru import logger -from pydantic import BaseModel +from typing import Any, List, Callable from ajet.context_tracker.multiagent_tracking import ( MultiAgentContextTracker, ) from ajet.task_rollout.async_llm_bridge import OpenaiLlmProxyWithTracker -from ajet.utils.magic_mock import SpecialMagicMock from openai.types.chat.chat_completion import ChatCompletion -from openai.resources.chat.chat import Chat, AsyncChat +from openai.resources.chat.chat import AsyncChat from openai.resources.completions import AsyncCompletions -from openai import OpenAI, AsyncOpenAI - -if TYPE_CHECKING: - from ajet import Workflow +from openai import AsyncOpenAI class MockAsyncCompletions(AsyncCompletions): @@ -80,5 +73,3 @@ async def create( ) assert isinstance(response_gen, ChatCompletion) return response_gen - - diff --git a/ajet/tuner_lib/weight_tuner/experimental/as_oai_model_server.py b/ajet/tuner_lib/weight_tuner/experimental/as_oai_model_server.py index 0c652c69..089d11eb 100644 --- a/ajet/tuner_lib/weight_tuner/experimental/as_oai_model_server.py +++ b/ajet/tuner_lib/weight_tuner/experimental/as_oai_model_server.py @@ -24,7 +24,7 @@ from loguru import logger from pydantic import BaseModel -from fastapi import FastAPI, Header, HTTPException, Request, Body +from fastapi import FastAPI, Header, HTTPException, Request from contextlib import asynccontextmanager from multiprocessing import Process from concurrent.futures import ThreadPoolExecutor @@ -239,5 +239,3 @@ def start_interchange_server(config) -> int: # return port return port - - diff --git a/ajet/utils/async_utils.py b/ajet/utils/async_utils.py index 3bb1b67e..219aba9c 100644 --- a/ajet/utils/async_utils.py +++ b/ajet/utils/async_utils.py @@ -67,4 +67,4 @@ def _patched_del(self) -> None: AsyncHttpxClientWrapper.__del__ = _patched_del print("Applied httpx aclose patch.") except ImportError: - pass \ No newline at end of file + pass diff --git a/ajet/utils/lowlevel_hook.py b/ajet/utils/lowlevel_hook.py index bdd536d0..006f17b9 100644 --- a/ajet/utils/lowlevel_hook.py +++ b/ajet/utils/lowlevel_hook.py @@ -44,4 +44,4 @@ def debug_task_init(self, coro, loop=None, name=None, context=None): asyncio.create_task = debug_create_task asyncio.AbstractEventLoop.create_task = debug_loop_create_task -patch_task_creation() \ No newline at end of file +patch_task_creation() diff --git a/ajet/utils/metric_helper/__init__.py b/ajet/utils/metric_helper/__init__.py index a9702d5d..70ce2818 100644 --- a/ajet/utils/metric_helper/__init__.py +++ b/ajet/utils/metric_helper/__init__.py @@ -14,4 +14,4 @@ def update_metrics(context_tracker_arr, metrics:dict): metrics.update(tool_metrics) if reward_metrics: metrics.update(reward_metrics) - return \ No newline at end of file + return diff --git a/ajet/utils/metric_helper/reward_metric_helper.py b/ajet/utils/metric_helper/reward_metric_helper.py index b6cf5918..49e069bf 100644 --- a/ajet/utils/metric_helper/reward_metric_helper.py +++ b/ajet/utils/metric_helper/reward_metric_helper.py @@ -11,17 +11,17 @@ - judge_time/ Judge time consumption statistics """ -from typing import List, Dict, Any, Optional +from typing import List, Dict, Any import numpy as np def extract_reward_stats_from_trajectories(trajectories: List[Any]) -> List[Dict[str, Any]]: """ Extract reward_stats from trajectories list. - + Args: trajectories: List of trajectory objects containing workflow_metadata - + Returns: List of reward_stats dictionaries """ @@ -36,10 +36,10 @@ def extract_reward_stats_from_trajectories(trajectories: List[Any]) -> List[Dict def extract_reward_stats_from_cmts(cmts: List[Any]) -> tuple[List[Dict[str, Any]], Dict[str, int]]: """ Extract reward_stats from cmts list and return debug statistics. - + Args: cmts: List of cmt objects containing workflow_metadata - + Returns: Tuple of (reward_stats_list, debug_stats) """ @@ -49,47 +49,47 @@ def extract_reward_stats_from_cmts(cmts: List[Any]) -> tuple[List[Dict[str, Any] 'has_workflow_metadata': 0, 'has_reward_stats': 0, } - + for _cmt in cmts: if hasattr(_cmt, 'workflow_metadata') and _cmt.workflow_metadata: debug_stats['has_workflow_metadata'] += 1 if 'reward_stats' in _cmt.workflow_metadata: debug_stats['has_reward_stats'] += 1 reward_stats_list.append(_cmt.workflow_metadata['reward_stats']) - + return reward_stats_list, debug_stats def compute_reward_metrics(reward_stats_list: List[Dict[str, Any]], prefix: str = "") -> Dict[str, float]: """ Compute SwanLab metrics from reward_stats list. - + Supports two data sources: 1. RM Gallery RewardStats fields (rm_raw, etc.) 2. OpenJudge fields (openjudge_xxx_raw, openjudge_xxx_contribution, etc.) - + Args: reward_stats_list: List of reward_stats dictionaries prefix: Metric name prefix (e.g., "val/" for validation phase) - + Returns: Formatted metrics dictionary ready for SwanLab reporting """ if not reward_stats_list: return {} - + n = len(reward_stats_list) metrics = {} - + # ========== Top-level Scores (General) ========== final_reward_list = [rs.get('final_reward', 0.0) for rs in reward_stats_list] fused_reward_list = [rs.get('fused_reward', 0.0) for rs in reward_stats_list] penalty_list = [rs.get('penalty', 0.0) for rs in reward_stats_list] step_reward_list = [rs.get('step_reward', 0.0) for rs in reward_stats_list] - + # Penalty statistics non_zero_penalties = [p for p in penalty_list if p != 0.0] - + # Top-level metrics metrics[f"{prefix}rewards/final_reward_mean"] = float(np.mean(final_reward_list)) metrics[f"{prefix}rewards/fused_reward_mean"] = float(np.mean(fused_reward_list)) @@ -97,110 +97,110 @@ def compute_reward_metrics(reward_stats_list: List[Dict[str, Any]], prefix: str metrics[f"{prefix}rewards/step_reward_mean"] = float(np.mean(step_reward_list)) metrics[f"{prefix}rewards/penalty_count"] = len(non_zero_penalties) metrics[f"{prefix}rewards/penalty_rate"] = len(non_zero_penalties) / n * 100 if n > 0 else 0.0 - + # ========== Detect OpenJudge Usage ========== openjudge_enabled_count = sum(1 for rs in reward_stats_list if rs.get('openjudge_enabled', False)) - + if openjudge_enabled_count > 0: # ========== OpenJudge Metrics ========== metrics[f"{prefix}rewards/openjudge_enabled_rate"] = openjudge_enabled_count / n * 100 - + # Dynamically extract OpenJudge grader fields - # Currently supported graders: report_resolution, trajectory_faithfulness, + # Currently supported graders: report_resolution, trajectory_faithfulness, # rubrics_performance, trajectory_comprehensive, information_gain, action_loop openjudge_graders = [ "report_resolution", - "trajectory_faithfulness", + "trajectory_faithfulness", "rubrics_performance", "trajectory_comprehensive", "information_gain", "action_loop", ] - + for grader_name in openjudge_graders: raw_key = f"openjudge_{grader_name}_raw" contrib_key = f"openjudge_{grader_name}_contribution" - + raw_list = [rs.get(raw_key, 0.0) for rs in reward_stats_list] contrib_list = [rs.get(contrib_key, 0.0) for rs in reward_stats_list] - + # Only report when non-zero values exist if any(v != 0.0 for v in raw_list): metrics[f"{prefix}rewards/openjudge/{grader_name}_raw_mean"] = float(np.mean(raw_list)) if any(v != 0.0 for v in contrib_list): metrics[f"{prefix}rewards/openjudge/{grader_name}_contribution_mean"] = float(np.mean(contrib_list)) - + # OpenJudge time consumption statistics grading_time_list = [rs.get('grading_time', 0.0) for rs in reward_stats_list] if any(v != 0.0 for v in grading_time_list): metrics[f"{prefix}judge_time/openjudge_grading_time_mean"] = float(np.mean(grading_time_list)) metrics[f"{prefix}judge_time/openjudge_grading_time_max"] = float(np.max(grading_time_list)) - + # ========== RM Gallery Metrics ========== # RM Gallery rm_raw_list = [rs.get('rm_raw', 0.0) for rs in reward_stats_list] rm_contribution_list = [rs.get('rm_contribution', 0.0) for rs in reward_stats_list] - + # RefJudge ref_final_raw_list = [rs.get('ref_final_raw', 0.0) for rs in reward_stats_list] ref_citation_raw_list = [rs.get('ref_citation_raw', 0.0) for rs in reward_stats_list] ref_grounding_raw_list = [rs.get('ref_grounding_raw', 0.0) for rs in reward_stats_list] ref_contribution_list = [rs.get('ref_contribution', 0.0) for rs in reward_stats_list] - + # StructureJudge structure_raw_list = [rs.get('structure_raw', 0.0) for rs in reward_stats_list] structure_contribution_list = [rs.get('structure_contribution', 0.0) for rs in reward_stats_list] - + # dimensions/ raw scores metrics[f"{prefix}rewards/dimensions/rm_raw_mean"] = float(np.mean(rm_raw_list)) metrics[f"{prefix}rewards/dimensions/ref_final_raw_mean"] = float(np.mean(ref_final_raw_list)) metrics[f"{prefix}rewards/dimensions/ref_citation_raw_mean"] = float(np.mean(ref_citation_raw_list)) metrics[f"{prefix}rewards/dimensions/ref_grounding_raw_mean"] = float(np.mean(ref_grounding_raw_list)) metrics[f"{prefix}rewards/dimensions/structure_raw_mean"] = float(np.mean(structure_raw_list)) - + # contribution/ weighted contributions metrics[f"{prefix}rewards/contribution/rm_contribution_mean"] = float(np.mean(rm_contribution_list)) metrics[f"{prefix}rewards/contribution/ref_contribution_mean"] = float(np.mean(ref_contribution_list)) metrics[f"{prefix}rewards/contribution/structure_contribution_mean"] = float(np.mean(structure_contribution_list)) - + # Enabled state statistics ref_judge_enabled_count = sum(1 for rs in reward_stats_list if rs.get('ref_judge_enabled', False)) if ref_judge_enabled_count > 0: metrics[f"{prefix}rewards/ref_judge_enabled_rate"] = ref_judge_enabled_count / n * 100 - + structure_judge_enabled_count = sum(1 for rs in reward_stats_list if rs.get('structure_judge_enabled', False)) if structure_judge_enabled_count > 0: metrics[f"{prefix}rewards/structure_judge_enabled_rate"] = structure_judge_enabled_count / n * 100 - + # Time consumption statistics rm_time_list = [rs.get('rm_time', 0.0) for rs in reward_stats_list] refstruc_time_list = [rs.get('refstruc_time', 0.0) for rs in reward_stats_list] - + metrics[f"{prefix}judge_time/rm_time_mean"] = float(np.mean(rm_time_list)) metrics[f"{prefix}judge_time/refstruc_time_mean"] = float(np.mean(refstruc_time_list)) - + if rm_time_list: metrics[f"{prefix}judge_time/rm_time_max"] = float(np.max(rm_time_list)) if refstruc_time_list: metrics[f"{prefix}judge_time/refstruc_time_max"] = float(np.max(refstruc_time_list)) - + # ========== General Time Consumption Statistics ========== judge_total_time_list = [rs.get('judge_total_time', 0.0) for rs in reward_stats_list] if any(v != 0.0 for v in judge_total_time_list): metrics[f"{prefix}judge_time/judge_total_time_mean"] = float(np.mean(judge_total_time_list)) metrics[f"{prefix}judge_time/judge_total_time_max"] = float(np.max(judge_total_time_list)) - + return metrics def compute_reward_metrics_from_trajectories(trajectories: List[Any]) -> Dict[str, float]: """ Training phase: Extract reward_stats from trajectories and compute metrics. - + Args: trajectories: List of trajectory objects - + Returns: Formatted metrics dictionary """ @@ -211,21 +211,21 @@ def compute_reward_metrics_from_trajectories(trajectories: List[Any]) -> Dict[st def compute_reward_metrics_from_cmts(cmts: List[Any], print_debug: bool = True) -> Dict[str, float]: """ Validation phase: Extract reward_stats from cmts and compute metrics. - + Args: cmts: List of cmt objects print_debug: Whether to print debug information - + Returns: Formatted metrics dictionary (with "val_reward/" prefix) """ reward_stats_list, debug_stats = extract_reward_stats_from_cmts(cmts) - + if print_debug: print(f"\n[DEBUG eval_dataset()] reward_stats statistics:") print(f" - Total cmts count: {debug_stats['total_cmts']}") print(f" - Has workflow_metadata: {debug_stats['has_workflow_metadata']}") print(f" - Has reward_stats: {debug_stats['has_reward_stats']}") print(f" - Extracted samples count: {len(reward_stats_list)}") - + return compute_reward_metrics(reward_stats_list, prefix="val_") diff --git a/ajet/utils/metric_helper/save_trajectory_as_json.py b/ajet/utils/metric_helper/save_trajectory_as_json.py index 0e380abc..344a6ab4 100644 --- a/ajet/utils/metric_helper/save_trajectory_as_json.py +++ b/ajet/utils/metric_helper/save_trajectory_as_json.py @@ -53,4 +53,4 @@ def save_trajectory_as_json(ctx_trackers, global_steps, prefix="train"): # Print confirmation for evaluation trajectories if prefix != "train": - print(f"Saved trajectory to {traj_file_path}") \ No newline at end of file + print(f"Saved trajectory to {traj_file_path}") diff --git a/ajet/utils/msg_converter.py b/ajet/utils/msg_converter.py index 0437f5ca..46c02128 100644 --- a/ajet/utils/msg_converter.py +++ b/ajet/utils/msg_converter.py @@ -21,8 +21,7 @@ {"role": "user/assistant/system", "content": "..."} """ -import json -from typing import List, Dict, Any, Union +from typing import List, Dict, Any diff --git a/ajet/utils/networking.py b/ajet/utils/networking.py index 9ed29c74..f2fed5ac 100644 --- a/ajet/utils/networking.py +++ b/ajet/utils/networking.py @@ -34,4 +34,4 @@ def get_host_ip(interface=None): except Exception: - return "127.0.0.1" \ No newline at end of file + return "127.0.0.1" diff --git a/ajet/utils/testing_utils.py b/ajet/utils/testing_utils.py index 22be6092..31f006c2 100644 --- a/ajet/utils/testing_utils.py +++ b/ajet/utils/testing_utils.py @@ -11,7 +11,6 @@ from loguru import logger from ajet.utils.dynamic_import import dynamic_import -from ajet.utils.sington import singleton class TestSuccessException(Exception): @@ -19,7 +18,6 @@ class TestSuccessException(Exception): All test is done, end the program early with exception. """ - pass class TestFailException(Exception): @@ -27,7 +25,6 @@ class TestFailException(Exception): Test has failed, end the program early with exception. """ - pass class BaseProbe(object): diff --git a/ajet/utils/thread_executors.py b/ajet/utils/thread_executors.py index 9c8ea634..1ab02baf 100644 --- a/ajet/utils/thread_executors.py +++ b/ajet/utils/thread_executors.py @@ -19,4 +19,4 @@ def __init__(self, max_workers=64): self.executor = concurrent.futures.ThreadPoolExecutor(max_workers=max_workers) def get_shared_executor(self) -> concurrent.futures.ThreadPoolExecutor: - return self.executor \ No newline at end of file + return self.executor diff --git a/docs/_toc.yml b/docs/_toc.yml index ffa745f4..7eb76610 100644 --- a/docs/_toc.yml +++ b/docs/_toc.yml @@ -8,7 +8,7 @@ parts: - file: en/installation.md - file: en/quickstart.md - file: en/tune_your_first_agent.md - + - caption: Example chapters: - file: en/example_math_agent.md @@ -17,13 +17,13 @@ parts: - file: en/example_learning_to_ask.md - file: en/example_frozenlake.md - file: en/example_countdown.md - + - caption: Component chapters: - file: en/workflow.md - file: en/data_pipeline.md - file: en/task_judger.md - + - caption: Deep Dive chapters: - file: en/configuration.md @@ -31,7 +31,7 @@ parts: - file: en/beast_logger.md - file: en/data_generation.md - file: en/example_tracing_feedback_loop.md - + # --- 中文部分 --- - caption: 教程 @@ -40,7 +40,7 @@ parts: - file: zh/installation.md - file: zh/quickstart.md - file: zh/tune_your_first_agent.md - + - caption: 示例 chapters: - file: zh/example_math_agent.md @@ -49,13 +49,13 @@ parts: - file: zh/example_learning_to_ask.md - file: zh/example_frozenlake.md - file: zh/example_countdown.md - + - caption: 组件 chapters: - file: zh/workflow.md - file: zh/data_pipeline.md - file: zh/task_judger.md - + - caption: 深入探索 chapters: - file: zh/configuration.md @@ -63,4 +63,3 @@ parts: - file: zh/beast_logger.md - file: zh/data_generation.md - file: zh/example_tracing_feedback_loop.md - diff --git a/docs/en/debugging_guide.md b/docs/en/debugging_guide.md index 0a938004..ff7563a2 100644 --- a/docs/en/debugging_guide.md +++ b/docs/en/debugging_guide.md @@ -104,4 +104,3 @@ Then, the modified launch.json will be | **VSCode Extension** | Python | Python + Ray Distributed Debugger | | **Launch Mode** | `F5` standard launch (via `launch.json`) | Command line execution with `ajet ... --debug="TAG"` | | **Commandline** | `--backbone=debug` | `--debug="TAG1\|TAG2\|TAG3"` | - diff --git a/docs/en/example_countdown.md b/docs/en/example_countdown.md index e214e4d5..ff8ec4e3 100644 --- a/docs/en/example_countdown.md +++ b/docs/en/example_countdown.md @@ -201,4 +201,3 @@ However, tuning resolves these issues, as shown in the example below: ![After tuning](https://img.alicdn.com/imgextra/i4/O1CN01C3kUnV221zjPi30rd_!!6000000007061-2-tps-1650-730.png) > **Token-level Visualization:** These detailed logs are generated by Beast-Logger. See [Beast-Logger Usage](./beast_logger.md) for more details. - diff --git a/docs/en/example_learning_to_ask.md b/docs/en/example_learning_to_ask.md index c3d4bcf9..d5a17abe 100644 --- a/docs/en/example_learning_to_ask.md +++ b/docs/en/example_learning_to_ask.md @@ -135,7 +135,7 @@ We provide two implmentations of the agent based on AgentScope and langchain: ```python # get the trainable llm llm_info=tuner.as_oai_baseurl_apikey() - + # create the langchain agent llm=ChatOpenAI( base_url=llm_info.base_url, @@ -145,7 +145,7 @@ We provide two implmentations of the agent based on AgentScope and langchain: model=llm, system_prompt=system_prompt, ) - + # build messages and send to the agent msg=[ {"role": x["role"], "content": x["content"]} for x in messages @@ -153,7 +153,7 @@ We provide two implmentations of the agent based on AgentScope and langchain: result = agent.invoke({ "messages": msg, # type: ignore }) - + response = result["messages"][-1].content reward = await reward_fn_with_semaphore(msg, response, truth_action, truth_info) return WorkflowOutput(reward=reward) @@ -221,4 +221,4 @@ Agent: Has itching or reddening appeared around this bite site recently without The question becomes more precise and informative, guiding the user to provide clinically relevant details. -> To learn more about the task and results on larger models, refer to [Grounded in Reality: Learning and Deploying Proactive LLM from Offline Logs](https://arxiv.org/abs/2510.25441). \ No newline at end of file +> To learn more about the task and results on larger models, refer to [Grounded in Reality: Learning and Deploying Proactive LLM from Offline Logs](https://arxiv.org/abs/2510.25441). diff --git a/docs/en/hardware_related_solution.md b/docs/en/hardware_related_solution.md index c2cad4a3..9743d384 100644 --- a/docs/en/hardware_related_solution.md +++ b/docs/en/hardware_related_solution.md @@ -17,4 +17,4 @@ This document records a list of **Hardware Related** issues for future reference ```bash export NCCL_NVLS_ENABLE=0 - ``` \ No newline at end of file + ``` diff --git a/docs/en/support_agentscope.md b/docs/en/support_agentscope.md index b3129191..e551e4d9 100644 --- a/docs/en/support_agentscope.md +++ b/docs/en/support_agentscope.md @@ -223,4 +223,3 @@ This article introduce the way to convert different types of ways to convert you else: is_success = False return WorkflowOutput(reward=(1.0 if is_success else 0.0), metadata={"final_answer": final_answer}) ``` - diff --git a/docs/en/support_http.md b/docs/en/support_http.md index 32474904..0bf3ab3d 100644 --- a/docs/en/support_http.md +++ b/docs/en/support_http.md @@ -93,5 +93,3 @@ in this AI era, you can always start from scratch and build your own "high-scrap ... ``` - - diff --git a/docs/en/support_langchain.md b/docs/en/support_langchain.md index 6e645dcc..d1e12890 100644 --- a/docs/en/support_langchain.md +++ b/docs/en/support_langchain.md @@ -84,5 +84,3 @@ This article introduce the way to convert different types of ways to convert you ... ``` - - diff --git a/docs/en/support_oaisdk.md b/docs/en/support_oaisdk.md index 5268ab42..b60b03e3 100644 --- a/docs/en/support_oaisdk.md +++ b/docs/en/support_oaisdk.md @@ -88,6 +88,3 @@ This article introduce the way to convert different types of ways to convert you ... ``` - - - diff --git a/docs/index.md b/docs/index.md index ba98cd7f..5583fa69 100644 --- a/docs/index.md +++ b/docs/index.md @@ -170,4 +170,3 @@ The internal system orchestrates several specialized modules to handle the compl

查看中文文档

完整的中文教程和指南。

--> - diff --git a/docs/javascripts/animations.js b/docs/javascripts/animations.js index 00e3603b..a5dc584a 100644 --- a/docs/javascripts/animations.js +++ b/docs/javascripts/animations.js @@ -399,4 +399,3 @@ }; })(); - diff --git a/docs/javascripts/code-zoom.js b/docs/javascripts/code-zoom.js index e2a08f6d..22d3d624 100644 --- a/docs/javascripts/code-zoom.js +++ b/docs/javascripts/code-zoom.js @@ -1,2 +1 @@ /* Code zoom - placeholder */ - diff --git a/docs/javascripts/responsive.js b/docs/javascripts/responsive.js index 663e371f..d57c4db2 100644 --- a/docs/javascripts/responsive.js +++ b/docs/javascripts/responsive.js @@ -353,4 +353,3 @@ }; })(); - diff --git a/docs/javascripts/search-fix.js b/docs/javascripts/search-fix.js index e8436240..444f2af9 100644 --- a/docs/javascripts/search-fix.js +++ b/docs/javascripts/search-fix.js @@ -1,2 +1 @@ /* Search fix - placeholder */ - diff --git a/docs/javascripts/tabbed-code.js b/docs/javascripts/tabbed-code.js index 880ba944..cfd19559 100644 --- a/docs/javascripts/tabbed-code.js +++ b/docs/javascripts/tabbed-code.js @@ -174,4 +174,3 @@ // Export for manual re-initialization if needed window.initTabbedSets = initTabbedSets; })(); - diff --git a/docs/requirements.txt b/docs/requirements.txt index 968bb898..db4f637c 100644 --- a/docs/requirements.txt +++ b/docs/requirements.txt @@ -11,4 +11,3 @@ pymdown-extensions==10.16.1 # Syntax highlighting Pygments>=2.18.0 - diff --git a/docs/stylesheets/animations.css b/docs/stylesheets/animations.css index 2129b6d8..9d390ff7 100644 --- a/docs/stylesheets/animations.css +++ b/docs/stylesheets/animations.css @@ -875,4 +875,3 @@ img { .duration-fast { animation-duration: var(--rm-transition-fast); } .duration-normal { animation-duration: var(--rm-transition-normal); } .duration-slow { animation-duration: var(--rm-transition-slow); } - diff --git a/docs/stylesheets/feature-cards.css b/docs/stylesheets/feature-cards.css index 03fe0464..5865ca73 100644 --- a/docs/stylesheets/feature-cards.css +++ b/docs/stylesheets/feature-cards.css @@ -540,4 +540,3 @@ .dark { --inline-icon-filter: invert(1) hue-rotate(180deg); } - diff --git a/docs/stylesheets/flowchart.css b/docs/stylesheets/flowchart.css index 175dc123..345b94f1 100644 --- a/docs/stylesheets/flowchart.css +++ b/docs/stylesheets/flowchart.css @@ -400,4 +400,3 @@ font-size: 0.875rem; margin-bottom: 0.5rem; } - diff --git a/docs/stylesheets/jupyter-simple.css b/docs/stylesheets/jupyter-simple.css index 401abf67..864c59bd 100644 --- a/docs/stylesheets/jupyter-simple.css +++ b/docs/stylesheets/jupyter-simple.css @@ -256,4 +256,3 @@ article .cell.markdown ol:last-child { top: 0.75rem; } } - diff --git a/docs/stylesheets/syntax-highlight.css b/docs/stylesheets/syntax-highlight.css index 3c651185..7cfcf6ba 100644 --- a/docs/stylesheets/syntax-highlight.css +++ b/docs/stylesheets/syntax-highlight.css @@ -303,4 +303,3 @@ .dark .codehilite .language-json .nd { color: #79c0ff; } - diff --git a/docs/stylesheets/tuner_v2.md b/docs/stylesheets/tuner_v2.md index c8766e31..c19509cd 100644 --- a/docs/stylesheets/tuner_v2.md +++ b/docs/stylesheets/tuner_v2.md @@ -78,4 +78,4 @@ response = client.chat.completions.create( ) -``` \ No newline at end of file +``` diff --git a/install.sh b/install.sh index bf0400b6..2306bad0 100755 --- a/install.sh +++ b/install.sh @@ -203,7 +203,7 @@ download_binary_and_run_installer() { local _checksum_value # destructure selected archive info into locals - case "$_artifact_name" in + case "$_artifact_name" in "uv-aarch64-apple-darwin.tar.gz") _arch="aarch64-apple-darwin" _zip_ext=".tar.gz" @@ -529,7 +529,7 @@ replace_home() { json_binary_aliases() { local _arch="$1" - case "$_arch" in + case "$_arch" in "aarch64-apple-darwin") echo '{}' ;; @@ -612,7 +612,7 @@ aliases_for_binary() { local _bin="$1" local _arch="$2" - case "$_arch" in + case "$_arch" in "aarch64-apple-darwin") case "$_bin" in *) @@ -793,7 +793,7 @@ select_archive_for_arch() { # try each archive, checking runtime conditions like libc versions # accepting the first one that matches, as it's the best match - case "$_true_arch" in + case "$_true_arch" in "aarch64-apple-darwin") _archive="uv-aarch64-apple-darwin.tar.gz" if [ -n "$_archive" ]; then diff --git a/mkdocs.yml b/mkdocs.yml index 6a06d4ad..a6fa0585 100644 --- a/mkdocs.yml +++ b/mkdocs.yml @@ -147,4 +147,3 @@ extra_javascript: - javascripts/nav-scroll-fix.js - javascripts/animations.js - javascripts/responsive.js - diff --git a/pyproject.toml b/pyproject.toml index aee28b2b..856cddca 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -113,4 +113,4 @@ known_third_party = ["wandb"] [project.urls] -"Homepage" = "https://github.com/modelscope/AgentJet" \ No newline at end of file +"Homepage" = "https://github.com/modelscope/AgentJet" diff --git a/scripts/display_dataset.py b/scripts/display_dataset.py index e3132bc4..6d125e5c 100644 --- a/scripts/display_dataset.py +++ b/scripts/display_dataset.py @@ -1,10 +1,5 @@ import argparse -import glob -import os -import time -from beast_logger import print_list -from huggingface_hub import snapshot_download parser = argparse.ArgumentParser(description="download Hugging Face dataset") parser.add_argument("--target", default="openai/gsm8k", type=str, help="HuggingFace dataset name") diff --git a/tests/bench/benchmark_math/benchmark_math.py b/tests/bench/benchmark_math/benchmark_math.py index 9d8397ca..973f9ea2 100644 --- a/tests/bench/benchmark_math/benchmark_math.py +++ b/tests/bench/benchmark_math/benchmark_math.py @@ -1,5 +1,4 @@ # flake8: noqa -import os import time from ajet.utils.testing_utils import BenchmarkProbe, singleton diff --git a/tests/test_networking.py b/tests/test_networking.py deleted file mode 100644 index 913fc341..00000000 --- a/tests/test_networking.py +++ /dev/null @@ -1,56 +0,0 @@ -import socket -import unittest -import sys -import os -import importlib.util - -# Load the module directly to avoid top-level package import issues -# caused by broken dependencies in other parts of the codebase. -# We are testing a standalone utility, so we don't need the whole app context. -module_path = os.path.abspath(os.path.join(os.path.dirname(__file__), '..', 'ajet', 'utils', 'networking.py')) -spec = importlib.util.spec_from_file_location("networking", module_path) -networking = importlib.util.module_from_spec(spec) -spec.loader.exec_module(networking) - -find_free_port = networking.find_free_port -get_host_ip = networking.get_host_ip - -class TestNetworking(unittest.TestCase): - def test_find_free_port(self): - """Test that find_free_port returns a valid integer port.""" - port = find_free_port() - self.assertIsInstance(port, int) - self.assertGreater(port, 0) - self.assertLess(port, 65536) - - # Verify the port is valid to bind to (it should have been released) - with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s: - try: - s.bind(('', port)) - except OSError: - # It's possible the port was taken immediately by another process - # but unlikely in a test environment. - pass - - def test_get_host_ip(self): - """Test that get_host_ip returns a valid IP string.""" - ip = get_host_ip() - self.assertIsInstance(ip, str) - parts = ip.split('.') - self.assertEqual(len(parts), 4) - for part in parts: - if part == 'localhost': - continue - self.assertTrue(part.isdigit(), f"Part {part} is not a digit") - self.assertTrue(0 <= int(part) <= 255) - - def test_get_host_ip_with_interface(self): - """Test get_host_ip with a non-existent interface falls back to default behavior.""" - # This will likely fail the interface specific block and fall back to the connect method - ip = get_host_ip(interface="invalid_interface_XYZ") - self.assertIsInstance(ip, str) - parts = ip.split('.') - self.assertEqual(len(parts), 4) - -if __name__ == '__main__': - unittest.main() diff --git a/tutorial/README.md b/tutorial/README.md index e5811d8d..8e5288a9 100644 --- a/tutorial/README.md +++ b/tutorial/README.md @@ -8,4 +8,4 @@ Explore our rich library of examples to kickstart your journey. - Example Benchmark Tracking System: - https://benchmark.agent-matrix.com/examples \ No newline at end of file + https://benchmark.agent-matrix.com/examples diff --git a/tutorial/example_appworld/appworld.py b/tutorial/example_appworld/appworld.py index d8b647e7..01816e67 100644 --- a/tutorial/example_appworld/appworld.py +++ b/tutorial/example_appworld/appworld.py @@ -1,5 +1,4 @@ from agentscope.message import Msg -from pydantic import Field from ajet import AjetTuner, Workflow, WorkflowOutput, WorkflowTask diff --git a/tutorial/example_appworld/appworld_oai_sdk.py b/tutorial/example_appworld/appworld_oai_sdk.py index 534ec00b..dc18db34 100644 --- a/tutorial/example_appworld/appworld_oai_sdk.py +++ b/tutorial/example_appworld/appworld_oai_sdk.py @@ -1,5 +1,4 @@ from agentscope.message import Msg -from pydantic import Field from ajet import Workflow, WorkflowOutput, WorkflowTask from ajet import AjetTuner diff --git a/tutorial/example_learn2ask/data_preprocess/llm_info_extraction.py b/tutorial/example_learn2ask/data_preprocess/llm_info_extraction.py index 75e76c87..070b1612 100644 --- a/tutorial/example_learn2ask/data_preprocess/llm_info_extraction.py +++ b/tutorial/example_learn2ask/data_preprocess/llm_info_extraction.py @@ -145,4 +145,4 @@ def parse_llm_output(output_str): return result except Exception as e: - return f"Error parsing output: [{repr(output_str)}] error = {str(e)}" \ No newline at end of file + return f"Error parsing output: [{repr(output_str)}] error = {str(e)}" diff --git a/tutorial/example_learn2ask/data_preprocess/message_splitter.py b/tutorial/example_learn2ask/data_preprocess/message_splitter.py index a82506a4..06362b05 100644 --- a/tutorial/example_learn2ask/data_preprocess/message_splitter.py +++ b/tutorial/example_learn2ask/data_preprocess/message_splitter.py @@ -97,4 +97,4 @@ def split_session_to_json_lines(session): json_lines = split_session_to_json_lines(example_session) print("JSON lines output:") for i, line in enumerate(json_lines): - print(f"Line {i + 1}: {line}") \ No newline at end of file + print(f"Line {i + 1}: {line}") diff --git a/tutorial/example_learn2ask/data_preprocess/step1.py b/tutorial/example_learn2ask/data_preprocess/step1.py index d2ba27c6..d4533ffa 100644 --- a/tutorial/example_learn2ask/data_preprocess/step1.py +++ b/tutorial/example_learn2ask/data_preprocess/step1.py @@ -28,14 +28,14 @@ def process_jsonl_file( str: Success message or error information """ progress_file = output_file + ".progress" - + def load_progress(): """Load progress from progress file. Returns set of completed line numbers.""" if os.path.exists(progress_file): with open(progress_file, "r", encoding="utf-8") as f: return set(int(line.strip()) for line in f if line.strip()) return set() - + def process_single_session(args): """Worker function to process a single session.""" line_num, line = args @@ -54,41 +54,41 @@ def process_single_session(args): return line_num, None, f"Warning: Skipping invalid JSON at line {line_num}: {e}" except Exception as e: return line_num, None, f"Warning: Error processing session at line {line_num}: {e}" - + try: # Load previous progress completed_lines = load_progress() if completed_lines: print(f"Resuming from previous progress. {len(completed_lines)} lines already completed.") - + # Read all lines first with open(input_file, "r", encoding="utf-8") as infile: all_lines = list(enumerate(infile, 1)) - + total_lines = len(all_lines) # Filter out already completed lines lines_to_process = [(num, line) for num, line in all_lines if num not in completed_lines] - + if not lines_to_process: print("All lines already processed.") # Clean up progress file if os.path.exists(progress_file): os.remove(progress_file) return f"All lines already processed. Results in {output_file}" - + print(f"Processing {len(lines_to_process)} remaining lines out of {total_lines} total.") - + # State for ordered writing results_buffer = {} # line_num -> processed_lines next_line_to_write = min(num for num, _ in lines_to_process) write_lock = threading.Lock() progress_lock = threading.Lock() - + # Open output file in append mode if resuming, otherwise write mode file_mode = "a" if completed_lines else "w" outfile = open(output_file, file_mode, encoding="utf-8") progress_out = open(progress_file, "a", encoding="utf-8") - + def flush_buffer(): """Write all consecutive completed results from buffer to file.""" nonlocal next_line_to_write @@ -106,28 +106,28 @@ def flush_buffer(): # Skip lines that were already completed or empty while next_line_to_write <= total_lines and next_line_to_write not in dict(lines_to_process): next_line_to_write += 1 - + try: # Process sessions in parallel using ThreadPoolExecutor with ThreadPoolExecutor(max_workers=max_workers) as executor: futures = {executor.submit(process_single_session, item): item[0] for item in lines_to_process} - + for future in as_completed(futures): line_num, processed_lines, error = future.result() if error: print(error) - + with write_lock: results_buffer[line_num] = processed_lines flush_buffer() finally: outfile.close() progress_out.close() - + # Clean up progress file on successful completion if os.path.exists(progress_file): os.remove(progress_file) - + return f"Successfully processed. Results saved to {output_file}" except Exception as e: @@ -177,7 +177,7 @@ def process_session(session, model_call_mode="online_api", max_retries=3, **kwar print(f"Attempt {attempt + 1} failed with exception: {str(e)}") if attempt < max_retries - 1: time.sleep(24) # Shorter wait for testing - + if info_set is None: raise Exception(f"failed to generate {session}") data["info_set"] = info_set @@ -206,4 +206,4 @@ def process_session(session, model_call_mode="online_api", max_retries=3, **kwar model_call_mode=args.model_call_mode, # Additional parameters for API calls ) - ) \ No newline at end of file + ) diff --git a/tutorial/example_learn2ask/data_preprocess/step2.py b/tutorial/example_learn2ask/data_preprocess/step2.py index 849aa510..9d546b0c 100644 --- a/tutorial/example_learn2ask/data_preprocess/step2.py +++ b/tutorial/example_learn2ask/data_preprocess/step2.py @@ -26,7 +26,7 @@ def main(input_file_path, output_file_path): if_keep, info_set, decision = process_message(data) if not if_keep: continue - + new_item = { 'main_query':'[no query]', 'init_messages': data['messages'], @@ -56,4 +56,4 @@ def main(input_file_path, output_file_path): args = parser.parse_args() - main(args.input_file, args.output_file) \ No newline at end of file + main(args.input_file, args.output_file) diff --git a/tutorial/example_learn2ask/learn2ask.md b/tutorial/example_learn2ask/learn2ask.md index d5afd08f..811d37f9 100644 --- a/tutorial/example_learn2ask/learn2ask.md +++ b/tutorial/example_learn2ask/learn2ask.md @@ -99,4 +99,4 @@ The agent's question is more precise and informative, providing two specific and ## Next -To learn more about the task and results on larger models, refer to [Grounded in Reality: Learning and Deploying Proactive LLM from Offline Logs](https://arxiv.org/abs/2510.25441). \ No newline at end of file +To learn more about the task and results on larger models, refer to [Grounded in Reality: Learning and Deploying Proactive LLM from Offline Logs](https://arxiv.org/abs/2510.25441). diff --git a/tutorial/example_learn2ask/learn2ask_langchain.py b/tutorial/example_learn2ask/learn2ask_langchain.py index d728ac64..b15d7309 100644 --- a/tutorial/example_learn2ask/learn2ask_langchain.py +++ b/tutorial/example_learn2ask/learn2ask_langchain.py @@ -4,7 +4,6 @@ import asyncio import threading -from agentscope.message import Msg from loguru import logger from ajet import AjetTuner, Workflow, WorkflowOutput, WorkflowTask @@ -174,26 +173,26 @@ async def execute(self, workflow_task: WorkflowTask, tuner: AjetTuner) -> Workfl assert isinstance(messages, list) truth_action = workflow_task.task.metadata["decision_truth"] or "continue" truth_info = workflow_task.task.metadata["info_truth"] - + llm_info=tuner.as_oai_baseurl_apikey() - + llm=ChatOpenAI( base_url=llm_info.base_url, api_key=lambda:llm_info.api_key, ) - + agent=create_agent( model=llm, system_prompt=system_prompt, ) - + msg=[ {"role": x["role"], "content": x["content"]} for x in messages ] result = agent.invoke({ "messages": msg, # type: ignore }) - + response = result["messages"][-1].content reward = await reward_fn_with_semaphore(msg, response, truth_action, truth_info) return WorkflowOutput(reward=reward) diff --git a/tutorial/example_ma_deepresearch/ma_deepresearch.py b/tutorial/example_ma_deepresearch/ma_deepresearch.py index 9eaba34c..d044458b 100644 --- a/tutorial/example_ma_deepresearch/ma_deepresearch.py +++ b/tutorial/example_ma_deepresearch/ma_deepresearch.py @@ -2,13 +2,8 @@ from loguru import logger from pydantic import BaseModel, Field from ajet import AjetTuner, Workflow, WorkflowOutput, WorkflowTask -from openai.types.chat.chat_completion import ChatCompletion -from openai.types.chat import ChatCompletionMessageToolCall -from textwrap import dedent -import json import os -import asyncio import requests diff --git a/tutorial/example_math_agent/math_agent_langchain.py b/tutorial/example_math_agent/math_agent_langchain.py index c47fc355..4c99d240 100644 --- a/tutorial/example_math_agent/math_agent_langchain.py +++ b/tutorial/example_math_agent/math_agent_langchain.py @@ -1,12 +1,6 @@ -from loguru import logger from ajet import AjetTuner, Workflow, WorkflowOutput, WorkflowTask -from openai.types.chat.chat_completion import ChatCompletion -from openai.types.chat import ChatCompletionMessageToolCall from textwrap import dedent -import json -import asyncio -import requests from langchain.agents import create_agent @@ -30,7 +24,7 @@ async def execute(self, workflow_task: WorkflowTask, tuner: AjetTuner) -> Workfl url_and_apikey = tuner.as_oai_baseurl_apikey() base_url = url_and_apikey.base_url api_key = url_and_apikey.api_key - + from langchain_openai import ChatOpenAI llm=ChatOpenAI( base_url=base_url, @@ -40,10 +34,10 @@ async def execute(self, workflow_task: WorkflowTask, tuner: AjetTuner) -> Workfl model=llm, system_prompt=self.system_prompt, ) - + # take out query query = workflow_task.task.main_query - + response = agent.invoke({ "messages": [ { @@ -52,6 +46,6 @@ async def execute(self, workflow_task: WorkflowTask, tuner: AjetTuner) -> Workfl } ], }) - + final_answer = response['messages'][-1].content - return WorkflowOutput(reward=None, metadata={"final_answer": final_answer}) \ No newline at end of file + return WorkflowOutput(reward=None, metadata={"final_answer": final_answer}) diff --git a/tutorial/example_math_agent/math_agent_oai_sdk.py b/tutorial/example_math_agent/math_agent_oai_sdk.py index 8304f14d..24bf47ec 100644 --- a/tutorial/example_math_agent/math_agent_oai_sdk.py +++ b/tutorial/example_math_agent/math_agent_oai_sdk.py @@ -1,4 +1,3 @@ -from loguru import logger from ajet import AjetTuner, Workflow, WorkflowOutput, WorkflowTask from openai.types.chat.chat_completion import ChatCompletion from openai.types.chat import ChatCompletionMessageToolCall diff --git a/tutorial/example_math_agent/math_agent_raw_http.py b/tutorial/example_math_agent/math_agent_raw_http.py index 6608e2be..69dfd949 100644 --- a/tutorial/example_math_agent/math_agent_raw_http.py +++ b/tutorial/example_math_agent/math_agent_raw_http.py @@ -1,11 +1,6 @@ -from loguru import logger from ajet import AjetTuner, Workflow, WorkflowOutput, WorkflowTask -from openai.types.chat.chat_completion import ChatCompletion -from openai.types.chat import ChatCompletionMessageToolCall from textwrap import dedent -import json -import asyncio import requests @@ -57,8 +52,3 @@ async def execute(self, workflow_task: WorkflowTask, tuner: AjetTuner) -> Workfl ) final_answer = response.json()['choices'][0]['message']['content'] return WorkflowOutput(reward=None, metadata={"final_answer": final_answer}) - - - - - diff --git a/tutorial/example_werewolves/start.py b/tutorial/example_werewolves/start.py index 554b0977..879b6101 100644 --- a/tutorial/example_werewolves/start.py +++ b/tutorial/example_werewolves/start.py @@ -12,7 +12,7 @@ from agentscope.agent import ReActAgent from agentscope.formatter import DashScopeMultiAgentFormatter, OpenAIMultiAgentFormatter -from agentscope.model import DashScopeChatModel, OpenAIChatModel +from agentscope.model import OpenAIChatModel from loguru import logger from pydantic import Field