Pre-commit fix #3

binary-husky · 2026-01-16T09:25:30Z

No description provided.

Former-commit-id: 84192c2dec51311450d1653e97979a50f39354cf

Title: Fix review suggestions https://aliyuque.antfin.com/bayotg/vpbyh0/pguna8cwykkggv0h Link: https://code.alibaba-inc.com/DAIL-LLM/astune/codereview/24622110 Former-commit-id: 38a4ab61dedcf8a0dd8214e25f85d7df60d89998

Former-commit-id: d20ebec369ce1abbf40b9684bf412b50599c1937

Title: remove trash and remove chinese Link: https://code.alibaba-inc.com/DAIL-LLM/astune/codereview/24641938 Former-commit-id: c4c183ad7d47c21dc969f151a86edc859bde6641

Added detailed observations on agent behavior and performance in werewolf simulations, including win rates and strategies.

Added a centered image and a tagline to the README.

Added detailed explanation of the Werewolves role-playing game as a multi-agent environment, including terms, workflow setup, configuration, debugging, and training instructions.

* Dev werewolves: change yaml and retry logic (#3) * Expand Werewolves documentation with multi-agent setup (#2) Added detailed explanation of the Werewolves role-playing game as a multi-agent environment, including terms, workflow setup, configuration, debugging, and training instructions. * rpg werewolves yaml update * werewolf expected_steps=3 (there are 3 werewolves, meaning usually 3 samples are generated per run) * fix all pre-commit (#4) * Document Math Agent training process This document outlines the steps to prepare data, build an Agent and Workflow, set up Rewards, and train a Math Agent from scratch, including code snippets and configuration details. * Revise AppWorld training documentation This update enhances the documentation for the AppWorld training process, providing detailed steps on preparing the environment, configuring the AgentScope workflow, and setting up the reward module. It also includes instructions for debugging and starting training, along with reference results and case observations. * Add Data Generator Reader (#5) * Add data generator reader --------- Co-authored-by: Xuchen Pan <32844285+pan-x-c@users.noreply.github.com> Co-authored-by: Yorick He <908536269@qq.com>

* rpg werewolves yaml update * werewolf expected_steps=3 (there are 3 werewolves, meaning usually 3 samples are generated per run) * for no-eos message, pad eos but loss_mask=0 * align math * add swanlab

* feat: add deduplication filter * fix: spacing in deduplication filter db_path configuration * refactor: remove unused chroma_id variable in embedding client

* docs: update docs * docs: fix images, formats, typos * docs: add hyperlinks to documentation sections in quickstart

* commit bad math agent * benchmark successful * connect test bot to test data server * patch native rollout * simplify tool call logging * enforce ray kill in autokill

feat: Add comprehensive metric tracking and trajectory persistence

… scripts

…sion

gemini-code-assist · 2026-01-16T09:29:05Z

Summary of Changes

Hello @binary-husky, I'm Gemini Code Assist¹! I'm currently reviewing this pull request and will post my feedback shortly. In the meantime, here's a summary to help you and other reviewers quickly get up to speed!

This pull request rebrands the project from 'AgentScope Tune' to 'AgentJet' and introduces a comprehensive refactoring and expansion of its core functionalities. The changes focus on enhancing the agent training framework by updating development dependencies, improving configuration management, and modularizing various components related to task reading, judging, and model tuning. The new structure supports multiple training backbones (VERL, Trinity, and a debug VLLM mode) and introduces advanced features for data generation, context tracking, and experimental interchange servers.

Highlights

Pre-commit Configuration Update: The .pre-commit-config.yaml file has been significantly updated to include a broader set of pre-commit hooks from pre-commit/pre-commit-hooks (e.g., trailing-whitespace, end-of-file-fixer, check-yaml, check-added-large-files, check-ast, check-json, check-merge-conflict, detect-private-key). The previous ruff and ruff-format hooks have been removed, and black formatter has been added. Several other hooks (isort, flake8, mypy) are commented out.
Project Renaming and Restructuring: The project name has been changed from 'AgentScope Tune' (ASTune) to 'AgentJet' (AJet) across the README.md and various internal file paths and module names. This involves renaming files and updating import paths to reflect the new branding and modular structure.
New Core Modules and Utilities: Several new Python files and directories have been added, introducing new functionalities and restructuring existing ones. Key additions include new backbone implementations for trinity, verl, and vllm modes, comprehensive context tracking modules, a programmatic training entry point (AgentJetJob), data generation and augmentation components, and a wide array of utility functions for configuration, debugging, and metric reporting.
Gitignore Expansion: The .gitignore file has been expanded to include new entries such as good_trinity_checkpoints, _resources, auto/auto_grader.json, tutorial/example_math_agent/math_agent_debug.yaml, saved_experiments, tests/temp, vsdb.py, appworld_pack_v2.tar*, saved_checkpoints, data, datasets, tutorial2, site, and dump.rdb.

🧠 New Feature in Public Preview: You can now enable Memory to help Gemini Code Assist learn from your team's feedback. This makes future code reviews more consistent and personalized to your project's style. Click here to enable Memory in your admin console.

Ignored Files

Ignored by pattern: .github/workflows/** (5)
- .github/workflows/doc.yaml
- .github/workflows/docker.yaml
- .github/workflows/docker/docker-compose.yaml
- .github/workflows/pre-commit.yaml
- .github/workflows/unittest.yaml

Using Gemini Code Assist

The full guide for Gemini Code Assist can be found on our documentation page, here are some quick tips.

Invoking Gemini

You can request assistance from Gemini at any point by creating a comment using either /gemini <command> or @gemini-code-assist <command>. Below is a summary of the supported commands on the current page.

Feature	Command	Description
Code Review	`/gemini review`	Performs a code review for the current pull request in its current state.
Pull Request Summary	`/gemini summary`	Provides a summary of the current pull request in its current state.
Comment	@gemini-code-assist	Responds in comments when explicitly tagged, both in pull request comments and review comments.
Help	`/gemini help`	Displays a list of available commands.

Customization

To customize Gemini Code Assist for GitHub experience, repository maintainers can create a configuration file and/or provide a custom code review style guide (such as PEP-8 for Python) by creating and adding files to a .gemini/ folder in the base of the repository. Detailed instructions can be found here.

Limitations & Feedback

Gemini Code Assist may make mistakes. Please leave feedback on any instances where its feedback is incorrect or counter productive. You can react with 👍 and 👎 on @gemini-code-assist comments. If you're interested in giving your feedback about your experience with Gemini Code Assist for Github and other Google products, sign up here.

You can also get AI-powered code generation, chat, as well as code reviews directly in the IDE at no cost with the Gemini Code Assist IDE Extension.

Review the Privacy Notices, Generative AI Prohibited Use Policy, Terms of Service, and learn how to configure Gemini Code Assist in GitHub here. Gemini can make mistakes, so double check it and use code with caution. ↩

gemini-code-assist

Code Review

This pull request introduces a major refactoring, renaming the project from "ASTune" to "AgentJet" and restructuring it into a proper Python package. The changes significantly improve the project's organization, clarity, and user-friendliness, especially with the updated README and the introduction of a programmatic API. My review focuses on a few key areas: the pre-commit configuration, a placeholder in the documentation, and potential issues in the rollout and data processing logic that could impact functionality and user experience.

gemini-code-assist · 2026-01-16T09:34:21Z

.pre-commit-config.yaml

+    hooks:
+    - id: black
+      language_version: python3.10
+      args: [--line-length=999999]


Setting the line length to an extremely high value like 999999 effectively disables line length checking. This can lead to code that is difficult to read and review. It's a standard practice to enforce a reasonable line length to ensure code consistency and maintainability. I recommend setting this to a more conventional value, such as 120.

args: [--line-length=120]

gemini-code-assist · 2026-01-16T09:34:21Z

ajet/backbone/main_verl.py

-                if config.astune.rollout.mode == "async"
-                else ActorRolloutRefWorker
-            )
+            actor_rollout_cls = AsyncActorRolloutRefWorker


The logic for selecting actor_rollout_cls has been simplified to always use AsyncActorRolloutRefWorker. The previous implementation allowed for a synchronous ActorRolloutRefWorker based on config.astune.rollout.mode. If synchronous rollouts are still intended to be supported, this change represents a regression in functionality. Please clarify if this change was intentional.

gemini-code-assist · 2026-01-16T09:34:21Z

ajet/schema/trajectory.py

        if len(self.prompt_ids) > self.max_prompt_len:
            truncate_any = True
-            print(f"-------------------------------------------------------------------------------------------------------")
-            print(f"Warning: prompt_ids length {len(self.prompt_ids)} exceeds max_prompt_len {self.max_prompt_len}, truncating.")
-            print(f"-------------------------------------------------------------------------------------------------------")
-            raise RuntimeError("Prompt length exceeds maximum allowed length. Please adjust the input data.")
-            self.prompt_ids = self.prompt_ids[-self.max_prompt_len:]
-            self.prompt_attention_mask = self.prompt_attention_mask[-self.max_prompt_len:]
-            self.prompt_position_ids = self.prompt_position_ids[-self.max_prompt_len:]
-            self.prompt_loss_mask = self.prompt_loss_mask[-self.max_prompt_len:]
-            self.prompt_logprobs = self.prompt_logprobs[-self.max_prompt_len:]
-
+            raise RuntimeError(f"Warning: prompt_ids length {len(self.prompt_ids)} exceeds max_prompt_len {self.max_prompt_len}, truncating.")


The code raises a RuntimeError if the prompt length exceeds max_prompt_len, but the error message misleadingly says "truncating." A hard crash can provide a poor user experience. Consider either implementing truncation as the commented-out code in the previous version suggested, or at least improving the error message to guide the user on how to resolve the issue (e.g., by suggesting they adjust max_prompt_length in their config or shorten their input data).

gemini-code-assist · 2026-01-16T09:34:21Z

README.md

+  author = {The AgentJet Team},
+  url    = {https://modelscope.github.io/AgentJet/},
+  month  = {01},
+  year   = {2026}


The citation year is set to 2026, which is in the future. This appears to be a placeholder and should be updated to the correct year to avoid confusion in academic or professional contexts.

Suggested change

year = {2026}

year = {2024}

binary-husky and others added 30 commits November 28, 2025 10:39

feat: Resolve conflict, auto committed by CodeFlow

e396ba8

Former-commit-id: 84192c2dec51311450d1653e97979a50f39354cf

Merge branch review_pxc into main

3df121b

Title: Fix review suggestions https://aliyuque.antfin.com/bayotg/vpbyh0/pguna8cwykkggv0h Link: https://code.alibaba-inc.com/DAIL-LLM/astune/codereview/24622110 Former-commit-id: 38a4ab61dedcf8a0dd8214e25f85d7df60d89998

remove trash and remove chinese

72dc650

Former-commit-id: d20ebec369ce1abbf40b9684bf412b50599c1937

Merge branch remove_trash_and_chinese into main

6dc5e1c

Title: remove trash and remove chinese Link: https://code.alibaba-inc.com/DAIL-LLM/astune/codereview/24641938 Former-commit-id: c4c183ad7d47c21dc969f151a86edc859bde6641

skip gpu check in debug mode

91f28b8

improving configuration sys

67f8230

renaming classes

5df0f74

Document agent performance and behavior in werewolf game

236752a

Added detailed observations on agent behavior and performance in werewolf simulations, including win rates and strategies.

Enhance README with image and tagline

6c749e0

Added a centered image and a tagline to the README.

Add section on expanding Qwen2-7B to Qwen2-14B (#1)

39eafc1

Expand Werewolves documentation with multi-agent setup (#2)

ef0320b

Added detailed explanation of the Werewolves role-playing game as a multi-agent environment, including terms, workflow setup, configuration, debugging, and training instructions.

Align math & add swanlab trinity compat (#6)

8e5b7af

* rpg werewolves yaml update * werewolf expected_steps=3 (there are 3 werewolves, meaning usually 3 samples are generated per run) * for no-eos message, pad eos but loss_mask=0 * align math * add swanlab

add additional dataset shuffle in task reader router

958cfc9

add deduplication filter (#8)

dd2bd2c

* feat: add deduplication filter * fix: spacing in deduplication filter db_path configuration * refactor: remove unused chroma_id variable in embedding client

merge TaskReaderRouter and TaskReaderRouterV2 into one

299968a

re-align trinity parameters

e2e87bb

fix test expectation

d9755a2

improve dynamic import

daea110

update docs (#9)

748da76

* docs: update docs * docs: fix images, formats, typos * docs: add hyperlinks to documentation sections in quickstart

fix docs

bc59397

fix pre-commit

76fcdd8

setup unittest auto benchmarking

d112987

commit bad math agent

2448142

benchmark successful

7cd2779

connect test bot to test data server

c7cf55c

patch native rollout

31adeee

simplify tool call logging

21f6eee

enforce ray kill in autokill

91c32ff

Add test bot server (#12)

0031eff

* commit bad math agent * benchmark successful * connect test bot to test data server * patch native rollout * simplify tool call logging * enforce ray kill in autokill

binary-husky and others added 23 commits January 13, 2026 17:19

amend config gap

a7dc96d

update benchmark test robot

69ca4d8

Merge pull request #1 from modelscope/dev/shuchang

a23a867

feat: Add comprehensive metric tracking and trajectory persistence

remove user_workflow reference and increase robustness

9e62b88

refactor(gc): use a lock for garbage collection in BaseAgentRunner

c8efc5e

redis

855cb2f

maybe stable

fc0a680

fix ctrl + c exit problem

1da86e2

optimize parallel performance with zmq

2c05d11

optimize further

8eae43c

update yaml configuration

75a4321

fix interchange server boot

666e8d3

httpx aclose patch

d0b6cae

keyurl multinode training support

0782173

fix trinity baseurl support

afeb5ed

refactor: update documentation deployment process and remove obsolete…

d02bf54

… scripts

fix mkdocs auto build

394b278

refactor: streamline MkDocs deployment workflow and update Python ver…

1ff504d

…sion

fix: correct site_url in mkdocs configuration

6906d3a

fix: update documentation links to point to the new modelscope URLs

c7f15db

docs: add citation section to README for research reference

0142ccf

fix invalid urls

0b13540

remove trailing-whitespace

9d0c77e

update doc: platform compare

0dfbc58

binary-husky force-pushed the pre-commit-fix branch from ae3ca83 to 0dfbc58 Compare January 16, 2026 09:29

stage

576bc48

gemini-code-assist bot reviewed Jan 16, 2026

View reviewed changes

binary-husky force-pushed the main branch 2 times, most recently from cf6110e to 1fc0663 Compare January 16, 2026 09:36

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Pre-commit fix #3

Pre-commit fix #3

Uh oh!

binary-husky commented Jan 16, 2026

Uh oh!

gemini-code-assist bot commented Jan 16, 2026

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

gemini-code-assist bot Jan 16, 2026

Uh oh!

gemini-code-assist bot Jan 16, 2026

Uh oh!

gemini-code-assist bot Jan 16, 2026

Uh oh!

gemini-code-assist bot Jan 16, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

9 participants

Pre-commit fix #3

Are you sure you want to change the base?

Pre-commit fix #3

Uh oh!

Conversation

binary-husky commented Jan 16, 2026

Uh oh!

gemini-code-assist bot commented Jan 16, 2026

Summary of Changes

Highlights

Footnotes

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

gemini-code-assist bot Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

gemini-code-assist bot Jan 16, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

9 participants