Deep Finance Update with New Judge (#7)

* feat(finworld): Added AgentScope learning protocol and OpenJudge evaluation functionality to the FinWorld task.

- Added the ExampleAgentScopeLearnProtocol class to implement the AgentScope execution flow for multi-turn interactions.

- Integrated semaphore control to manage the parallelism of environment calls, improving environment stepping performance.

- Implemented a mechanism for detecting context overflows and quickly terminating during environment interactions to prevent blocking.

- Added a finworld.yaml configuration file to define project training and rollout parameters.

- Added the FinWorldJudgeByOpenJudge class, integrating multiple evaluators including RM Gallery and OpenJudge (@Haoran).

- Implemented a mechanism for converting task output, asynchronous calls, and retrying to ensure evaluation stability.

- Weight normalization manages the contributions of each evaluator, merging them to calculate the final reward and success determination.

* Precommit fix (#4)

* fix end of files

* autoflake import fix

* add mypy check

* fix test bench import

* refactor(finworld): Replace agent protocol and unify configuration updates

- Renamed ExampleAgentScopeLearnProtocol to ExampleDeepResearchProtocol and modified the execute method signature.
- Unified the parameter name of the model tuner to `tuner` and its related attribute references.
- Optimized the multi-turn interaction step configuration, changing it to use `tuner.config.ajet.rollout.multi_turn.max_steps`.
- Modified the context overflow judgment logic to prevent tool call blocking.
- Updated the finworld.yaml configuration, replacing astune with ajet-related configurations, and adjusted the workflow protocol and environment parameters.
- Modified the default environment variable values ​​and log saving paths in finworld_judge.py.
- Added and improved multi-machine and single-machine startup scripts, supporting dynamic generation of MCP configuration and environment variable loading.
- Added the finworld_single.yaml template to adapt to single-machine training configurations.
- Adjusted the key reference for multi-turn step configuration in ma_deepresearch.py, using the ajet configuration path.

* feat(finworld): Added FinWorld training environment configuration scripts and templates

- Added bash startup scripts for multi-machine, multi-GPU training, supporting dynamic configuration generation and environment variable import.
- Implemented training configuration file templates, supporting automatic injection of various weight parameters and model paths.
- Adjusted the default request timeout of EnvClient from 30 seconds to 300 seconds to accommodate long training requests.
- Added a new finworld example directory and related documentation, improving the example project structure.

* refactor(utils): Remove unused extract and compute functions `extract_tool_stats_from_cmts`

* refactor(finworld): Replace the old model with OpenJudge, update evaluation configuration and scripts

- Replaced model initialization in FinWorldJudgeByOpenJudge with the `_init_openjudge_model` method
- Read Judge model parameters from the configuration file first, using environment variables as a fallback
- Optimized RM Gallery initialization, using configuration-first logic, and improved exception stack trace printing
- Cleaned up and removed the old `_init_model` singleton method and related code
- Updated the example startup script `ajet_finworld.sh`, adding OPENJUDGE_LLM and RM_LLM configurations
- Modified YAML templates and configuration files to unify the structure and field naming of Judge configuration items
- Deleted the outdated `cc_rm4_res2cit2fai2_30b.sh` script
- Adjusted the `env_service` startup path to improve environment activation compatibility
- Adjusted script log output format and content to enhance the clarity of configuration parameter printing

* feat(task_reader): Support data reading of type jsonl_with_env_service

- Added the jsonl_with_env_service type, which allows loading data from jsonl files while calling tools via env_service.
- Extended ResourceKeeper to handle the creation and release logic of environment instances for jsonl_with_env_service.
- Maintained the env_service type logic, calling create_instance to register instances and initializing them using init_messages from the jsonl file.
- Added an example protocol, ExampleDeepResearchProtocol, to implement multi-turn interaction and environment call coordination.
- Provided training scripts and YAML configuration templates for finworld, supporting the jsonl_with_env_service mode training environment.
- Optimized scripts to support multi-node multi-GPU training, including environment variables and Ray cluster configuration.

* feat(core): add finworld task reader support to framework

* feat(finworld): implement specialized data reader and openjudge-based grading logic

* refactor(finworld): optimize configuration templates and prompt engineering

* chore(finworld): update launch scripts and add variant experiment scripts

* feat(finworld): Added support for multi-machine, multi-GPU training scripts and configuration templates:

* chore(git): ignore finworld/yaml/*

* fix(metrics): Fix and enhance the compatibility and debugging output of the metrics update logic

- Modified the `update_metrics` function, adding a `prefix` parameter to distinguish between training and validation metrics.
- Adjusted the data source for extracting `reward_stats` and `tool_stats`, migrating from `workflow_metadata` to `log_metrics`.
- Added debug printing to output the `log_metrics` content and metric key names at key steps for easier troubleshooting.
- Used the appropriate prefix when calling `update_metrics` in `trainer_verl.py`, and added multiple debug prints.
- Modified `WorkflowOutput` to place `tool_stats` and `reward_stats` into the `log_metrics` field.
- Removed redundant and deprecated code for extracting `reward_stats` and calculation functions.
- Added debug information output to the `finworld` and `finworld_judge` modules to track log metrics and scoring data.

* fix(metrics): Remove debug prints and synchronize reward statistics

- Removed debug print statements before and after the `update_metrics` call in `trainer_verl.py`
- Removed debug print statements related to the `log_metrics` key in `finworld.py`
- Removed debug print statements before updating `metadata_stats` in `finworld_judge.py`
- Added logic in `general_runner.py` to synchronize `reward_stats` from `metadata` to `log_metrics` after the judge calculation
- Cleaned up debug print statements within `update_metrics` in `metric_helper`, improving code readability.

* chore: "Stop tracking existing yaml files in tutorial directory"

* fix(task_runner): Synchronize reward_stats to log_metrics

feat(tutorial): Added FinWorld multi-machine multi-GPU training startup script

* refactor(script): Refactored the finworld training script, integrating configuration and startup processes.

* Refactor(deep_finance): Replace and remove finworld-related implementations

- Switched the example directory from example_finworld to example_deep_finance
- Modified startup parameters and logic to support deep_finance, replacing the finworld option
- Replaced finworld_reader with deep_finance_reader in the task reader
- Adjusted environment client configuration in resource management, using deep_finance instead of finworld-related checks
- Updated reward metric tool documentation to support deep_finance
- Deleted finworld-related configuration files, scripts, code, and evaluation modules, cleaning up leftover files and scripts
- Replaced the keyword "finworld" with "deep_finance" in comments and logs

* refactor(deepfinance): Rename and unify DeepFinance module and config references

- Replace all "finworld" and "deep_finance" names with the unified "deepfinance" format.
- Modify command-line arguments to `--with-deepfinance` for consistency.
- Adjust the class name in `task_reader` from `deep_financeReader` to `DeepFinanceReader`.
- Update the documentation description and file name of the `metric_helper` module to DeepFinance.
- Modify environment variables and configuration paths in the example script `deep_finance.sh` to use the `DEEPFINANCE` prefix.
- Update `judge_protocol` to `DeepFinanceJudgeByOpenJudge` in the `deep_finance.yaml` configuration.
- Refactor the `FinWorldJudgeByOpenJudge` class in `deep_finance_judge.py` to `DeepFinanceJudgeByOpenJudge`.
- Rename the `FinworldReader` class in `deep_finance_reader.py` to `DeepFinanceReader`.
- Modify the debug log identifier and corresponding environment variable name to `DEEPFINANCE_DEBUG`.
- Update the evaluation protocol in the `deep_finance_template.yaml` template to `DeepFinanceJudgeByOpenJudge`.
- Ensure that internal references and comments in all modules are updated to use DeepFinance and deepfinance-related names.

* refactor(tutorial): Optimize dynamic generation logic for configuration file paths

* fix(deep_finance): argparse: with-deepfinance

* fix(tutorial): Fixed issues with multi-machine training environment variable settings

* fix(env): Corrected the assignment logic for reward and info when returning environment state

- Corrected the `env_output` return value structure in `BaseGymEnv` to ensure correct assignment of `reward` and `info` fields.
- Removed `RefJudge` and `StructureJudge` related metric calculations and statistics from `reward_metric_helper`.
- Cleaned up redundant code in `reward_metric_helper`, removing invalid comments and statistical items.
- Modified `save_trajectory_as_json` to always print trajectory saving confirmation information.
- Corrected log comments in `example_deep_finance` to avoid meaningless log output.
- Added the `save_trajectory_as_json_file` configuration item to `deep_finance_template.yaml` to support trajectory saving functionality.

* chore(config): Update example_deep_finance configuration and clean up files

- Added a new ignore rule for config file paths in .gitignore
- Deleted the automatically generated mcp_finance_tool_generated.json file in example_deep_finance
- Refactored the deep_finance.yaml configuration file, adjusting project and experiment names
- Reorganized Judge configuration, clarifying openjudge_llm and rm_llm models
- Optimized model paths and training parameter configurations, adding parallel and batch processing settings
- Adjusted data reading methods and training/validation set path placeholders
- Reduced GPU memory usage ratio for rollout to 0.8
- Updated the default save directory path for the trainer to a placeholder variable
- Cleaned up unused and commented-out code to improve configuration file conciseness

* Refactor(metric): Optimize tool metric calculation and data saving logic

- Corrected the data source field for timeline data used during trajectory saving.
- Removed redundant fields in tool execution time, cache hit rate, and error rate statistics.
- Updated .gitignore to add ignore rules for the example script directory.
- Removed unnecessary debugging information from logs to reduce log noise.
- Adjusted log printing in the multi-round interaction execution process to simplify output content.
- Streamlined log code for environment observation and termination checks to improve code readability.

* fix(metric_helper): fix tool cache metric

* fix little bug

* fix(utils): Suppress httpx AsyncClient.aclose() exception warnings

* comments to english

* feat: 支持服务名称前缀功能

- 在 launcher 中添加 --prefix 参数支持
- 在 pty_launch 函数中实现前缀逻辑
- 更新 deep_finance.sh 脚本以使用前缀功能
- 允许在同一环境中运行多个服务实例

* fix: 改进 MultiAgent 消息内容解析逻辑

- 支持 tool_result 格式的消息内容块
- 改进非文本内容的处理逻辑，继续处理其他项而非跳过整个消息
- 添加 tool_use 类型的处理（跳过，因为已通过 tool_calls 字段处理）
- 优化代码结构和注释，提高可读性

* fix: 优化 DeepFinance 判断逻辑和配置

- 修复 tool_stats 提取逻辑，从 log_metrics 中正确获取数据
- 添加惩罚项调试信息输出
- 启用 tool calls 功能（force_disable_toolcalls: False）
- 确保奖励计算准确性

* chore(deps): bump agentscope from 1.0.7 to 1.0.8

* fix(metric_helper): correct trajectory save path and add tool call metric

- Change trajectory save directory from "ctx_trackers" to "trajectory" to organize files better
- Add recording of tool call counts alongside error rates in tool metrics
- Update experiment suffix in deep finance example script for clearer naming convention

* revise message parsing

* fix(metric_helper): update openjudge graders list in reward metric helper

* feat(deep_finance): replace OpenJudge graders with PresentationQualityGrader

- Remove legacy graders and integrate PresentationQualityGrader and GroundingGrader
- Update grader weights and disable unused graders in config and code
- Simplify grader configuration creation with new mappers for report content and traj
- Refactor DeepFinanceJudgeByOpenJudge to support new grading scheme
- Add PresentationQualityGrader implementation with strict JSON output format
- Include utilities for JSON parsing and validation in presentation quality grader
- Add prompt templates for presentation quality grading criteria and instructions
- Provide example script to run PresentationQualityGrader with OpenAIChatModel
- Add traj_adapter utilities to normalize and extract user query and final report
- Update YAML template to replace old grader weights with presentation quality weight
- Create init files to expose PresentationQualityGrader in judge package

* feat(grounding): implement grounding grader for citation compliance evaluation

- add GroundingGrader class to evaluate citation coverage and truthfulness based on dialogue traj
- provide default OpenAIChatModel creation with deterministic options
- implement prompt construction and JSON parsing utilities for model interaction
- calculate scores including coverage, grounding, and invalid citation penalties
- add detailed json_utils module for strict JSON extraction and validation
- introduce prompt templates defining citation auditing rules and user prompts
- supply reference.py with related grounding evaluation logic and RefJudgeEvaluator class
- create __init__.py to expose GroundingGrader module
- add presentation_quality module __init__.py with PresentationQualityGrader export

* fix(deep_finance_judge): add debug logging for OpenJudge evaluation process

* feat(deep_finance): enhance reward metadata and zero score debugging

- Add populate_reward_metadata_from_stats to copy reward stats into reward metadata
- Populate reward metadata in GeneralRunner if reward_stats present in workflow output
- Refine compute_reward_metrics with updated OpenJudge graders: presentation_quality, grounding, planning
- Add _save_zero_score_debug method in DeepFinanceJudgeByOpenJudge to save debug info for zero grader scores
- Remove deprecated RewardStats usage in deep_finance_judge
- Update judge __init__ to export GroundingGrader alongside PresentationQualityGrader
- Clean up debug print statements and logging in deep_finance_judge.py
- Update .gitignore to exclude prepare_data and judge/analytical_sufficiency folders in example_deep_finance tutorial

* feat(presentation_quality): upgrade grading to 1/3/5 scoring system with markdown cleanup

- Add function to strip markdown code block fences in grounding and presentation_quality modules
- Change presentation quality grader to score each of 8 criteria on a 1/3/5 scale instead of pass/fail
- Normalize total score by dividing sum of item scores by max (40), improving granularity
- Update reasoning output to list lowest scoring items with notes for focused feedback
- Revise presentation quality prompt to reflect new 1/3/5 scoring rubric with detailed instructions
- Adjust JSON output schema accordingly, replacing boolean pass with numeric score fields
- Add get_score utility in JSON utils to extract and validate scores from graded items
- Clean report input by removing markdown fences before grading to avoid markup noise
- Add grounding weight configuration in YAML template for improved modular judge weighting

* chore(config): update experiment suffix, prefix and reward weights in deep_finance.sh

* fix(deep_finance): update environment variables and training launch options

* chore(config): parameterize deep finance training configuration

* chore(config): update experiment suffix, prefix, and weight parameters

* fix(example_deep_finance): update dynamic config file generation path

* refactor(judge): remove deprecated presentation quality script

---------

Co-authored-by: binary-husky <[email protected]>
Co-authored-by: Qingxu Fu <[email protected]>
Co-authored-by: qingxu.fu <[email protected]>

Author: TaoShuchang

.gitignore

@@ -158,6 +158,9 @@ tutorial/example_deep_finance/yaml/*
 tutorial/example_deep_finance/config/*
 tutorial/example_deep_finance/scripts/*
 flash_attn-2.8.*.whl
+tutorial/example_deep_finance/prepare_data/*
+tutorial/example_deep_finance/judge/analytical_sufficiency/*
+
 .dockerignore
 benchmark_datasets
 modelscope_cache

ajet/context_tracker/multiagent_tracking.py

@@ -82,6 +82,18 @@ def extract_text_content_from_content_dict(self, msg):
         #        },
         #    ],
         # }
+        # or tool_result format?? not observed yet:
+        # msg = {
+        #    "role": "tool",
+        #    "content": [
+        #        {
+        #           "type": "tool_result",
+        #           "id": "call_xxx",
+        #           "output": "tool output content",
+        #           "name": "tool_name"
+        #        },
+        #    ],
+        # }
 
 
         str_content = ""

ajet/task_runner/general_runner.py

@@ -9,6 +9,7 @@
 from ajet.schema.trajectory import Reward
 from ajet.task_runner.base_runner import BaseAgentRunner
 from ajet.utils.dynamic_import import dynamic_import
+from ajet.utils.metric_helper.reward_metric_helper import populate_reward_metadata_from_stats
 
 
 class GeneralRunner(BaseAgentRunner):
@@ -73,6 +74,10 @@ def execute(self, workflow_task: WorkflowTask) -> BaseContextTracker:
             madness=0,
             description="",
         )
+        
+        # Populate reward metadata with deep_finance reward stats if available
+        if "reward_stats" in workflow_output.metadata:
+            populate_reward_metadata_from_stats(reward, workflow_output.metadata["reward_stats"])
         context_tracker.process_reward(reward)
         # generate token before merging
         context_tracker.group_merge()

ajet/utils/metric_helper/reward_metric_helper.py

@@ -11,9 +11,12 @@
 - judge_time/                 Judge time consumption statistics
 """
 
-from typing import List, Dict, Any
+from typing import List, Dict, Any, TYPE_CHECKING
 import numpy as np
 
+if TYPE_CHECKING:
+    from ajet.schema.trajectory import Reward
+
 
 def extract_reward_stats_from_trajectories(trajectories: List[Any]) -> List[Dict[str, Any]]:
     """
@@ -72,22 +75,15 @@ def compute_reward_metrics(reward_stats_list: List[Dict[str, Any]], prefix: str
     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 Metrics (PresentationQualityGrader, GroundingGrader) ==========
     openjudge_enabled_count = sum(1 for rs in reward_stats_list if rs.get('openjudge_enabled', False))
 
     if openjudge_enabled_count > 0:
-        # ========== OpenJudge Metrics ==========
-
-        # Dynamically extract OpenJudge grader fields
-        # Currently supported graders: report_resolution, trajectory_faithfulness,
-        # rubrics_performance, trajectory_comprehensive, information_gain, action_loop
+        # OpenJudge graders: presentation_quality, grounding
         openjudge_graders = [
-            "report_resolution",
-            "trajectory_faithfulness",
-            "rubrics_performance",
-            "trajectory_comprehensive",
-            "information_gain",
-            "action_loop",
+            "presentation_quality",
+            "grounding",
+            "planning"
         ]
 
         for grader_name in openjudge_graders:
@@ -151,3 +147,18 @@ def compute_reward_metrics_from_trajectories(trajectories: List[Any], prefix: st
     reward_stats_list = extract_reward_stats_from_trajectories(trajectories)
     return compute_reward_metrics(reward_stats_list, prefix=prefix)
 
+
+def populate_reward_metadata_from_stats(reward: "Reward", reward_stats: Dict[str, Any]) -> None:
+    """
+    Populate Reward.metadata with all reward statistics.
+    
+    Args:
+        reward: The Reward object to populate
+        reward_stats: The reward_stats dictionary from judge
+    """
+    if not reward_stats:
+        return
+    
+    # Directly copy all reward_stats into metadata
+    reward.metadata.update(reward_stats)
+

tutorial/example_deep_finance/__init__.py

@@ -0,0 +1 @@
+# tutorial/example_deep_finance package

tutorial/example_deep_finance/deep_finance.sh

@@ -1,29 +1,29 @@
 #!/bin/bash
-set -e
+set -e  
 #===============================================================================
 # 1. 配置区域 - 用户只需修改这里
 #===============================================================================
-SUFFIX="deep_finance"     # 实验后缀，影响所有日志和实验名称
-PREFIX="open"                        # 实验前缀，影响日志和实验所在文件夹
+SUFFIX="newjudge"     # 实验后缀，影响所有日志和实验名称
+PREFIX="ajet_newjudge"                        # 实验前缀，影响日志和实验所在文件夹
 
 # OpenJudge 模型配置
 OPENJUDGE_LLM='qwen-flash'        # OpenJudge 评分模型
 RM_LLM='qwen-max'                 # RM Gallery 评分模型
 JUDGE_CONCURRENCY=10
 
 # 奖励权重配置
-RM_WEIGHT=0.4
-CITATION_AUDIT_WEIGHT=0.2
-REPORT_RESOLUTION_WEIGHT=0.2
-TRAJECTORY_FAITHFULNESS_WEIGHT=0.2
+RM_WEIGHT=0.5
+PRESENTATION_QUALITY_WEIGHT=0.25
+GROUNDING_WEIGHT=0.25
 
 # 训练参数配置
 NUM_REPEAT=4        # group size，每个query rollout NUM_REPEAT次
 TRAIN_BATCH_SIZE=32  # 训练batchsize
 NUM_STEPS=6         # 每个样本step轮数
 DEEPFINANCE_TOOL_RESULT_MAX_CHARS=10000
 
-# 主目录
+# 主目录（需要更改）
+export AJET_ROOT="/mnt/data_cpfs/taoshuchang.tsc/deepresearch/AgentJet_new"
 
 NNODES=${WORLD_SIZE}
 
@@ -46,7 +46,7 @@ fi
 # 2. 动态生成配置文件 (从yaml template生成yaml)
 #===============================================================================
 # 修改：配置文件生成路径，现在动态生成到 yaml 目录下
-CONFIG_TEMPLATE="tutorial/example_deep_finance/yaml_template/deep_finance_template.yaml"
+CONFIG_TEMPLATE="tutorial/example_deep_finance/deep_finance.yaml"
 CONFIG_FILE="${AJET_ROOT}/tutorial/example_deep_finance/yaml/${SUFFIX}.yaml"
 mkdir -p $(dirname ${CONFIG_FILE})
 
@@ -55,12 +55,11 @@ sed -e "s|{{SUFFIX}}|${SUFFIX}|g" \
     -e "s|{{MODEL_PATH}}|${MODEL_PATH}|g" \
     -e "s|{{NNODES}}|${NNODES}|g" \
     -e "s|{{RM_WEIGHT}}|${RM_WEIGHT}|g" \
-    -e "s|{{CITATION_AUDIT_WEIGHT}}|${CITATION_AUDIT_WEIGHT}|g" \
+    -e "s|{{PRESENTATION_QUALITY_WEIGHT}}|${PRESENTATION_QUALITY_WEIGHT}|g" \
+    -e "s|{{GROUNDING_WEIGHT}}|${GROUNDING_WEIGHT}|g" \
     -e "s|{{OPENJUDGE_LLM}}|${OPENJUDGE_LLM}|g" \
     -e "s|{{RM_LLM}}|${RM_LLM}|g" \
     -e "s|{{JUDGE_CONCURRENCY}}|${JUDGE_CONCURRENCY}|g" \
-    -e "s|{{REPORT_RESOLUTION_WEIGHT}}|${REPORT_RESOLUTION_WEIGHT}|g" \
-    -e "s|{{TRAJECTORY_FAITHFULNESS_WEIGHT}}|${TRAJECTORY_FAITHFULNESS_WEIGHT}|g" \
     -e "s|{{NUM_REPEAT}}|${NUM_REPEAT}|g" \
     -e "s|{{NUM_STEPS}}|${NUM_STEPS}|g" \
     -e "s|{{TRAIN_BATCH_SIZE}}|${TRAIN_BATCH_SIZE}|g" \
@@ -72,7 +71,7 @@ sed -e "s|{{SUFFIX}}|${SUFFIX}|g" \
     ${AJET_ROOT}/${CONFIG_TEMPLATE} > ${CONFIG_FILE}
 
 echo "配置文件已生成: ${CONFIG_FILE}"
-echo "参数确认: RM=${RM_WEIGHT}, Citation=${CITATION_AUDIT_WEIGHT}, OpenJudge=${OPENJUDGE_LLM}, RM_LLM=${RM_LLM}"
+echo "参数确认: RM=${RM_WEIGHT}, PresentationQuality=${PRESENTATION_QUALITY_WEIGHT}, Grounding=${GROUNDING_WEIGHT}, OpenJudge=${OPENJUDGE_LLM}, RM_LLM=${RM_LLM}"
 
 #===============================================================================
 # 3. 环境配置
@@ -106,15 +105,15 @@ export DEEPFINANCE_MCP_CONFIG  DEEPFINANCE_TOOL_RESULT_MAX_CHARS
 # 其他服务配置
 HF_ENDPOINT="https://hf-mirror.com"
 ES_HOSTS="http://11.160.132.46:8200"
-export HF_ENDPOINT ES_HOSTS
+export HF_ENDPOINT ES_HOSTS 
 
 # log 文件位置
 CURRENT_TIME=$(date "+%Y%m%d_%H%M%S")
 LOG_DIR="${AJET_ROOT}/logs/${PREFIX}"
 MASTER_IP_FILE="${LOG_DIR}/master-ip_${SUFFIX}.log"
 ENV_SERVICE_LOG="${LOG_DIR}/env_service_${SUFFIX}_${CURRENT_TIME}.log"
 TRAIN_LOG="${LOG_DIR}/train_${SUFFIX}_${CURRENT_TIME}.log"
-
+env_log_prefix="${SUFFIX}__${CURRENT_TIME}"
 # 多机训练参数配置
 GPUS_PER_NODE=8
 EXPECTED_WORKERS=$WORLD_SIZE
@@ -156,6 +155,8 @@ export NCCL_ASYNC_ERROR_HANDLING=1
 
 export PYTHONPATH="${AJET_ROOT}:${PYTHONPATH}"
 export RAY_CLUSTER_MODE="multi_node"
+export DEEPFINANCE_PATH="${ENV_SERVICE_ROOT}" # AgentJet 内部可能使用此路径
+export DEEPFINANCE_SCRIPT="source /mnt/data/taoshuchang.tsc/anaconda3/etc/profile.d/conda.sh && conda activate finworld_1209  && cd ${ENV_SERVICE_ROOT} && DEEPFINANCE_TOOL_RESULT_MAX_CHARS=${DEEPFINANCE_TOOL_RESULT_MAX_CHARS} DEEPFINANCE_MCP_CONFIG=${DEEPFINANCE_MCP_CONFIG} CACHE_TYPE=${CACHE_TYPE} MONGO_URI=${MONGO_URI} MONGO_DB_NAME=${MONGO_DB_NAME} MONGO_COLLECTION_NAME=${MONGO_COLLECTION_NAME} python -m env_service.env_service --env finworld --portal 0.0.0.0 --port 8080"
 
 
 #===============================================================================
@@ -202,11 +203,12 @@ if [[ $HOSTNAME == *"-master-"* ]]; then
 
     # 启动训练任务（最核心）
     python ajet/launcher.py \
+        --with-deepfinance \
         --conf ${CONFIG_FILE} \
         --backbone="verl" \
-        --prefix=${SUFFIX} \
+        --prefix=${env_log_prefix} \
         2>&1 | tee ${TRAIN_LOG}
-
+    
 
 #===============================================================================
 # 6.2 Worker 节点启动流程
@@ -218,4 +220,4 @@ else
     ray stop || true
     ray start --address $MASTER_ADDR:6379 --num-gpus 8
     while true; do sleep 60; done
-fi
+fi

tutorial/example_deep_finance/deep_finance.yaml

@@ -1,27 +1,26 @@
 # ------------------ 主要配置 ------------------
 ajet:
-  project_name: ajet_deep_finance
-  experiment_name: "ajet_deep_finance"
+  project_name: "{{PREFIX}}"
+  experiment_name: "{{SUFFIX}}"
   # Judge 配置（嵌套结构，对应 self.config.ajet.judge.*）
   judge:
-    openjudge_llm: qwen-flash     # OpenJudge 模型
-    rm_llm: qwen-max                   # RM Gallery 模型
-    concurrency: 10   # Judge 并发数
+    openjudge_llm: {{OPENJUDGE_LLM}}     # OpenJudge 模型
+    rm_llm: {{RM_LLM}}                   # RM Gallery 模型
+    concurrency: {{JUDGE_CONCURRENCY}}   # Judge 并发数
     train_ref_ans_path: {{TRAIN_REF_ANS_PATH}}   # 训练集 Reference Answer 路径
     val_ref_ans_path: {{VAL_REF_ANS_PATH}}       # 验证集 Reference Answer 路径
   # OpenJudge 权重配置
-  report_resolution_weight: 0.2         # 报告质量评估
-  trajectory_faithfulness_weight: 0.2   # 事实准确性评估
-  citation_audit_weight: 0.2               # 引用审计评估 (覆盖率 + 真实性)
-  rm_weight: 0.4                                       # RM Gallery 权重
+  presentation_quality_weight: {{PRESENTATION_QUALITY_WEIGHT}}   # 报告呈现质量评估
+  grounding_weight: {{GROUNDING_WEIGHT}}                         # 引用规范性评估
+  rm_weight: {{RM_WEIGHT}}                                       # RM Gallery 权重
   task_judge:
     # 使用本地 DeepFinanceJudge 进行评估（解耦远程 env_service）
     judge_protocol: tutorial.example_deep_finance.deep_finance_judge->DeepFinanceJudgeByOpenJudge
   model:
     # ✨✨✨✨ 设置待训练的模型
     path: {{MODEL_PATH}}
   trainer_common:
-    nnodes: 8
+    nnodes: {{NNODES}}
     n_gpus_per_node: 8
     val_before_train: True
     val_pass_n: 8
@@ -32,44 +31,42 @@ ajet:
   rollout:
     # ✨✨✨✨ 编写并选择Agent
     user_workflow: tutorial.example_deep_finance.deep_finance->ExampleDeepResearchProtocol
-    force_disable_toolcalls: True
+    force_disable_toolcalls: False
     enable_oversample: False
     tensor_model_parallel_size: 8
-    num_repeat: 4
+    num_repeat: {{NUM_REPEAT}}
     max_env_worker: 64  # 增加环境并行数
     max_num_seqs: 64    # 增加VLLM并发序列数
     max_response_length_in_one_turn: 8000
     max_model_len: 50000
     agent_madness_reward: 0.0
     compute_madness_checklist: None
     multi_turn:
-      max_steps: 6
+      max_steps: {{NUM_STEPS}}
   interchange_server:
     interchange_method: 'tcp' # options: 'tcp' (multi-nodes) or  'ipc' (1 node)
   debug:
     debug_max_parallel: 1  # 增加并行任务数，充分利用GPU
     debug_first_n_tasks: 100  # 增加处理的任务数
   data:
-    train_batch_size: 32
+    train_batch_size: {{TRAIN_BATCH_SIZE}}
     max_prompt_length: 8000
     max_response_length: 41000
 
   task_reader:
     type: deep_finance  # 数据从 JSON 加载并组装 init_messages，工具调用走 env_service
     deep_finance:
       training:
-        file_path: {{TRAIN_PATH}}
+        file_path: {{TRAIN_DATA_PATH}}
       validation:
-        file_path: {{VAL_PATH}}
+        file_path: {{VAL_DATA_PATH}}
     # env_service 仍需配置（用于工具调用）
     env_service:
       env_type: "finworld"
       env_url: {{ENV_SERVICE_URL}}
       env_action_preference: code
-
-
 trainer:
-  default_local_dir: {{CKPT_SAVE_PATH}}
+  default_local_dir: "{{CKPT_SAVE_PATH}}/{{PREFIX}}/{{SUFFIX}}"
   # resume_mode: disable  # 禁用自动恢复，从头开始训练
 actor_rollout_ref:
   rollout: