mirror of
https://github.com/agentscope-ai/ReMe.git
synced 2026-08-28 05:25:04 +00:00
935e886af3
Some checks failed
Pre-commit / run (ubuntu-latest) (push) Has been cancelled
* feat(reme): 添加配置选项以启用或禁用个人资料功能 - 在 ReMe 初始化方法中添加 enable_profile 参数，默认值为 True - 根据 enable_profile 设置决定是否创建 profile 目录和设置 profile_dir - 在 PersonalSummarizer 中根据 enable_profile 条件性地添加个人资料相关工具 - 在 PersonalRetriever 中根据 enable_profile 条件性地添加 ReadAllProfiles 工具 - 修改 profile_path 属性以在禁用个人资料时返回 None - 修改 get_profile_handler 方法以在禁用个人资料时返回 None - 为 enable_profile 参数添加文档说明其用于云向量存储场景 * refactor(benchmark): 重构LongMemEval基准测试中的ReMe实例管理 - 移除未使用的shutil导入 - 将固定的ReMe实例改为每个问题创建独立实例以实现隔离 - 更新LLM配置名称从qwen3-max-think到qwen-max-t - 修改模型调用逻辑使用正确的model_name参数 - 添加qwen-flash和GPT-4o-mini等新模型配置 - 统一使用"User"作为用户名，通过集合名实现隔离 - 调整并发处理数从4降至1，批处理大小从10增至30 - 每个问题类型采样数从2增至4 - 添加异步上下文管理确保资源正确释放 * reformat 2 files * refactor(benchmark): 重构长记忆评估中的模型配置 - 将原有的 eval_model_name 替换为专门的 retrieve_model_name 用于检索操作 - 添加对 qwen-max 模型配置的支持 - 更新参数解析器以支持新的检索模型参数 - 修改最大并发数默认值从 1 提升到 4 - 调整样本数量默认值从 4 减少到 1 - 统一模型参数命名规范，区分摘要、检索和评估模型 - 优化内存处理器初始化逻辑，支持独立的检索模型配置 * fix(benchmark): 移除数据路径默认值并设为必填参数 - 将LongMemEval评估脚本中的data_path参数改为必需参数 - 将HaluMem评估脚本中的data_path参数改为必需参数 - 删除了硬编码的默认文件路径配置 - 强制用户显式指定数据集文件路径以避免路径错误 * Update __init__.py * Update __init__.py * fix(benchmark): 修复ReMe评估中的模型配置和空值处理问题 - 移除了retrieve_memory调用中不需要的llm_config_name参数 - 修复了长字符串打印的换行格式问题 - 添加了eval_result为空时的初始化处理 - 在accuracy评估中加入了eval_model_name参数传递 * style(benchmark): 格式化模型名称打印输出 - 移除了多行字符串中的换行符和多余空格 - 将模型名称信息合并为单行连续显示 - 保持了原有的打印格式和信息完整性 * docs(readme): 更新文档添加实验结果表格 - 在英文版 README 中添加 🧪 Experiments 章节 - 添加 LoCoMo 和 HaluMem 两个基准测试的结果表格 - 在中文版 README_ZH 中添加 🧪 实验 章节 - 添加 LoCoMo 和 HaluMem 测试集的实验配置说明 - 添加完整的实验数据对比表格和评估协议说明 * docs(readme): 更新文档中的内存系统链接 - 为基于文件的记忆系统添加锚点链接 - 为基于向量库的记忆系统添加锚点链接 - 修复英文文档中的链接格式 - 修复中文文档中的链接格式和空行问题 * docs(readme): update experimental results section in documentation - Remove outdated experimental data placeholder "Coming soon..." - Add complete evaluation results for LoCoMo and HaluMem benchmarks - Include detailed performance metrics tables for all memory methods - Update experimental settings description with ReMe backbone details - Align evaluation protocol information with LLM-as-a-Judge approach - Maintain consistent formatting between English and Chinese documentation * docs(benchmark): add quick start guides for halumem and longmemeval experiments - Created HaluMem experiment quick start guide with ReMe integration setup - Added detailed steps for installing ReMe environment using conda - Included repository cloning instructions for HaluMem benchmark - Provided complete command examples for running HaluMem experiments - Created LongMeMEval quick start guide with data download procedures - Added wget commands for downloading cleaned dataset files - Included evaluation script instructions for computing experiment statistics - Documented parameter configurations for different model types and batch sizes * docs(longmemeval): update quickstart guide documentation - Changed project name from Halumem to Longmemeval in title - Updated description to reference Longmemeval experiments instead of Halumem - Maintained existing ReMe integration instructions unchanged
1.5 KiB
1.5 KiB
Longmemeval
Experiment Quick Start Guide This guide helps you quickly set up and run Longmemeval experiments with ReMe integration.
1. Start ReMe Service
Install ReMe (if not already installed) If you haven't installed the ReMe environment yet, follow these steps:
# Create ReMe environment
conda create -p ./reme-env python==3.12
conda activate ./reme-env
# Install ReMe
pip install .
2. Clone the Repository
cd ./benchmark/longmemeval
mkdir -p data/
cd data/
wget https://huggingface.co/datasets/xiaowu0162/longmemeval-cleaned/resolve/main/longmemeval_oracle.json
wget https://huggingface.co/datasets/xiaowu0162/longmemeval-cleaned/resolve/main/longmemeval_s_cleaned.json
wget https://huggingface.co/datasets/xiaowu0162/longmemeval-cleaned/resolve/main/longmemeval_m_cleaned.json
cd ..
3. Run Experiments
Launch the ReMe service to enable memory library functionality:
clear && python benchmark/longmemeval/eval_longmemeval_reme.py \
--data_path benchmark/longmemeval/data/longmemeval_s_cleaned.json \
--reme_model_name qwen-flash \
--reme_model_name retrieve_model_name \
--eval_model_name gpt-4o-mini-2024-07-18 \
--batch_size 20 \
--algo_version default
4. Evaluate Results
Evaluate the results of the experiments:
python benchmark/longmememeval/compute_stats.py \
--results_dir bench_results/longmemeval_reme \
--output_file bench_results/longmemeval_reme/statistics.json
The compute_stats.py script computes various statistics from the evaluation results.