ViperEkura
|
2b3bf442e9
|
feat: add SFT process_batch for parallel tokenization + short QA filter script
|
2026-07-29 22:01:09 +08:00 |
|
ViperEkura
|
545104ba70
|
feat: add Alpaca-GPT4-ZH SFT export script
|
2026-07-29 21:23:20 +08:00 |
|
ViperEkura
|
e999629098
|
移除缓存脚本中的 data_files 和 max_chunks 限制参数
|
2026-07-26 10:57:30 +08:00 |
|
ViperEkura
|
33c8720d69
|
feat: 并行 batch tokenization + cache_jsonl 批处理支持
- pipeline/tokenize/tokenizer.py: encode() 全部走 encode_batch(支持单条/批量)
- pipeline/processors/base.py: BaseProcessor 新增 process_batch()
- pipeline/processors/pretrain.py: PreTrainProcessor 覆盖 process_batch() 批量编码
- pipeline/io/export.py: cache_jsonl 新增 batch_size 参数默认 1000, 批量处理
- scripts/cache_h5.py: 新增 --batch-size 参数, 默认 tokenizer 路径改为 ../AstrAI/params
|
2026-07-25 12:19:45 +08:00 |
|
ViperEkura
|
e6787a2036
|
feat: add BELLE SFT export script, remove Firefly-1.1M-Rephrased
|
2026-07-17 23:15:18 +08:00 |
|
ViperEkura
|
fec37545e2
|
Add max_chunks=18 to dolma-30b export, remove max_chunks=5 from english-wiki, rename project to datapipline
|
2026-07-06 08:53:18 +08:00 |
|
ViperEkura
|
900cd91798
|
feat: MinHash+LSH 去重 + Strategy/Factory 存储后端
|
2026-07-04 14:47:38 +08:00 |
|
ViperEkura
|
816c02dab0
|
refactor: SFT 统一 messages 格式 + ChatML 纯 jinja 渲染
|
2026-07-04 14:32:35 +08:00 |
|
ViperEkura
|
06735b9cb3
|
fix: 修复 nl_id + BFD 按 group_size 分批打包(存盘不拆分文件)
|
2026-07-03 17:07:26 +08:00 |
|
ViperEkura
|
598e1ce4ae
|
refactor: 重构打包模块,新增 BFD/FFD/Greedy 三种 bin-packing 算法,默认 BFD
- 将 pipeline/packing.py 拆分为 packing/ 子包 (base/stream/binpack)
- 新增 BfdPacker(默认)/FfDPacker/GreedyPacker,移除 StreamingPacker
- 超长序列直接截断至 pack_size
- group_size 语义改为"每 N 个 chunk 合并为一块",默认 1000
- 新增 AutoTokenizer.token_to_id(),修复 ChatML 中 hacky 的 nl_id 获取
- pad_value 默认改为 2(pad_token_id),position_ids pad=0, loss_mask pad=False
- 新增 position_ids 打包后归零一致性测试
- scripts/cache_h5.py 新增 --pack-algo 参数
|
2026-07-03 16:17:27 +08:00 |
|
ViperEkura
|
2f919e9243
|
feat: add MetaMathQA SFT export script and fix ChatML newline encoding
|
2026-07-02 17:40:03 +08:00 |
|
ViperEkura
|
fd99b9d852
|
feat: add OpenHermes-2.5 SFT export script
|
2026-06-04 14:27:04 +08:00 |
|
ViperEkura
|
aa2ea4f3a6
|
feat: SFT 增加 position_ids 边界处理
- SFTProcessor 输出 per-sample position_ids(torch.arange),每个样本从 0 开始
- position_ids 与 sequence/loss_mask 一同打包,边界处自然重置
- PT 路径不生成 position_ids
- 新增 TestPositionIds 测试及 SFTProcessor 相关测试
|
2026-06-04 14:19:32 +08:00 |
|
jiachun
|
efba7a009f
|
feat: add sft scripts for Magicoder-Evol-Instruct-110K and alpaca-gpt4-data
- sft_magicoder.py: export ise-uiuc/Magicoder-Evol-Instruct-110K
- sft_alpaca_gpt4.py: export llm-wizard/alpaca-gpt4-data with instruction+input
- remove deprecated SFT scripts
|
2026-06-01 15:50:13 +08:00 |
|
ViperEkura
|
e1125be3a7
|
fix: mask 全链路保持 bool dtype(创建→打包→HDF5落盘/读盘)
|
2026-05-15 17:23:51 +08:00 |
|
ViperEkura
|
625695fd71
|
fix: padding 默认值改为 0,避免 loss_mask 误计 padding 位置
|
2026-05-15 16:59:42 +08:00 |
|
ViperEkura
|
cb6bfcb976
|
refactor: 重构流水线架构,添加Pipeline抽象并拆分IOHandler
|
2026-04-23 19:45:57 +08:00 |
|
ViperEkura
|
a38334f4ce
|
refactor: 修改模型架构
|
2026-04-16 18:57:21 +08:00 |
|
ViperEkura
|
cb93f8219e
|
chore: 更新.gitigore
|
2026-04-02 18:48:54 +08:00 |
|
ViperEkura
|
f44ad6912e
|
fix: 修复特殊token 问题
|
2026-04-02 16:16:02 +08:00 |
|
ViperEkura
|
e01ec081b3
|
fix: 修复打包策略问题
|
2026-03-30 21:08:22 +08:00 |
|
ViperEkura
|
35963bcb08
|
reafactor: 重构项目
|
2026-03-30 20:58:51 +08:00 |
|
ViperEkura
|
f67bad0d8b
|
perf: 优化 processors、cache、packing 模块性能并简化 README
|
2026-03-30 17:50:26 +08:00 |
|
ViperEkura
|
7baa3ea0c3
|
feat: 增加日志管理
|
2026-03-30 16:28:56 +08:00 |
|
ViperEkura
|
71887bb4bb
|
fix: 修复 pipeline 模块中的打包逻辑缺陷并完善测试覆盖
|
2026-03-30 12:22:37 +08:00 |
|
ViperEkura
|
07ef471aa6
|
chore: 增加 h5_cache 脚本
|
2026-03-21 21:33:38 +08:00 |
|
ViperEkura
|
909f1b400c
|
refactor: 修改项目结构
|
2026-03-21 21:25:59 +08:00 |
|
ViperEkura
|
5b47ab1dea
|
chore: 更新.gitigore
|
2026-03-21 16:46:53 +08:00 |
|
ViperEkura
|
3fb2326115
|
refactor: 重构数据管道模块
|
2026-03-20 21:36:37 +08:00 |
|
ViperEkura
|
7045de9c32
|
fix: 修复存储和加载逻辑
|
2026-03-06 11:03:42 +08:00 |
|
ViperEkura
|
fbda7163a0
|
reafactor: 放弃文件压缩设置
|
2026-03-01 15:34:06 +08:00 |
|
ViperEkura
|
3a8edc156d
|
refactor: 更新dump_files函数
|
2026-02-27 19:46:00 +08:00 |
|
ViperEkura
|
eb2c6d4ffb
|
refactor: 更新 gitignore
|
2026-02-27 19:03:09 +08:00 |
|
ViperEkura
|
faa391c115
|
fix: 修复文件缀问题
|
2026-02-27 19:02:12 +08:00 |
|
ViperEkura
|
5d32b01860
|
build: 增加build 设置
|
2026-02-27 17:38:57 +08:00 |
|
ViperEkura
|
c32784bf9f
|
reafactor: 使用h5 文件格式存储
|
2026-02-27 17:31:54 +08:00 |
|
ViperEkura
|
617ef7164d
|
feat(tokenizer): 优化BPE分词器的预处理和训练配置
|
2025-12-21 21:50:53 +08:00 |
|
ViperEkura
|
397c8e5d0f
|
fix(module): 修复一些错误
|
2025-10-30 15:17:06 +08:00 |
|
ViperEkura
|
1deb6d59a3
|
fix(tokenizer): 更新特殊token定义以支持新的对话格式
|
2025-10-25 16:50:41 +08:00 |
|
ViperEkura
|
491b39b23f
|
fix(utils): 优化序列打包逻辑以提高效率和正确性
|
2025-10-07 12:11:40 +08:00 |
|
ViperEkura
|
cd3a22c1ac
|
feat(utils): 为缓存文件函数添加 packing 和 padding 参数支持
|
2025-10-02 22:01:00 +08:00 |
|
ViperEkura
|
2ccb26add4
|
feat(tokenizer): 调整 BPE 分词器的字母表限制并新增用户与系统标识符属性
|
2025-10-02 22:00:52 +08:00 |
|
ViperEkura
|
85bdd2bf4e
|
refactor(.gitignore): 更新输出目录忽略规则
|
2025-09-29 16:59:29 +08:00 |
|
ViperEkura
|
e844680025
|
fix(utils): 重命名变量并统一张量转换逻辑
|
2025-09-29 14:13:46 +08:00 |
|
ViperEkura
|
c3ebc4f681
|
refactor(utils): 简化SFT处理器逻辑
|
2025-09-28 15:50:10 +08:00 |
|
ViperEkura
|
44be73d255
|
refactor(supervised_finetuning): 重构 Chinese Instruct 数据集处理逻辑
|
2025-09-12 13:46:04 +08:00 |
|
ViperEkura
|
a976dc931b
|
refactor(utils): 更新 comprehensive_normalization 函数
|
2025-08-25 19:36:59 +08:00 |
|
ViperEkura
|
be79108f84
|
refactor(utils): 更新缓存文件键名
|
2025-08-11 13:49:58 +08:00 |
|
ViperEkura
|
96f3eb27fb
|
refactor(reforce_learning): 修改 DPO 数据集处理输出目录
|
2025-08-11 11:35:56 +08:00 |
|
ViperEkura
|
fcd2b66b5c
|
feat(utils): 添加 DPO 训练模式的处理器
|
2025-08-10 14:22:48 +08:00 |
|