Commit Graph
17 Commits
Author SHA1 Message Date
ViperEkura 45cc048fe9 fix: resolve audited training, import, and serving bugs
- shard the Muon Newton-Schulz orthogonalization over the FSDP mesh instead of partial local slices
- import HF checkpoints faithfully: per-head RoPE permutation for q/k projections and qk-norm, qwen3, shared experts, and qk-norm before RoPE (changes numerics for existing use_qk_norm checkpoints)
- make preprocessing and resume self-contained: backfill realigned bucket keys by semantics (masks ones, rest zeros) and snapshot tokenizer files into every checkpoint
- keep RL consistent: sync the offline GRPO old_model each optimizer step and validate online strategies through a public one-off-rollout hook that leaves the replay cache untouched
- fix streaming serving: withhold partial tool-call prefixes with a stream-end flush, stream tool-call arguments from the raw source span, and terminate SSE frames with a blank line
- fix sampling semantics: capture logprobs before top-k/top-p mutate logits in place and detect greedy pipelines polymorphically instead of isinstance bookkeeping
2026-09-03 20:27:41 +08:00
ViperEkuraand0z5a 1fad50d847 fix: publish checkpoints atomically
- Write checkpoint payloads to a hidden sibling staging directory, add a versioned checksum manifest, fsync the completed payload, and publish it with an atomic rename
- Republishing an existing step retires the old payload under a hidden sibling name before the atomic rename, so re-runs into the same output directory replace the previous checkpoint instead of raising FileExistsError
- Keep legacy checkpoints loadable, add optional checksum verification, and align metric flushing with checkpoint publication

Co-authored-by: 0z5a <dezhen.lu@student.uni-tuebingen.de>
2026-09-02 15:29:22 +08:00
0z5a 08721f6d31 fix: save checkpoints after optimizer steps
- add a post-step callback hook for checkpoint saves
- preserve updated model, optimizer, and scheduler state
- cover checkpoint ordering with a regression test
2026-09-01 12:25:27 +08:00
ViperEkura dcc96de12a test: refactor tests and fix inference edge cases
- Convert protocol and MoE test classes to plain functions
- Add real server/engine integration and generate_async tests
- Isolate test model per test and use pytest tmp_path
- Reset FastAPI engine state after inference tests
- Fix generate_async StopIteration handling on Python 3.12
- Fix HF adapter MoE dense/shared and Gemma qk_norm mapping
- Correct dev dependency httpx2 to httpx
2026-08-21 22:59:51 +08:00
ViperEkura 5ba21f4eb3 refactor: eliminate test duplication via shared helpers
- Add tests/helpers.py with shared config, dataset, tokenizer, executor, and assertion helpers
- Replace 15 copies of device one-liner with session-scoped fixture
- Collapse 5 near-identical Dataset subclasses into RandomTokenDataset
- Remove duplicate _make_config/_make_model/_make_frozen and FakeTokenizer/FakeExecutor definitions
- Make test_callbacks and test_early_stopping use existing train_config_factory
- Replace 6 duplicate meta.json read blocks with load_shard_meta
- Fix mkdtemp leaks in test_lora.py with TemporaryDirectory
2026-07-27 22:34:53 +08:00
ViperEkura 6ae1828449 refactor : 清理工厂和配置系统中的死代码与冗余抽象
- 删除 Registry 中未使用的 category/priority 字段,_entries 简化为直接存储类引用
- 修正 __init_subclass__ 避免叶子类(AutoRegressiveLM 等)创建空注册表
- 删除 5 个工厂的薄 create() 覆写,统一使用 BaseFactory.create(name, *args, **kwargs)
- 删除 3 处零调用的 available_types/available_strategies 别名死代码
- 删除零调用的 BaseModelConfig.to_file 死代码
- 将 BaseConfig.from_json/to_json 重命名为 from_file/to_file,消除与子类重复
- 移除两个 inference builder 中总是被覆写的 prompt_tokens=0
2026-06-07 11:39:50 +08:00
ViperEkura 4145d35e3c refactor: 检查点加载重构,路径替代对象传递
- model: nn.Module -> model_fn 工厂函数,spawn 边界只传字符串
- Trainer.train(resume_dir=path) — Checkpoint 不再通过 pickle 传递
- TrainContextBuilder.with_resume_dir(path) — 自动检测 meta.json 分流 resume/from-scratch
- CheckpointCallback: 拆分 state_dict 收集(全 rank)与磁盘写入(rank-0),修复 FSDP 死锁
- serialization: load_torch 支持 broadcast,消除 _load_extra/_load_torch_broadcast
- optimizer/scheduler 恢复逻辑内联到 build(),在 executor.prepare() 之后执行
- pyproject.toml: ruff exclude build/ 避免 CI 扫描构建产物
2026-05-27 20:15:29 +08:00
ViperEkura 7fa69572c0 fix: 测试日志写入临时目录避免冗余文件 2026-05-24 20:54:59 +08:00
ViperEkura 2c2697390d feat: 新增 GradientCheckpointingCallback
- TrainConfig.gradient_checkpointing_modules 指定模块类型
- apply 递归遍历,兼容 DDP,不硬编码模型结构
- modules=None 时静默跳过,零开销
2026-05-17 18:21:05 +08:00
ViperEkura d7a7f570ed refactor: 训练循环改为两重迭代并统一参数命名
- 训练循环从三重(epoch→batched→batch)改为二重(epoch→batch)
- batch_size → batch_per_device, accumulation_steps → grad_accum_steps
- scheduler 移入 step block 对齐 optimizer 更新步
- GradientClippingCallback 改用 on_step_begin 避免零梯度裁剪
- 移除 _train_impl 误导性的 -> Checkpoint 标注
- total_steps 修除为向下取整并精简为一行
- warmup_steps 改为 warmup_ratio (默认0.05)
2026-05-16 21:27:35 +08:00
ViperEkura b531232a9b style: 修改为显式导入 2026-04-04 16:02:49 +08:00
ViperEkura 0852b852f8 refactor: 优化参数传递,清理导入样式 2026-04-03 22:06:32 +08:00
ViperEkura 2e009cf59a chore: 更新项目名称 2026-03-31 09:34:11 +08:00
ViperEkura 426af2d75f style: 使用ruff 工具优化代码风格 2026-03-30 23:32:28 +08:00
ViperEkura 345fd2f091 fix: 修复参数传递问题 2026-03-30 22:22:36 +08:00
ViperEkura 6089a12cef fix: 修复参数传递问题并更新测试单元 2026-02-28 19:01:16 +08:00
ViperEkura 4da70785b5 refactor(tests): 重构测试文件目录结构 2026-01-08 21:34:52 +08:00