# Training Guide / 训练指南

This document describes the reproducible training workflow for AniFileBERT.

本文档记录 AniFileBERT 的可复现训练流程。

## 1. Environment / 环境

Use `uv` for all dependency and command execution.

所有依赖和命令优先使用 `uv`。

```powershell
uv sync
uv run python -c "import torch; print(torch.__version__, torch.cuda.is_available())"
```

Recommended GPU configuration:

推荐 GPU 配置：

- RTX 3080 class GPU or better
- batch size `192` to `256` for full char training
- `fp16` enabled automatically when CUDA is available
- `--num-workers 4` or `8` when the local disk can keep up

## 2. Dataset / 数据集

The authoritative dataset lives in the nested submodule:

权威数据集位于嵌套子模块：

```text
datasets/AnimeName/dmhy_weak.jsonl
datasets/AnimeName/dmhy_weak_char.jsonl
datasets/AnimeName/vocab.json
datasets/AnimeName/vocab.char.json
```

Current expected properties:

当前期望属性：

- rows / 行数: `632002`
- strict BIO violations / 严格 BIO 违规: `0`
- character vocab / 字符词表: `6199`
- character coverage / 字符覆盖率: `100%`

## 3. Relabel Full Dataset / 全量重标注

Use this when weak-label rules changed in `dmhy_dataset.py` or `label_repairs.py`.

当 `dmhy_dataset.py` 或 `label_repairs.py` 的弱标注规则改变时，使用此流程。

```powershell
uv run python relabel_dataset_from_filenames.py `
  --input datasets/AnimeName/dmhy_weak.jsonl `
  --output datasets/AnimeName/dmhy_weak.relabel.jsonl `
  --manifest-output datasets/AnimeName/dmhy_weak.relabel.manifest.json `
  --vocab-output datasets/AnimeName/vocab.relabel.json `
  --base-vocab datasets/AnimeName/vocab.json `
  --max-vocab-size 8000 `
  --progress 50000
```

After checking the manifest and sample labels, replace the authoritative files:

检查 manifest 和样本标注后，再替换权威文件：

```powershell
Move-Item datasets/AnimeName/dmhy_weak.relabel.jsonl datasets/AnimeName/dmhy_weak.jsonl -Force
Move-Item datasets/AnimeName/vocab.relabel.json datasets/AnimeName/vocab.json -Force
Move-Item datasets/AnimeName/dmhy_weak.relabel.manifest.json datasets/AnimeName/dmhy_weak.manifest.json -Force
```

## 4. Convert to Character Dataset / 转换为字符数据集

The published checkpoint uses the character tokenizer.

当前发布模型使用字符级 tokenizer。

```powershell
uv run python convert_to_char_dataset.py `
  --input datasets/AnimeName/dmhy_weak.jsonl `
  --output datasets/AnimeName/dmhy_weak_char.jsonl `
  --vocab-output datasets/AnimeName/vocab.char.json `
  --manifest-output datasets/AnimeName/dmhy_weak_char.manifest.json `
  --progress 50000
```

## 5. Full Training / 全量训练

Recommended RTX 3080 run:

推荐 RTX 3080 训练命令：

```powershell
uv run python train.py --tokenizer char `
  --data-file datasets/AnimeName/dmhy_weak_char.jsonl `
  --vocab-file datasets/AnimeName/vocab.char.json `
  --save-dir checkpoints/dmhy-char-full `
  --init-model-dir . `
  --epochs 2 `
  --batch-size 256 `
  --learning-rate 0.00008 `
  --warmup-steps 300 `
  --max-seq-length 128 `
  --train-split 0.98 `
  --num-workers 4 `
  --checkpoint-steps 1000 `
  --save-total-limit 3 `
  --parse-eval-limit 2048 `
  --case-eval-file data/parser_regression_cases.json `
  --seed 52 `
  --experiment-name dmhy-char-full
```

Training outputs:

训练输出：

- `checkpoints/<run>/checkpoint-*`: resumable checkpoints / 可恢复 checkpoint
- `checkpoints/<run>/final`: final Hugging Face checkpoint / 最终 checkpoint
- `final/run_metadata.json`: run configuration / 训练配置
- `final/trainer_eval_metrics.json`: seqeval metrics / token/entity 指标
- `final/parse_eval_metrics.json`: held-out parser exact-match / held-out 解析准确率
- `final/case_metrics.json`: fixed real-world case regression / 固定真实 case 回归
- TensorBoard logs unless `--no-tensorboard` is set / 默认写 TensorBoard

## 6. Thin Hard-Case Fine-Tuning / 薄层困难样本微调

Use hard-case fine-tuning only after a specific real-world failure pattern has
been confirmed, fixed in the weak labels, and added to
`data/parser_regression_cases.json`.

只有在确认某类真实失败样式、修复弱标注并加入
`data/parser_regression_cases.json` 后，才使用困难样本微调。

```powershell
uv run python build_repair_focus_dataset.py `
  --input datasets/AnimeName/dmhy_weak_char.jsonl `
  --output data/thin_hard_focus_char.jsonl `
  --context-samples 30000 `
  --repeat-focus 3 `
  --repeat-manual 240 `
  --seed 57

uv run python train.py --tokenizer char `
  --data-file data/thin_hard_focus_char.jsonl `
  --vocab-file datasets/AnimeName/vocab.char.json `
  --save-dir checkpoints/dmhy-char-thin-hardfocus `
  --init-model-dir . `
  --epochs 2 `
  --batch-size 256 `
  --learning-rate 0.00004 `
  --warmup-steps 80 `
  --max-seq-length 128 `
  --train-split 0.95 `
  --num-workers 4 `
  --checkpoint-steps 300 `
  --save-total-limit 2 `
  --parse-eval-limit 1024 `
  --case-eval-file data/parser_regression_cases.json `
  --seed 58 `
  --experiment-name dmhy-char-thin-hardfocus
```

The default quality gate is model-led parsing:

默认质量门槛以模型主导解析为准：

- fixed regression `model_only >= 85%`
- held-out parse `model_only >= 75%`
- `normalized_only` is the default thin runtime metric
- structural filename assists are not part of training or release metrics

- 固定回归 `model_only >= 85%`
- held-out 解析 `model_only >= 75%`
- `normalized_only` 是默认薄层运行时指标
- 结构化文件名辅助不属于训练或发布指标

## 7. Publish to Repository Root / 发布到仓库根目录

The repository root is the Hugging Face checkpoint surface.

仓库根目录就是 Hugging Face checkpoint 发布面。

```powershell
$final = "checkpoints/dmhy-char-full/final"
Copy-Item "$final/config.json" . -Force
Copy-Item "$final/model.safetensors" . -Force
Copy-Item "$final/tokenizer_config.json" . -Force
Copy-Item "$final/training_args.bin" . -Force
Copy-Item "$final/vocab.json" . -Force
Copy-Item "$final/run_metadata.json" . -Force
Copy-Item "$final/trainer_eval_metrics.json" . -Force
Copy-Item "$final/parse_eval_metrics.json" . -Force
Copy-Item "$final/case_metrics.json" . -Force
Copy-Item datasets/AnimeName/vocab.char.json .\vocab.char.json -Force
```

Then export ONNX:

然后导出 ONNX：

```powershell
uv run python export_onnx.py --model-dir . --output exports/anime_filename_parser.onnx --max-length 128
```

## 8. Validation Checklist / 验证清单

Run these before committing:

提交前执行：

```powershell
uv run python -m py_compile tokenizer.py dataset.py dmhy_dataset.py label_repairs.py train.py inference.py export_onnx.py onnx_inference.py
uv run python evaluate_parser_cases.py --model-dir . --case-file data/parser_regression_cases.json --output case_metrics.json
uv run python inference.py --model-dir . "[GM-Team][国漫][神印王座][Throne of Seal][2022][200][AVC][GB][1080P].mp4"
uv run python onnx_inference.py "[YYDM&VCB-Studio] Shinsekai Yori [NCED02][Ma10p_1080p][x265_flac].mkv"
uv run python benchmark_inference.py --model-dir . --onnx exports/anime_filename_parser.onnx --case-file data/parser_regression_cases.json --repeat 20 --warmup 20 --torch-threads 1 --ort-threads 1 --output benchmark_results.json
```

## 9. Git and LFS Order / Git 与 LFS 顺序

If the dataset submodule changed:

如果数据集子模块有变动：

```powershell
git -C datasets/AnimeName add dmhy_weak.jsonl dmhy_weak.manifest.json dmhy_weak_char.jsonl dmhy_weak_char.manifest.json vocab.json vocab.char.json
git -C datasets/AnimeName commit -m "Update anime filename labels"
git -C datasets/AnimeName lfs push origin main --all
git -C datasets/AnimeName push origin main
```

Then commit the model repo:

再提交模型仓库：

```powershell
git add README.md MAINTENANCE.md ANDROID.md docs/training.md docs/onnx.md `
  config.json model.safetensors tokenizer_config.json training_args.bin vocab.json vocab.char.json `
  exports/anime_filename_parser.onnx exports/anime_filename_parser.metadata.json `
  train.py inference.py export_onnx.py onnx_inference.py data/parser_regression_cases.json datasets/AnimeName
git commit -m "Update AniFileBERT model and documentation"
git lfs push origin main --all
git push origin main
```