---
license: apache-2.0
library_name: transformers
pipeline_tag: token-classification
tags:
  - anime
  - filename-parsing
  - bert
  - token-classification
  - onnx
datasets:
  - ModerRAS/AnimeName
language:
  - en
  - ja
  - zh
model-index:
  - name: AniFileBERT
    results:
      - task:
          type: token-classification
          name: Anime filename token classification
        dataset:
          name: AniFileBERT fixed parser regression cases
          type: parser-regression
        metrics:
          - type: accuracy
            name: Fixed parser model-only full-match accuracy
            value: 0.9231
          - type: accuracy
            name: Fixed parser thin-runtime full-match accuracy
            value: 1.0
---

# AniFileBERT

**中文**：AniFileBERT 是一个面向番剧发布文件名的轻量级 BERT token-classification 解析器。它把常见发布名解析为结构化字段：字幕组、标题、季、集数、分辨率、来源和 special tag。

**English**: AniFileBERT is a lightweight BERT token-classification parser for anime release filenames. It extracts structured fields: release group, title, season, episode, resolution, source, and special tags.

This repository is the Hugging Face model repo used by MiruPlay as `tools/anime_parser`.

## Model Details / 模型信息

| Item | Value |
| --- | --- |
| Architecture / 架构 | `BertForTokenClassification` |
| Tokenizer / 分词器 | Custom character tokenizer in `anifilebert/tokenizer.py` |
| Parameters / 参数量 | 4,783,631 |
| Hidden size / 隐层维度 | 256 |
| Layers / 层数 | 4 |
| Attention heads / 注意力头 | 8 |
| Max sequence length / 最大长度 | 128 |
| Labels / 标签 | BIO labels for `TITLE`, `SEASON`, `EPISODE`, `GROUP`, `RESOLUTION`, `SOURCE`, `SPECIAL` |
| Default checkpoint / 默认权重 | Repository root files (`config.json`, `model.safetensors`, `vocab.json`, `tokenizer_config.json`) |
| ONNX export / ONNX 导出 | `exports/anime_filename_parser.onnx` |
| Training lineage / 训练链路 | `reports/training_lineage.json` |

**中文**：根目录就是发布 checkpoint，不再保留旧的 `model/` 重复副本。默认解析路径是“模型 logits + 约束 BIO + 薄字段规范化”，不再默认启用重结构规则；直接 `from_pretrained()` 只能加载 token-classification 权重。

**English**: The repository root is the published checkpoint. The default parser is model logits + constrained BIO + thin field normalization; heavy structural assist is not enabled by default. `from_pretrained()` only loads token-classification weights.

## Intended Use / 使用场景

**中文**

- 解析番剧/动画发布文件名，用于媒体库刮削、归类、搜索和展示。
- 覆盖常见结构：`[GROUP] TITLE - EP [META]`、点分隔 `S01E07`、国漫多括号标题、BD 特典 `NCOP/NCED/IV05`、长集数、第二季别名等。
- 不适合泛化为自然语言 NER；这是结构化文件名解析任务。

**English**

- Parse anime release filenames for media library scraping, classification, search, and display.
- Covers common layouts: `[GROUP] TITLE - EP [META]`, dotted `S01E07`, Chinese animation bracket layouts, BD extras such as `NCOP/NCED/IV05`, long-running episode numbers, and season aliases.
- This is not a general natural-language NER model; it is a structured filename parser.

## Install / 安装

```powershell
uv sync
```

If the dataset submodule is missing:

```powershell
git submodule update --init --recursive
```

## Quick Start / 快速使用

Run the Python parser:

```powershell
uv run python -m anifilebert.inference --model-dir . "[GM-Team][国漫][神印王座][Throne of Seal][2022][200][AVC][GB][1080P].mp4"
```

Expected output:

```json
{"title":"神印王座","season":null,"episode":200,"group":"GM-Team","resolution":"1080P","source":"GB","special":null}
```

Load the raw Transformers model:

```python
from transformers import BertForTokenClassification

model = BertForTokenClassification.from_pretrained("ModerRAS/AniFileBERT")
```

**中文**：如果需要完整字段解析，请 clone 本仓库并使用 `python -m anifilebert.inference`，因为分词器和后处理是自定义的。

**English**: For complete field parsing, clone this repo and use `python -m anifilebert.inference`; the tokenizer and postprocessing are custom.

## ONNX Usage / ONNX 使用

The ONNX graph outputs token logits only. A complete parser still needs:

1. custom character tokenization,
2. constrained BIO decoding,
3. field aggregation and thin string/number normalization.

本仓库提供最小可运行示例：

```powershell
uv run python -m tools.onnx_inference "[YYDM&VCB-Studio] Shinsekai Yori [NCED02][Ma10p_1080p][x265_flac].mkv"
```

Static graph shapes:

- `input_ids`: `int64[1,128]`
- `attention_mask`: `int64[1,128]`
- `logits`: `float32[1,128,15]`

More details: [`docs/onnx.md`](docs/onnx.md) and [`docs/android.md`](docs/android.md).

## Evaluation / 评估

Current published checkpoint:

| Metric / 指标 | Value / 数值 |
| --- | --- |
| Fixed regression, model-only / 固定回归，纯模型聚合 | 24/26 full match = `92.31%` |
| Fixed regression, default thin runtime / 固定回归，默认薄层运行时 | 26/26 full match = `100%` |
| Held-out parse, model-only / held-out 解析，纯模型聚合 | 1962/2048 full match = `95.80%` |
| Held-out parse, default thin runtime / held-out 解析，默认薄层运行时 | 1988/2048 full match = `97.07%` |
| Token/entity eval / token/entity 评估 | F1 `0.9844`, token accuracy `0.9961` |
| ONNX parity / ONNX 误差 | max abs diff `4.0054e-05` |
| CPU thin-runtime latency / CPU 薄层运行时延迟 | ONNX avg `12.04 ms`, P95 `13.81 ms` |

**中文**：当前发布模型是“两阶段训练”产物：先用 Rust 预生成 `20,439,848` 行虚拟 BIO shard，在 RTX 5070 Ti 上完整训练 10 epoch / `114,070` optimizer steps；再接 1 epoch light hard-case focus 微调。细节见 `reports/training_lineage.json`。README 主指标以 `model-only` 和默认薄层 `normalized-only` 为准；旧版结构规则辅助层已移除，不再作为运行时或质量对照。

**English**: The published checkpoint was trained in two stages: a 10-epoch CUDA fine-tune over `20,439,848` Rust-generated virtual BIO shard rows (`114,070` optimizer steps) on the RTX 5070 Ti, followed by a 1-epoch light hard-case focus fine-tune. See `reports/training_lineage.json` for details. README quality numbers prioritize `model-only` and the default thin `normalized-only` runtime; structural filename assists have been removed from the runtime and quality reports.

Run regression:

```powershell
uv run python -m tools.evaluate_parser_cases --model-dir . --case-file data/parser_regression_cases.json --output reports/case_metrics.json
```

## Performance / 性能

Benchmark command:

性能测试命令：

```powershell
uv run python -m tools.benchmark_inference --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 reports/benchmark_results.json
```

Local CPU benchmark on the 26 fixed real-world cases, single-threaded, using the
default thin runtime: tokenization, model/session forward, constrained BIO
decoding, entity aggregation, and light string/number normalization:

本地 CPU 单线程测试，使用 26 条固定真实 case，默认薄层运行时，包含 tokenizer、
模型/session 前向、约束 BIO 解码、实体聚合和轻量字符串/数字规范化：

| Backend / 后端 | Load ms / 加载 ms | Avg ms / 平均 ms | P50 ms | P95 ms | P99 ms | files/s |
| --- | ---: | ---: | ---: | ---: | ---: | ---: |
| PyTorch | 46.35 | 15.36 | 14.25 | 22.27 | 29.75 | 65.1 |
| ONNX Runtime | 50.92 | 12.04 | 11.90 | 13.81 | 15.38 | 83.1 |

**中文**：这是完整薄层 parser 的端到端延迟，不是只测模型 forward。移动端实现应复用 ONNX session，并保持 tokenizer/BIO/薄规范化逻辑一致。

**English**: This is end-to-end thin-parser latency, not model-forward-only timing. Mobile code should keep the ONNX session reusable and keep tokenizer/BIO/thin-normalization behavior aligned.

## Training / 训练

Training uses the dataset submodule at `datasets/AnimeName`.

Recommended virtual-shard character-token run on the Windows RTX 5070 Ti worker:

```powershell
@'
import random
from pathlib import Path

source = Path("datasets/AnimeName/dmhy_weak_char.jsonl")
target = Path("data/generated/virtual_source_train_seed105.jsonl")
rows = [line for line in source.read_text(encoding="utf-8").splitlines() if line]
random.Random(105).shuffle(rows)
target.parent.mkdir(parents=True, exist_ok=True)
target.write_text("\n".join(rows[: int(len(rows) * 0.98)]) + "\n", encoding="utf-8")
'@ | .\.venv\Scripts\python.exe -

cargo build --release --manifest-path tools/virtual_dataset_generator/Cargo.toml
.\tools\virtual_dataset_generator\target\release\anifilebert-virtual-dataset-generator.exe `
  --input data/generated/virtual_source_train_seed105.jsonl `
  --vocab-file datasets/AnimeName/vocab.char.json `
  --output-dir data/generated/virtual_char_sps32_seed105 `
  --max-length 128 `
  --samples-per-source 32 `
  --seed 105 `
  --threads 20 `
  --separator-mode per-gap `
  --bracket-mode per-part

.\.venv\Scripts\python.exe -m anifilebert.train --tokenizer char `
  --data-file datasets/AnimeName/dmhy_weak_char.jsonl `
  --vocab-file datasets/AnimeName/vocab.char.json `
  --virtual-dataset-dir data/generated/virtual_char_sps32_seed105 `
  --save-dir checkpoints/dmhy-char-virtual-sps32-10epoch-lr1e5 `
  --init-model-dir . `
  --epochs 10 `
  --batch-size 1792 `
  --learning-rate 0.00001 `
  --warmup-steps 2000 `
  --max-seq-length 128 `
  --train-split 0.98 `
  --num-workers 4 `
  --prefetch-factor 4 `
  --persistent-workers `
  --checkpoint-steps 5000 `
  --save-total-limit 3 `
  --parse-eval-limit 2048 `
  --case-eval-file data/parser_regression_cases.json `
  --bf16 `
  --no-periodic-eval `
  --perf-log-steps 1000 `
  --perf-sample-interval 0.5 `
  --seed 105 `
  --experiment-name dmhy-char-virtual-sps32-10epoch-lr1e5
```

`python -m anifilebert.train` writes:

- Hugging Face checkpoints under `--save-dir`,
- `final/run_metadata.json`,
- `final/trainer_eval_metrics.json`,
- `final/parse_eval_metrics.json`,
- `final/case_metrics.json` unless `--no-case-eval` is used,
- `final/perf_metrics.json` when `--perf-log-steps` is set,
- TensorBoard logs unless `--no-tensorboard` is used.

Full workflow: [`docs/training.md`](docs/training.md).

## Dataset / 数据集

Authoritative dataset snapshot:

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

Current snapshot:

- rows / 行数: `632002`
- failed relabel rows / 重标注失败行: `0`
- strict BIO violations / 严格 BIO 违规: `0`
- character vocab / 字符词表: `6199`
- character coverage / 字符覆盖率: `100%`

**中文**：`datasets/AnimeName` 是嵌套数据集仓库。更新数据后需要先提交/推送子仓库，再提交父仓库的 submodule pointer。

**English**: `datasets/AnimeName` is a nested dataset repository. Commit and push the dataset repo first, then commit the updated submodule pointer in this model repo.

## Repository Layout / 仓库结构

```text
config.json
model.safetensors
tokenizer_config.json
vocab.json
training_args.bin
anifilebert/
tools/
data/parser_regression_cases.json
datasets/AnimeName/
exports/anime_filename_parser.onnx
docs/
reports/
```

## Maintenance / 维护

See [`docs/maintenance.md`](docs/maintenance.md) for release steps, LFS order, dataset submodule updates, and MiruPlay integration notes.

## Limitations / 局限

**中文**

- 发布命名没有统一标准，极端 OCR 噪声、乱码、非动画命名仍可能失败。
- ONNX 只包含模型 logits，不包含 tokenizer、BIO decode 和薄字段规范化；移动端必须保持 tokenizer/vocab/config 一致。
- `source` 当前是单值字段，复杂文件名里可能同时存在平台、发布源、编码器和语言标签。

**English**

- Anime release names are not standardized; extreme OCR noise, mojibake, or non-anime names can still fail.
- ONNX contains logits only. Mobile runtimes must keep tokenizer, vocabulary, config, BIO decode, and thin normalization in sync.
- `source` is currently a single field, while real filenames may contain platform, release source, codec, and language tags together.