Aiden1020

LocalSubs-EN-ZH-TW-0.6B

README

License: apache-2.0

Model Details

Table with columns: Item, Value
Item	Value
Base model	NiuTrans/LMT-60-0.6B
Tokenizer	domain specific 64k tokenizer
Fine-tuning method	Supervised Fine-Tuning (SFT), causal LM loss on assistant turn only
Task	EN → TW subtitle translation, cue-level, 0–3 cues of context
Quantization	Q5_K_M (GGUF, via llama.cpp)
Best checkpoint	Step 22,000 · eval loss 1.4804
Precision	BF16 during training
Context window	256 tokens (subtitle cues are short)

Intended Use

This model is designed for one task only:

Given recent English subtitle cues as context and the current English subtitle cue, output only the Taiwan Traditional Chinese translation of the current cue.

In scope:

Real-time local subtitle overlay(avg. latency 52ms test on Macbook air M2)
Cue-level streaming inference
English → Taiwan Traditional Chinese (繁體中文・台灣用語)

Out of scope:

General-purpose translation
Long-document translation
Simplified Chinese output
General instruction following

Input / Output Format

markdown
CTX:
<0–3 previous English subtitle cues>

CUR:
<current English subtitle cue>

Output: Taiwan Traditional Chinese translation of CUR only.

Example:

markdown
CTX:
I sent you the file.
Check your inbox.

CUR:
You got it?

Expected output:

markdown
你收到了嗎？

Usage (llama.cpp)

bash
./llama-cli \
  -m LocalSubs-EN-ZH-TW-0.6B-Q5_K_M.gguf \
  --temp 0.0 \
  -p "CTX:
I sent you the file.
Check your inbox.

CUR:
You got it?" \
  -n 64

Usage (Python · llama-cpp-python)

python
from llama_cpp import Llama

llm = Llama(model_path="SubtitleEN2TW-0.6B-Q5_K_M.gguf", n_ctx=256)

ctx_cues = ["I sent you the file.", "Check your inbox."]
cur_cue  = "You got it?"

ctx_block = "
".join(ctx_cues)
prompt = f"CTX:
{ctx_block}

CUR:
{cur_cue}"

out = llm(prompt, max_tokens=64, temperature=0.0, stop=["

"])
print(out["choices"][0]["text"].strip())

Training Data

The SFT dataset was built from two public subtitle corpora. The actual dataset is not redistributed due to upstream licensing constraints. Pipeline details and the diagnostic test set are available at Aiden1020/SubtitleEN2TW-SFT-Pipeline.

Table with columns: Source, Format, Language pair, Scale (before filtering)
Source	Format	Language pair	Scale (before filtering)
OpenSubtitles v2024 (OPUS)	Moses parallel text	`en` – `zh_TW`	~18.6 M sentence pairs
TVSub	Timestamped subtitle cues	`en` – `zh`	~2.2 M cue pairs

Filtering steps applied:

Text cleaning: remove HTML/ASS/VTT tags, music symbols, encoding-corruption markers (PUA chars, kana, Cyrillic), subtitle-group watermarks
Simplified Chinese rejection: OpenCC s2t diff ratio > 3% → dropped
CTX-leakage filter: discard samples where the Chinese target is disproportionately longer than the English cue, or closely matches the previous Chinese cue
English-echo filter: discard samples where the English source appears verbatim in the Chinese output
Length validation: English 1–200 chars, Chinese 1–120 chars
LLM-filtering: using Qwen3 35b as a judge to filter the split-cue.
Cue length balance: balancing the weight of long cue in the data distribution. Context size distribution in training set (0–3 cues): 30% / 30% / 25% / 15%.

A small manually curated diagnostic set (~3% mix-in) was included to cover leakage probes, short responses, ambiguous phrases, and Taiwan terminology.

Limitations

Trained on subtitle-domain data only; may produce unnatural output for other domains.
The training data is primarily from OpenSubtitles which contains crowd-sourced subtitles; quality varies.
Output targets Taiwan Traditional Chinese style but may occasionally produce Hong Kong or neutral Traditional Chinese forms.
Context crossing episode boundaries is treated as acceptable noise; no document-level segmentation is applied.
Not suitable for general chat or instruction following.

Citation

If you use this model, please also cite the model:

bibtex
@misc{wong2026localsubs,
  title         = {Workload-Driven Optimization for On-Device Real-Time Subtitle Translation},
  author        = {Tsz-To Wong},
  year          = {2026},
  eprint        = {2607.09957},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  doi           = {10.48550/arXiv.2607.09957},
  url           = {https://arxiv.org/abs/2607.09957}
}

also cite the base model:

bibtex
@misc{luoyf2025lmt,
  title={NiuTrans.LMT: Toward Inclusive and Scalable Multilingual Machine Translation with LLMs},
  author={Yingfeng Luo, Ziqiang Xu, Yuxuan Ouyang, Murun Yang, Dingyang Lin, Kaiyan Chang, Tong Zheng, Bei Li, Peinan Feng, Quan Du, Tong Xiao, Jingbo Zhu},
  year={2025},
  eprint={2511.07003},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2511.07003}
}

License

This model is released under the Apache License 2.0.

See NOTICE for third-party attribution.

Available on FriendliAI

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Model Details

Model Provider

Aiden1020

Model Tree

Base

NiuTrans/LMT-60-0.6B

Quantized

this model

Input Modalities

Text

Output Modalities

Text

Supported Functionality

Dedicated Endpoints

Container

Explore FriendliAI today

Get started Talk to an engineer

README

License: apache-2.0

Model Details

Table with columns: Item, Value
Item	Value
Base model	NiuTrans/LMT-60-0.6B
Tokenizer	domain specific 64k tokenizer
Fine-tuning method	Supervised Fine-Tuning (SFT), causal LM loss on assistant turn only
Task	EN → TW subtitle translation, cue-level, 0–3 cues of context
Quantization	Q5_K_M (GGUF, via llama.cpp)
Best checkpoint	Step 22,000 · eval loss 1.4804
Precision	BF16 during training
Context window	256 tokens (subtitle cues are short)

Intended Use

This model is designed for one task only:

Given recent English subtitle cues as context and the current English subtitle cue, output only the Taiwan Traditional Chinese translation of the current cue.

In scope:

Real-time local subtitle overlay(avg. latency 52ms test on Macbook air M2)
Cue-level streaming inference
English → Taiwan Traditional Chinese (繁體中文・台灣用語)

Out of scope:

General-purpose translation
Long-document translation
Simplified Chinese output
General instruction following

Input / Output Format

markdown
CTX:
<0–3 previous English subtitle cues>

CUR:
<current English subtitle cue>

Output: Taiwan Traditional Chinese translation of CUR only.

Example:

markdown
CTX:
I sent you the file.
Check your inbox.

CUR:
You got it?

Expected output:

markdown
你收到了嗎？

Usage (llama.cpp)

bash
./llama-cli \
  -m LocalSubs-EN-ZH-TW-0.6B-Q5_K_M.gguf \
  --temp 0.0 \
  -p "CTX:
I sent you the file.
Check your inbox.

CUR:
You got it?" \
  -n 64

Usage (Python · llama-cpp-python)

python
from llama_cpp import Llama

llm = Llama(model_path="SubtitleEN2TW-0.6B-Q5_K_M.gguf", n_ctx=256)

ctx_cues = ["I sent you the file.", "Check your inbox."]
cur_cue  = "You got it?"

ctx_block = "
".join(ctx_cues)
prompt = f"CTX:
{ctx_block}

CUR:
{cur_cue}"

out = llm(prompt, max_tokens=64, temperature=0.0, stop=["

"])
print(out["choices"][0]["text"].strip())

Training Data

Table with columns: Source, Format, Language pair, Scale (before filtering)
Source	Format	Language pair	Scale (before filtering)
OpenSubtitles v2024 (OPUS)	Moses parallel text	`en` – `zh_TW`	~18.6 M sentence pairs
TVSub	Timestamped subtitle cues	`en` – `zh`	~2.2 M cue pairs

Filtering steps applied:

Text cleaning: remove HTML/ASS/VTT tags, music symbols, encoding-corruption markers (PUA chars, kana, Cyrillic), subtitle-group watermarks
Simplified Chinese rejection: OpenCC s2t diff ratio > 3% → dropped
CTX-leakage filter: discard samples where the Chinese target is disproportionately longer than the English cue, or closely matches the previous Chinese cue
English-echo filter: discard samples where the English source appears verbatim in the Chinese output
Length validation: English 1–200 chars, Chinese 1–120 chars
LLM-filtering: using Qwen3 35b as a judge to filter the split-cue.
Cue length balance: balancing the weight of long cue in the data distribution. Context size distribution in training set (0–3 cues): 30% / 30% / 25% / 15%.

A small manually curated diagnostic set (~3% mix-in) was included to cover leakage probes, short responses, ambiguous phrases, and Taiwan terminology.

Limitations

Trained on subtitle-domain data only; may produce unnatural output for other domains.
The training data is primarily from OpenSubtitles which contains crowd-sourced subtitles; quality varies.
Output targets Taiwan Traditional Chinese style but may occasionally produce Hong Kong or neutral Traditional Chinese forms.
Context crossing episode boundaries is treated as acceptable noise; no document-level segmentation is applied.
Not suitable for general chat or instruction following.

Citation

If you use this model, please also cite the model:

bibtex
@misc{wong2026localsubs,
  title         = {Workload-Driven Optimization for On-Device Real-Time Subtitle Translation},
  author        = {Tsz-To Wong},
  year          = {2026},
  eprint        = {2607.09957},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  doi           = {10.48550/arXiv.2607.09957},
  url           = {https://arxiv.org/abs/2607.09957}
}

also cite the base model:

bibtex
@misc{luoyf2025lmt,
  title={NiuTrans.LMT: Toward Inclusive and Scalable Multilingual Machine Translation with LLMs},
  author={Yingfeng Luo, Ziqiang Xu, Yuxuan Ouyang, Murun Yang, Dingyang Lin, Kaiyan Chang, Tong Zheng, Bei Li, Peinan Feng, Quan Du, Tong Xiao, Jingbo Zhu},
  year={2025},
  eprint={2511.07003},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2511.07003}
}

License

This model is released under the Apache License 2.0.

See NOTICE for third-party attribution.