Hy-MT2-1.8B WMT26 — Video Subtitle Translation (zh→en)
Fine-tuned from Hy-MT2-1.8B using QLoRA on TVsub subtitle data for the WMT26 Video Subtitle Translation shared task.
Training Details
- Base model: tencent/Hy-MT2-1.8B
- Fine-tuning: 4-bit QLoRA (rank=16, alpha=32, dropout=0.05)
- Data: 500K sentence pairs from TVsub (processed split)
- Hardware: 2× NVIDIA 3080Ti (12GB) - Local
- Training time: ~5 hours
Usage
from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("ShivRamSaud/hy-mt2-1.8b-wmt26")
tokenizer = AutoTokenizer.from_pretrained("ShivRamSaud/hy-mt2-1.8b-wmt26")
prompt = tokenizer.apply_chat_template([
{"role": "system", "content": "Translate the following Chinese subtitle to English."},
{"role": "user", "content": "你好，世界"},
], tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt")
output = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(output[0], skip_special_tokens=True))
Performance
- BLEU: 32.48
- chrF: 44.07
- TER: 55.99
Evaluation on a 200-pair dev split sampled from TVsub training data.
License
Apache 2.0