---
license: other
license_name: tencent-hy-community
license_link: https://huggingface.co/tencent/HY-MT1.5-1.8B/blob/main/License.txt
base_model: tencent/HY-MT1.5-1.8B
tags:
- gguf
- lora
- llama.cpp
- translation
- zh-vi
- xianxia
---

# HY-MT Xianxia LoRA VI GGUF

GGUF LoRA adapter for `DanVP/hy-mt-xianxia-lora-vi`, intended for use with
the official `tencent/HY-MT1.5-1.8B-GGUF` base model.

## Files

- `hy-mt-xianxia-lora-vi-f16.gguf` - f16 LoRA adapter converted from the PEFT
  adapter in `DanVP/hy-mt-xianxia-lora-vi`.

## Recommended Base

For CPU demos, start with:

- `tencent/HY-MT1.5-1.8B-GGUF/HY-MT1.5-1.8B-Q4_K_M.gguf`

Higher quality local runs can try `Q6_K` or `Q8_0` if RAM and latency allow.

## llama.cpp Usage

```bash
llama-cli \
  -m HY-MT1.5-1.8B-Q4_K_M.gguf \
  --lora hy-mt-xianxia-lora-vi-f16.gguf \
  -p "<|prompt|>" \
  -n 1024 \
  --temp 0
```

This model uses HY-MT chat special tokens. The demo Space formats prompts as:

```text
<｜hy_begin▁of▁sentence｜><｜hy_User｜>{instruction_and_source}<｜hy_Assistant｜>
```

## License

This adapter is a derivative of Tencent HY-MT1.5 and inherits the Tencent HY
Community License. See the base model license for full terms and restrictions.