Hy-MT2-1.8B-1.25Bit-GGUF

STQ1_0 quantized GGUF of Tencent's Hy-MT2 1.8B translation model for on-device inference.

Base Model

• Model: Hy-MT2 1.8B (33-language translation)
• Provider: Tencent
• Source: https://huggingface.co/tencent/Hy-MT2-1.8B
• License: Apache 2.0
• Parameters: 1.8B

Quantization Details

• Format: STQ1_0 (1.25-bit Sherry quantization)
• GGUF Size: ~441 MB
• Original source: AngelSlim/Hy-MT2-1.8B-1.25Bit-GGUF (commit 5946a42)
• Hosted by: Zanish Labs / Voco

Runtime

• Compatible with llama.cpp (CPU/NEON)
• Requires STQ1_0 generic kernel dispatch on Apple Silicon (NEON kernel broken)
• Tested on Voco iOS app (llama.cpp PR #22836)

Attribution

This is a quantized derivative. The original model was created by Tencent. See LICENSE for full license text.

Links

• Original model: https://huggingface.co/tencent/Hy-MT2-1.8B
• Voco app: https://github.com/Zane-dev16/Voco
• llama.cpp: https://github.com/Zane-dev16/llama.cpp-voco