Hy MT2 7B Abliterated Bf16 needs about 6.2 GB of memory at Q4_K_M (a 4.6 GB file) with an 8K context window, so it fits on a 8 GB GPU. This repository has 4 quants, from 4.6 GB (Q4_K_M) to 8.0 GB (Q8_0). Longer context needs more: at 32K tokens the KV cache alone adds about 4.3 GB.
Hy MT2 7B Abliterated Bf16 – a 7.5B multilingual translation model in GGUF format
Hy MT2 7B Abliterated Bf16 is a 7.5B parameter model based on Tencent Hunyuan’s Hy‑MT2. It is an abliterated derivative that removes certain refusal directions. The model is available in GGUF format and can be run with llama.cpp or MLX.
Good at
- Multilingual translation
- Low‑refusal behavior
- High‑precision bf16 weights
Limits
- No vision support
- Requires 8‑k context for estimates
- License: Apache‑2.0
Best for
- Translating legal or medical documents
- Processing quoted speech
- Researching refusal‑related behavior
Who it’s for. Users who need a multilingual translation model with reduced refusal tendencies and who are comfortable with the Apache‑2.0 license. It is suitable for developers and researchers working on translation pipelines.
Quants & memory
| Quant | File | Est. memory | vs VRAM | Result | |
|---|---|---|---|---|---|
| Q4_K_M | 4.6 GB | 6.2 GB | Fits | Get file → | |
| Q5_K_M | 5.4 GB | 6.9 GB | Fits | Get file → | |
| Q6_K | 6.2 GB | 7.7 GB | Tight | Get file → | |
| Q8_0 | 8.0 GB | 9.6 GB | Too big | Get file → |
For 8‑GB VRAM use Q4_K_M (file size 4.62 GB, estimated memory 6.2 GB). For 12‑GB VRAM use Q5_K_M (5.37 GB, 6.9 GB). For 16‑GB VRAM use Q6_K (6.16 GB, 7.7 GB). For 24‑GB VRAM use Q8_0 (7.98 GB, 9.6 GB). Memory estimates are approximate.
What GPU do you need for Hy MT2 7B Abliterated Bf16?
These figures add the file size, an fp16 KV cache at an 8K context window (worked out from this model’s own header) and about 0.5 GB for the runtime. They are estimates, so leave a little headroom, and expect to use less context or a smaller quant if you are right at the limit.
- 4 GB VRAM: no quant fits at an 8K context.
- 6 GB VRAM: no quant fits at an 8K context.
- 8 GB VRAM: the best quant that fits is Q6_K (6.2 GB file, about 7.7 GB in use).
- 12 GB VRAM: the best quant that fits is Q8_0 (8.0 GB file, about 9.6 GB in use).
Which Hy MT2 7B Abliterated Bf16 quant should you choose?
Q4_K_M is the usual starting point for Hy MT2 7B Abliterated Bf16: it balances file size and quality well. Here it is a 4.6 GB file.
The Q8_0 file is 8.0 GB, about 1.7 times the size of Q4_K_M. It keeps the weights closer to the original, which only matters if you have the memory to spare.
New to quant names? Read GGUF quantization explained.
Specifications
Architecture from the GGUF header
Source
Frequently asked questions about Hy MT2 7B Abliterated Bf16
How much VRAM does Hy MT2 7B Abliterated Bf16 need?
About 6.2 GB at Q4_K_M with an 8K context window. That is the 4.6 GB file, plus the KV cache and roughly 0.5 GB of runtime overhead. Smaller quants need less and longer context needs more.
Can I run Hy MT2 7B Abliterated Bf16 on an 8 GB GPU?
Yes, with Q6_K: the 6.2 GB file needs about 7.7 GB at 8K context. Larger quants in this repository do not fit in 8 GB.
What is the file size of Hy MT2 7B Abliterated Bf16 GGUF?
It depends on the quant: Q4_K_M is 4.6 GB, Q5_K_M is 5.4 GB, Q6_K is 6.2 GB, Q8_0 is 8.0 GB.
What context length does Hy MT2 7B Abliterated Bf16 support?
The GGUF header lists a trained context length of 262,144 tokens (256K). The KV cache grows with the context you use: about 1.1 GB at 8K and 4.3 GB at 32K.
Where can I download Hy MT2 7B Abliterated Bf16 GGUF?
The files are on Hugging Face in the alexwkleung/Hy-MT2-7B-Abliterated-GGUF repository. The download buttons on this page open the file on Hugging Face. ggufmodels.com does not host any model files.
What license does Hy MT2 7B Abliterated Bf16 use?
Hugging Face lists the license as apache-2.0. Licenses can differ between the quantized files and the original model, so check the repository and the base model before commercial use.
Can I run this model on 8 GB VRAM?
Yes, the Q4_K_M quantization requires about 6.2 GB of VRAM at 8‑k context, which fits in 8 GB.
Does this model support vision tasks?
No, it is a text‑only translation model with no vision capabilities.
What license applies to this model?
The model is released under the Apache‑2.0 license.