File size: 1,522 Bytes
1e260b9 c3fb9c1 1e260b9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 | ---
pipeline_tag: text-generation
license: other
license_name: modified-mit
license_link: https://github.com/MiniMax-AI/MiniMax-M2.1/blob/main/LICENSE
base_model:
- MiniMaxAI/MiniMax-M2.1
tags:
- smoothie-qwen
---
# Smoothie-MiniMax-M2.1
## Overview
This is a modified version of [MiniMax-M2.1](https://huggingface.co/MiniMaxAI/MiniMax-M2.1), using [Smoothie-Qwen](https://github.com/dnotitia/smoothie-qwen).
## What is it?
Reduced probability of Kanji, Hanja, Chinese character(radical) tokens to reduce sudden language mixing.
## For who?
If you see Chinese characters during non-Chinese conversation, this model will **help** in this case.
It does not *"solve"* the main problem, just improve its occurrence.
**For Chinese and Japanese users: Use original model!** This model will behave worse in these languages.
## Result
From my testing:
* Chinese character did not appear on Korean conversation.
* When I ask about Japanese topic, model sucessfully answered with Kanji and Hiragana (although I can't test correctness of response)
## How I did it?
I tried to replicate Unsloth's UD quant as possible because my system only can handle up to 3-bit quants.
1. Download original model
2. Apply Smoothie-qwen (See configs/config.yaml for reference)
3. Convert to GGUF (BF16)
4. Run llama-quantize with Unsloth imatrix and manual override to tensor type from UD quants
5. Run llama-gguf-split (max size 50GB)
## Recommendation
At temperature 1.0, tool calling is bit unstable. I recommend temperature=0.7. |