File size: 1,567 Bytes
ba16ea6 ec17444 ba16ea6 ec17444 ba16ea6 ec17444 18779b8 ba16ea6 4bac1f3 59520b3 ba16ea6 ec17444 | 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 | ---
base_model: unsloth/qwen2.5-3b-instruct-unsloth-bnb-4bit
tags:
- text-generation-inference
- transformers
- unsloth
- qwen2
- grpo
license: apache-2.0
language:
- en
datasets:
- facebook/natural_reasoning
---
# Coma 3B
Coma is based on Qwen 2.5 3B, GRPO-fine tuned on the natural reasoning data set from Meta.
## GGUF
There are quantized versions available at [theprint/Coma-3B-GGUF](https://huggingface.co/theprint/Coma-3B-GGUF) in GGUF format.
## Testing
The following system prompt was used in testing of the model:
```
Between the tags <think> and </think>: You will first think through your answer carefully step by step, including any potential risks or pit falls, the context of the user's request, and how best to present this. These are notes for yourself, so be detailed and honest in your assessment.
When you are done with the analysis, you must use your own guidelines from the previous section to construct the final response. The user will only see this section.
In summary, respond using this pattern:
<think>
First, ...
</think>
[your response]
```
Testing was done at `temperature=1.0`, `top_k=45` and `top_p=0.95`.
- **Developed by:** theprint
- **License:** apache-2.0
- **Finetuned from model :** unsloth/qwen2.5-3b-instruct-unsloth-bnb-4bit
This qwen2 model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth) |