---
tags:
- gguf
- llama.cpp
- unsloth
- Llama
license: mit
datasets:
- Unshifts/CodeFeedback-Alpaca
language:
- en
base_model:
- meta-llama/Llama-3.2-3B-Instruct
pipeline_tag: text-generation
---

# Llama-3B-Coder : GGUF

This model was fine tuned using 1 Billion tokens of Alpaca format code feedback (the dataset is linked). This model is the first of many, I plan to run a full epoch of this dataset soon, and update the model along with it, currently ive only done around 10% of an epoch.



**Example usage**:
- For text only LLMs:    `llama-cli -hf koshuro/Llama-3B-Coder --jinja`
- For multimodal models: `llama-mtmd-cli -hf koshuro/Llama-3B-Coder --jinja`



# Benchmarks
On basic reasoning, math, and ela benchmarks, this model scored close to its base model, and near the same score as [google/gemma-3n-e4b](https://huggingface.co/google/gemma-3n-E4B).

![compare_bar_objective](https://cdn-uploads.huggingface.co/production/uploads/683150652abe09530517e8ed/yp6mo_4d6VR2O3dye5-Bc.png)





## Available Model files:
- `llama-3.2-3b-instruct.Q5_K_M.gguf`
- `llama-3.2-3b-instruct.F16.gguf`
- `llama-3.2-3b-instruct.Q4_K_M.gguf`
- `llama-3.2-3b-instruct.Q8_0.gguf`