README.md

---
tags:
- gguf
- llama.cpp
- agent
- tool-use
base_model: WeiboAI/VibeThinker-3B
pipeline_tag: reinforcement-learning
---

# mythos_fast : GGUF

## Model Description

mythos_fast is a fine-tuned version of [WeiboAI/VibeThinker-3B](https://huggingface.co/WeiboAI/VibeThinker-3B), adapted for custom tool-use and agentic task execution. The base model was trained on a rich dataset of approximately 2 million samples covering multi-step tool calls, function-calling formats, and agent-style reasoning traces, then converted to GGUF format for efficient local inference with llama.cpp.

## Training Details

- **Base model:** WeiboAI/VibeThinker-3B
- **Fine-tuning focus:** tool-use / function calling, agentic task completion
- **Dataset size:** ~2,000,000 samples
- **Output format:** GGUF (F16)

## Available Model Files

- `VibeThinker-3B.F16.gguf`

## Usage

For text-only LLMs:
```
llama-cli -hf Shadow0482/mythos_fast --jinja
```

For multimodal models:
```
llama-mtmd-cli -hf Shadow0482/mythos_fast --jinja
```

## Intended Use

This model is intended for local inference scenarios that require tool-calling and agent-style task execution, such as autonomous agents, function-calling pipelines, and multi-step reasoning workflows.

## Limitations

Performance on tool-use tasks depends on the format and structure of tool definitions provided at inference time. Results may vary outside the training distribution covered by the fine-tuning dataset.