50 lines
1.9 KiB
Markdown
50 lines
1.9 KiB
Markdown
|
---
|
||
|
|
license: mit
|
||
|
|
language:
|
||
|
|
- en
|
||
|
|
base_model:
|
||
|
|
- WeiboAI/VibeThinker-1.5B
|
||
|
|
tags:
|
||
|
|
- gguf
|
||
|
|
- llama.cpp
|
||
|
|
- unsloth
|
||
|
|
- math
|
||
|
|
- code
|
||
|
|
- gpqa
|
||
|
|
- reasoning
|
||
|
|
- python
|
||
|
|
pipeline_tag: text-generation
|
||
|
|
library_name: transformers
|
||
|
|
---
|
||
|
|
|
||
|
|
# PyThink-1.5B v0.0.1 : safetensors
|
||
|
|
|
||
|
|
<blockquote style="border-left: 4px solid #ff6b6b; background-color: #fff5f5; padding: 10px 15px; margin: 10px 0; color: #cc3333;">
|
||
|
|
<span style="font-weight: bold;">🚨</span> This model was built to output "reasoning plans" in Python, and does not always output normal responses. This model is intended for research into alternative ways to make LLMs do structured reasoning.
|
||
|
|
</blockquote>
|
||
|
|
|
||
|
|
This model was made to generate more training data and to start experimenting with LLM reasoning in code. Code executes faster and with less compute than LLMs, is deterministic, and is easier to audit. This small model can run on my laptop, is blazing fast, and is already close to good for generating new training data.
|
||
|
|
|
||
|
|
I will be releasing more versions soon.
|
||
|
|
|
||
|
|
- Training dataset: https://huggingface.co/datasets/NuclearManD/pythink-20k
|
||
|
|
|
||
|
|
- GGUF 4bit quantized version: https://huggingface.co/NuclearManD/pythink-qwen2-1.5b-v0.0.1-Q4_K_M-GGUF
|
||
|
|
|
||
|
|
Follow me on X for updates: https://x.com/NuclearManD
|
||
|
|
|
||
|
|
This model was finetuned and converted using [Unsloth](https://github.com/unslothai/unsloth).
|
||
|
|
|
||
|
|
**Example usage**:
|
||
|
|
- For text only LLMs: `llama-cli -hf NuclearManD/pythink-qwen2-1.5b-v0.0.1-Q4_K_M-GGUF --jinja`
|
||
|
|
|
||
|
|
## Available Model files:
|
||
|
|
- `checkpoint-360.Q4_K_M.gguf`
|
||
|
|
This was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth)
|
||
|
|
[<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)
|
||
|
|
|
||
|
|
Fine-tuned from (WeiboAI/VibeThinker-1.5B)[https://huggingface.co/WeiboAI/VibeThinker-1.5B].
|
||
|
|
|
||
|
|
## License
|
||
|
|
The model repository is licensed under the MIT License.
|