36 lines
961 B
Markdown
36 lines
961 B
Markdown
|
---
|
||
|
|
library_name: transformers
|
||
|
|
tags:
|
||
|
|
- trl
|
||
|
|
- sft
|
||
|
|
base_model:
|
||
|
|
- HuggingFaceTB/SmolLM2-1.7B-Instruct
|
||
|
|
datasets:
|
||
|
|
- ngxson/MiniThinky-dataset
|
||
|
|
---
|
||
|
|
|
||
|
|
# MiniThinky 1.7B (based on SmolLM2)
|
||
|
|
|
||
|
|
> [!IMPORTANT]
|
||
|
|
> This checkpoint still have a high loss value, so the model will hallucinate the response quite a lot.
|
||
|
|
|
||
|
|
My first trial to fine tune a small model to add reasoning capability.
|
||
|
|
|
||
|
|
Chat template is the same with llama 3, but the response will be as follow:
|
||
|
|
|
||
|
|
```
|
||
|
|
<|thinking|>{thinking_process}
|
||
|
|
<|answer|>
|
||
|
|
{real_answer}
|
||
|
|
```
|
||
|
|
|
||
|
|
## IMPORTANT: System message
|
||
|
|
|
||
|
|
The model is **very sensitive** to system message. Make sure you're using this system message (system role) at the beginning of the conversation:
|
||
|
|
|
||
|
|
`You are MiniThinky, a helpful AI assistant. You always think before giving the answer. Use <|thinking|> before thinking and <|answer|> before giving the answer.`
|
||
|
|
|
||
|
|
---
|
||
|
|
|
||
|
|
TODO: include more info here + maybe do some benchmarks? (Plz add a discussion if you're interested)
|