---
base_model: Qwen/Qwen2.5-1.5B
tags:
  - reasoning
  - chain-of-thought
  - distillation
  - s1
  - limo
  - qlora
  - text-generation
datasets:
  - simplescaling/s1K
  - GAIR/LIMO
pipeline_tag: text-generation
language:
  - en
license: apache-2.0
library_name: transformers
---

![Reasoning Rob](reasoning-rob.png)

# Reasoning Rob

*A Qwen2.5-1.5B base model fine-tuned to reason with chain-of-thought traces from s1K + LIMO.*

[![Model Size](https://img.shields.io/badge/params-1.5B-blue)]()
[![Base Model](https://img.shields.io/badge/base-Qwen2.5--1.5B-green)](https://huggingface.co/Qwen/Qwen2.5-1.5B)
[![License](https://img.shields.io/badge/license-Apache--2.0-orange)]()
[![Downloads](https://img.shields.io/huggingface/downloads/dustarrr/reasoning-rob)](https://huggingface.co/dustarrr/reasoning-rob)
[![Likes](https://img.shields.io/huggingface/likes/dustarrr/reasoning-rob)](https://huggingface.co/dustarrr/reasoning-rob)

---

## Summary

| | |
|---|---|
| **Base model** | [`Qwen/Qwen2.5-1.5B`](https://huggingface.co/Qwen/Qwen2.5-1.5B) |
| **Parameters** | ~1.5B (LoRA r=16, merged) |
| **Context length** | 2048 tokens |
| **Training data** | s1K (1,000 traces) + LIMO (817 traces) = ~1,800 CoT samples |
| **Method** | s1-style distillation + budget forcing via QLoRA SFT |
| **Compute** | Google Colab T4 GPU, ~16 min |
| **Special tokens** | `<think>` `</think>` for reasoning trace delimiters |

---

## Evaluation Results

| Benchmark | Reasoning Rob |
|-----------|--------------:|
| GSM8K (50 samples) | 10.00% |

---

## Usage

```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained(
    "dustarrr/reasoning-rob",
    torch_dtype=torch.float16,
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained("dustarrr/reasoning-rob")
model.eval()

messages = [
    {"role": "system", "content": "You are a helpful assistant that thinks step by step."},
    {"role": "user", "content": "If a train travels 60 km in 1.5 hours, what is its speed?"},
]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt")

with torch.no_grad():
    outputs = model.generate(**inputs, max_new_tokens=1024, do_sample=False)

response = tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=False)
print(response)
```

### Budget Forcing (s1-style)

Extend the model's thinking phase by injecting `"Wait"` before the `</think>` token
to force longer reasoning before the final answer. This is the test-time scaling
trick from the [s1 paper](https://arxiv.org/abs/2501.19393).

---

## Training Details

| Hyperparameter | Value |
|---|---|
| LoRA rank | 16 |
| LoRA alpha | 32 |
| LoRA dropout | 0.05 |
| Learning rate | 0.0001 |
| LR scheduler | cosine |
| Warmup ratio | 0.03 |
| Weight decay | 0.01 |
| Batch size | 2 |
| Gradient accumulation | 8 |
| Max sequence length | 2048 |
| Epochs | 1 |
| Quantization | NF4 (4-bit, double quant) |
| Optimizer | adamw_torch |

---

## Attribution

Reasoning Rob is a QLoRA fine-tune of **[Qwen/Qwen2.5-1.5B](https://huggingface.co/Qwen/Qwen2.5-1.5B)**
(base, not instruct) trained on:

- **[s1K](https://huggingface.co/datasets/simplescaling/s1K)** - 1,000 curated reasoning traces
- **[LIMO](https://huggingface.co/datasets/GAIR/LIMO)** - 817 "Less Is More" reasoning traces

Using the **[s1](https://arxiv.org/abs/2501.19393)** distillation + budget-forcing method
and **[LIMO](https://arxiv.org/abs/2502.03387)** "less is more" reasoning transfer approach.

All credit to:
- The **Qwen Team** (Alibaba) for the base model
- The **s1 authors** (Stanford) for the training methodology and dataset
- The **LIMO authors** (GAIR) for the reasoning dataset

This model would not exist without their work.

---

## Limitations

- **Small model**: At 1.5B parameters, Reasoning Rob has limited capacity.
- **Hallucination**: The model may still produce incorrect reasoning or fabricate facts.
- **Short context**: Max sequence length is 2048 tokens.
- **English only**: Training data is predominantly English.

---

## License

Apache 2.0 (inherited from Qwen2.5 base model).

---

*Generated on 2026-06-23*