File size: 11,383 Bytes
f07d1df |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 |
Quantization made by Richard Erkhov.
[Github](https://github.com/RichardErkhov)
[Discord](https://discord.gg/pvy7H8DZMG)
[Request more models](https://github.com/RichardErkhov/quant_request)
rho-math-7b-v0.1 - GGUF
- Model creator: https://huggingface.co/microsoft/
- Original model: https://huggingface.co/microsoft/rho-math-7b-v0.1/
| Name | Quant method | Size |
| ---- | ---- | ---- |
| [rho-math-7b-v0.1.Q2_K.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q2_K.gguf) | Q2_K | 2.53GB |
| [rho-math-7b-v0.1.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.IQ3_XS.gguf) | IQ3_XS | 2.81GB |
| [rho-math-7b-v0.1.IQ3_S.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.IQ3_S.gguf) | IQ3_S | 2.96GB |
| [rho-math-7b-v0.1.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q3_K_S.gguf) | Q3_K_S | 2.95GB |
| [rho-math-7b-v0.1.IQ3_M.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.IQ3_M.gguf) | IQ3_M | 3.06GB |
| [rho-math-7b-v0.1.Q3_K.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q3_K.gguf) | Q3_K | 3.28GB |
| [rho-math-7b-v0.1.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q3_K_M.gguf) | Q3_K_M | 3.28GB |
| [rho-math-7b-v0.1.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q3_K_L.gguf) | Q3_K_L | 3.56GB |
| [rho-math-7b-v0.1.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.IQ4_XS.gguf) | IQ4_XS | 3.67GB |
| [rho-math-7b-v0.1.Q4_0.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q4_0.gguf) | Q4_0 | 3.83GB |
| [rho-math-7b-v0.1.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.IQ4_NL.gguf) | IQ4_NL | 3.87GB |
| [rho-math-7b-v0.1.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q4_K_S.gguf) | Q4_K_S | 3.86GB |
| [rho-math-7b-v0.1.Q4_K.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q4_K.gguf) | Q4_K | 4.07GB |
| [rho-math-7b-v0.1.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q4_K_M.gguf) | Q4_K_M | 4.07GB |
| [rho-math-7b-v0.1.Q4_1.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q4_1.gguf) | Q4_1 | 4.24GB |
| [rho-math-7b-v0.1.Q5_0.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q5_0.gguf) | Q5_0 | 4.65GB |
| [rho-math-7b-v0.1.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q5_K_S.gguf) | Q5_K_S | 4.65GB |
| [rho-math-7b-v0.1.Q5_K.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q5_K.gguf) | Q5_K | 4.78GB |
| [rho-math-7b-v0.1.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q5_K_M.gguf) | Q5_K_M | 4.78GB |
| [rho-math-7b-v0.1.Q5_1.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q5_1.gguf) | Q5_1 | 5.07GB |
| [rho-math-7b-v0.1.Q6_K.gguf](https://huggingface.co/RichardErkhov/microsoft_-_rho-math-7b-v0.1-gguf/blob/main/rho-math-7b-v0.1.Q6_K.gguf) | Q6_K | 5.53GB |
Original model description:
---
license: mit
tags:
- nlp
- math
language:
- en
pipeline_tag: text-generation
---
<h1 align="center">
Rho-1: Not All Tokens Are What You Need
</h1>
<p align="center">
<a href="https://arxiv.org/abs/2404.07965"><b>[📜 Arxiv]</b></a> •
<a href="https://huggingface.co/papers/2404.07965"><b>[💬 HF Paper]</b></a> •
<a href="https://huggingface.co/microsoft/rho-math-1b-v0.1"><b>[🤗 Models]</b></a> •
<a href="https://github.com/microsoft/rho"><b>[🐱 GitHub]</b></a>
</p>
<p align="center">
<img src="https://github.com/microsoft/rho/blob/main/docs/static/images/acc_vs_tokens_1b_7b.png?raw=true" width="1000">
<br>
<em>Figure 1: Rho-1 is pre-trained with Selective Language Modeling (SLM). SLM improves average few-shot accuracy on GSM8k and MATH by over 16%, achieving the baseline performance 5-10x faster.</em>
</p>
## 🔥 News
- [2024/04/12] 🔥🔥🔥 Rho-Math-v0.1 models released at 🤗 HuggingFace!
- [Rho-Math-1B](https://huggingface.co/microsoft/rho-math-1b-v0.1) and [Rho-Math-7B](https://huggingface.co/microsoft/rho-math-7b-v0.1) achieve 15.6% and 31.0% few-shot accuracy on MATH dataset, respectively — matching DeepSeekMath with only 3\% of the pretraining tokens.
- [Rho-Math-1B-Interpreter](https://huggingface.co/microsoft/rho-math-1b-interpreter-v0.1) is the first 1B LLM that achieves over 40% accuracy on MATH.
- [Rho-Math-7B-Interpreter](https://huggingface.co/microsoft/rho-math-7b-interpreter-v0.1) achieves 52% on MATH dataset, using only 69k samples for fine-tuning.
- [2024/04/11] Rho-1 paper and repo released.
## 💡 Introduction
Rho-1 base models employ Selective Language Modeling (SLM) for pretraining, which selectively trains on clean and useful tokens that aligned with the desired distribution.
### Selective Lanugage Modeling (SLM)
<p align="center">
<img src="https://github.com/microsoft/rho/blob/main/docs/static/images/example.png?raw=true" width="1000">
<br>
<em>Figure 2:
<b>Upper:</b> Even an extensively filtered pretraining corpus contains token-level noise.
<b>Left:</b> Previous Causal Language Modeling (CLM) trains on all tokens.
<b>Right:</b> Our proposed Selective Language Modeling (SLM) selectively applies loss on those useful and clean tokens.</em>
</p>
<p align="center">
<img src="https://github.com/microsoft/rho/blob/main/docs/static/images/pipeline.png?raw=true" width="1000">
<br>
<em>Figure 3: <b>The pipeline of Selective Language Modeling.</b>
SLM optimizes language model performance by concentrating on valuable, clean tokens during pre-training.
It involves three steps:
(Step 1) Initially, train a reference model on high-quality data.
(Step 2) Then, score each token's loss in a corpus using the reference model.
(Step 3) Finally, train the language model selectively on tokens that show higher excess loss compared to the reference loss.</em>
</p>
<!-- results: -->
### Evaluation Results
Base models (Few-shot CoT):
| **Model** | **Size** | **Data** | **Uniq. Token** | **Train Token** | **GSM8K** | **MATH** | **MMLU STEM** | **SAT** |
|:-----------------:|:--------:|:--------:|:---------------:|:---------------:|:---------:|:--------:|:-------------:|:--------:|
| 1-2B Base Models | | | | | | | | |
| Qwen1.5 | 1.8B | - | - | - | 36.1 | 6.8 | 31.3 | 40.6 |
| Gemma | 2.0B | - | - | - | 18.8 | 11.4 | **34.4** | 50.0 |
| DeepSeekMath | 1.3B | - | 120B | 150B | 23.8 | 13.6 | 33.1 | **56.3** |
| [Rho-Math-1B-v0.1](https://huggingface.co/microsoft/rho-math-1b-v0.1) | 1.1B | OWM | 14B | 30B | **36.2** | **15.6** | 23.3 | 28.1 |
| >= 7B Base Models | | | | | | | | |
| Mistral | 7B | | - | - | 41.2 | 11.6 | 49.5 | 59.4 |
| Minerva | 540B | - | 39B | 26B | 58.8 | 33.6 | **63.9** | - |
| LLemma | 34B | PPile | 55B | 50B | 54.2 | 23.0 | 54.7 | 68.8 |
| InternLM2-Math | 20B | - | 31B | 125B | 65.4 | 30.0 | 53.1 | 71.9 |
| DeepSeekMath | 7B | - | 120B | 500B | 64.1 | **34.2** | 56.4 | **84.4** |
| [Rho-Math-7B-v0.1](https://huggingface.co/microsoft/rho-math-7b-v0.1) | 7B | OWM | 14B | 10.5B | **66.9** | 31.0 | 54.6 | **84.4** |
[Tool-integrated reasoning](https://github.com/microsoft/ToRA) (Code Interpreter):
| **Model** | **Size** | **SFT Data** | **GSM8k** | **MATH** | **SVAMP** | **ASDiv** | **MAWPS** | **TabMWP** | **GSM-Hard** | **AVG** |
|------------------------------|----------|--------------|-----------|----------|-----------|-----------|-----------|------------|--------------|----------|
| gpt4-early (pal) | - | - | 94.2 | 51.8 | 94.8 | 92.6 | 97.7 | 95.9 | 77.6 | 86.4 |
| gpt-4-turbo-2024-04-09 (cot) | - | - | - | 73.4 | - | - | - | - | - |
| Open-Source Small Models | | | | | | | | | |
| MAmmoTH | 70B | MI-260k | 76.9 | 41.8 | 82.4 | - | - | - | - | - |
| ToRA | 7B | ToRA-69k | 68.8 | 40.1 | 68.2 | 73.9 | 88.8 | 42.4 | 54.6 | 62.4 |
| ToRA | 70B | ToRA-69k | 84.3 | 49.7 | **82.7** | 86.8 | 93.8 | 74.0 | **67.2** | **76.9** |
| DeepSeekMath | 7B | ToRA-69k | 79.8 | **52.0** | 80.1 | **87.1** | 93.8 | **85.8** | 63.1 | 77.4 |
| [Rho-Math-1B-Interpreter-v0.1](https://huggingface.co/microsoft/rho-math-1b-interpreter-v0.1) | 1B | ToRA-69k | 59.4 | 40.6 | 60.7 | 74.2 | 88.6 | 26.7 | 48.1 | 56.9 |
| [Rho-Math-7B-Interpreter-v0.1](https://huggingface.co/microsoft/rho-math-7b-interpreter-v0.1) | 7B | ToRA-69k | 81.3 | **51.8** | 80.8 | 85.5 | **94.5** | 70.1 | 63.1 | 75.3 |
## 🚀 Quick Start
### Evaluation
```sh
git clone git@github.com:microsoft/rho.git
cd rho-1/math-evaluation-harness
```
Base model few-shot evaluation:
```sh
bash scripts/run_eval.sh cot microsoft/rho-math-7b-v0.1
```
SFT model (code-interpreter) evaluation:
```sh
bash scripts/run_eval.sh tora microsoft/rho-math-7b-interpreter-v0.1
```
Our reproduced outputs are provided in `rho-1/outputs.zip`.
## ☕️ Citation
If you find this repository helpful, please consider citing our paper:
```
@misc{lin2024rho1,
title={Rho-1: Not All Tokens Are What You Need},
author={Zhenghao Lin and Zhibin Gou and Yeyun Gong and Xiao Liu and Yelong Shen and Ruochen Xu and Chen Lin and Yujiu Yang and Jian Jiao and Nan Duan and Weizhu Chen},
year={2024},
eprint={2404.07965},
archivePrefix={arXiv},
primaryClass={cs.CL}
}
```
|