Update README.md

README.md

@@ -1,199 +1,202 @@
 ---
+language:
+- en
+license: other
+license_name: deepseek
+license_link: https://github.com/deepseek-ai/DeepSeek-Math/blob/main/LICENSE-MODEL
 library_name: transformers
-tags: []
+tags:
+- mathematics
+datasets:
+- hkust-nlp/dart-math-hard
+metrics:
+- accuracy
+pipeline_tag: text-generation
+base_model: deepseek-ai/deepseek-math-7b-base
+model-index:
+  - name: dart-math-dsmath-7b-prop2diff
+    results:
+      - task:
+          type: text-generation
+          name: Mathematical Problem-Solving
+        dataset:
+          type: hendrycks/competition_math
+          name: MATH
+          split: test
+        metrics:
+          - type: accuracy
+            name: Pass@1 (0-shot CoT)
+            value: 53.6
+        source:
+          name: "Paper *DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving*"
+          url: "https://arxiv.org/abs/2407.13690"
+      - task:
+          type: text-generation
+          name: Mathematical Problem-Solving
+        dataset:
+          type: openai/gsm8k
+          name: GSM8K
+          config: main
+          split: test
+        metrics:
+          - type: accuracy
+            name: Pass@1 (0-shot CoT)
+            value: 86.8
+        source:
+          name: "Paper *DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving*"
+          url: "https://arxiv.org/abs/2407.13690"
+      - task:
+          type: text-generation
+          name: Mathematical Problem-Solving
+        dataset:
+          type: college-math
+          name: CollegeMath
+        metrics:
+          - type: accuracy
+            name: Pass@1 (0-shot CoT)
+            value: 40.7
+        source:
+          name: "Paper *DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving*"
+          url: "https://arxiv.org/abs/2407.13690"
+      - task:
+          type: text-generation
+          name: Mathematical Problem-Solving
+        dataset:
+          type: deepmind-mathematics
+          name: DeepMind-Mathematics
+        metrics:
+          - type: accuracy
+            name: Pass@1 (0-shot CoT)
+            value: 61.6
+        source:
+          name: "Paper *DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving*"
+          url: "https://arxiv.org/abs/2407.13690"
+      - task:
+          type: text-generation
+          name: Mathematical Problem-Solving
+        dataset:
+          type: Hothan/OlympiadBench
+          name: OlympiadBench-OE_TO_maths_en_COMP
+          config: OE_TO_maths_en_COMP
+          split: train
+        metrics:
+          - type: accuracy
+            name: Pass@1 (0-shot CoT)
+            value: 21.7
+        source:
+          name: "Paper *DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving*"
+          url: "https://arxiv.org/abs/2407.13690"
+      - task:
+          type: text-generation
+          name: Mathematical Problem-Solving
+        dataset:
+          type: TIGER-Lab/TheoremQA
+          name: TheoremQA
+          split: test
+        metrics:
+          - type: accuracy
+            name: Pass@1 (0-shot CoT)
+            value: 32.2
+        source:
+          name: "Paper *DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving*"
+          url: "https://arxiv.org/abs/2407.13690"
 ---
 
-# Model Card for Model ID
+# DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving
 
-<!-- Provide a quick summary of what the model is/does. -->
+📝 [Paper@arXiv](https://arxiv.org/abs/2407.13690) | 🤗 [Datasets&Models@HF](https://huggingface.co/collections/hkust-nlp/dart-math-665704599b35de59f8fdf6c1) | 🐱 [Code@GitHub](https://github.com/hkust-nlp/dart-math) | 🐦 [Thread@X(Twitter)](https://x.com/tongyx361/status/1811413243350454455) | 🐶 [中文博客@知乎](https://zhuanlan.zhihu.com/p/708371895) | 📑 [BibTeX](https://github.com/hkust-nlp/dart-math?tab=readme-ov-file#citation)
 
+## Models: `DART-Math`
 
+`DART-Math` models achieve performance **superior or competitive to previous SOTAs** on 2 in-domain and 4 challenging out-of-domain mathematical reasoning benchmarks, despite using **much smaller datasets** and **no proprietary model like GPT-4**.
 
-## Model Details
+| Model                                                                                                  | [MATH](https://huggingface.co/datasets/hendrycks/competition_math) | [GSM8K](https://huggingface.co/datasets/gsm8k) | [College](https://github.com/hkust-nlp/dart-math/tree/main/data/eval-dsets/mwpbench/college-math-test.jsonl) | [DM](https://github.com/hkust-nlp/dart-math/tree/main/data/eval-dsets/deepmind-mathematics.json) | [Olympiad](https://github.com/hkust-nlp/dart-math/tree/main/data/eval-dsets/olympiadbench/OE_TO_maths_en_COMP.json) | [Theorem](https://github.com/hkust-nlp/dart-math/tree/main/data/eval-dsets/theoremqa.json) |      AVG |
+| :----------------------------------------------------------------------------------------------------- | -----------------------------------------------------------------: | ---------------------------------------------: | -----------------------------------------------------------------------------------------------------------: | -----------------------------------------------------------------------------------------------: | ------------------------------------------------------------------------------------------------------------------: | -----------------------------------------------------------------------------------------: | -------: |
+| GPT-4 (0314)                                                                                           |                           [52.6](https://arxiv.org/abs/2403.04706) |       [94.7](https://arxiv.org/abs/2403.04706) |                                                                     [24.4](https://arxiv.org/abs/2403.02884) |                                                                                               -- |                                                                                                                  -- |                                                                                         -- |       -- |
+| Llama-3-70B-MetaMath                                                                                   |                                                               44.9 |                                           88.0 |                                                                                                         31.9 |                                                                                             53.2 |                                                                                                                11.6 |                                                                                       21.9 |     41.9 |
+| [`DART-Math-Llama-3-70B` (Uniform)](https://huggingface.co/hkust-nlp/dart-math-llama3-70b-uniform)     |                                                               54.9 |                                       **90.4** |                                                                                                     **38.5** |                                                                                         **64.1** |                                                                                                                19.1 |                                                                                       27.4 |     49.1 |
+| [`DART-Math-Llama-3-70B` (Prop2Diff)](https://huggingface.co/hkust-nlp/dart-math-llama3-70b-prop2diff) |                                                           **56.1** |                                           89.6 |                                                                                                         37.9 |                                                                                         **64.1** |                                                                                                            **20.0** |                                                                                   **28.2** | **49.3** |
+| DeepSeekMath-7B-MetaMath                                                                               |                                                               43.7 |                                           81.8 |                                                                                                         33.7 |                                                                                             53.0 |                                                                                                                13.6 |                                                                                       23.2 |     41.5 |
+| [DeepSeekMath-7B-RL](https://huggingface.co/deepseek-ai/deepseek-math-7b-rl)                           |                                                               53.1 |                                           88.4 |                                                                                                         41.3 |                                                                                             58.3 |                                                                                                                18.7 |                                                                                       35.9 |     49.3 |
+| [`DART-Math-DSMath-7B` (Uniform)](https://huggingface.co/hkust-nlp/dart-math-dsmath-7b-uniform)        |                                                               52.9 |                                       **88.2** |                                                                                                         40.1 |                                                                                             60.2 |                                                                                                                21.3 |                                                                                   **32.5** |     49.2 |
+| [`DART-Math-DSMath-7B` (Prop2Diff)](https://huggingface.co/hkust-nlp/dart-math-dsmath-7b-prop2diff)    |                                                           **53.6** |                                           86.8 |                                                                                                     **40.7** |                                                                                         **61.6** |                                                                                                            **21.7** |                                                                                       32.2 | **49.4** |
+| Mistral-7B-MetaMath                                                                                    |                                                               29.8 |                                           76.5 |                                                                                                         19.3 |                                                                                             28.0 |                                                                                                                 5.9 |                                                                                       14.0 |     28.9 |
+| [`DART-Math-Mistral-7B` (Uniform)](https://huggingface.co/hkust-nlp/dart-math-mistral-7b-uniform)      |                                                               43.5 |                                       **82.6** |                                                                                                         26.9 |                                                                                             42.0 |                                                                                                                13.2 |                                                                                       16.4 |     27.4 |
+| [`DART-Math-Mistral-7B` (Prop2Diff)](https://huggingface.co/hkust-nlp/dart-math-mistral-7b-prop2diff)  |                                                           **45.5** |                                           81.1 |                                                                                                     **29.4** |                                                                                         **45.1** |                                                                                                            **14.7** |                                                                                   **17.0** | **38.8** |
+| Llama-3-8B-MetaMath                                                                                    |                                                               32.5 |                                           77.3 |                                                                                                         20.6 |                                                                                             35.0 |                                                                                                                 5.5 |                                                                                       13.8 |     30.8 |
+| [`DART-Math-Llama-3-8B` (Uniform)](https://huggingface.co/hkust-nlp/dart-math-llama3-8b-uniform)       |                                                               45.3 |                                       **82.5** |                                                                                                         27.1 |                                                                                         **48.2** |                                                                                                                13.6 |                                                                                       15.4 |     38.7 |
+| [`DART-Math-Llama-3-8B` (Prop2Diff)](https://huggingface.co/hkust-nlp/dart-math-llama3-8b-prop2diff)   |                                                           **46.6** |                                           81.1 |                                                                                                     **28.8** |                                                                                             48.0 |                                                                                                            **14.5** |                                                                                   **19.4** | **39.7** |
 
-### Model Description
+***Abbreviations**: College (CollegeMath), DM (DeepMind Mathematics), Olympiad (OlympiadBench-Math), Theorem (TheoremQA).
+**Bold** means the best score by SFT on the respective base model here.
+To reproduce our results, please refer to [the `DART-Math` GitHub repository](https://github.com/hkust-nlp/dart-math).*
 
-<!-- Provide a longer summary of what this model is. -->
+## Prompt Template
 
-This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
+All the `DART-Math` models use the [Alpaca](https://github.com/tatsu-lab/stanford_alpaca) prompt template:
 
-- **Developed by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Model type:** [More Information Needed]
-- **Language(s) (NLP):** [More Information Needed]
-- **License:** [More Information Needed]
-- **Finetuned from model [optional]:** [More Information Needed]
+```
 
-### Model Sources [optional]
+Below is an instruction that describes a task. Write a response that appropriately completes the request.\n\n###Instruction:\n{query}\n\n### Response:\n
 
-<!-- Provide the basic links for the model. -->
+```
 
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
+## Training Dataset
 
-## Uses
+We construct our traning datasets by applying **Difficulty-Aware Rejection Sampling** (`DARS`) to the **MATH and GSM8K** training sets.
 
-<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+`DARS` tackle **severe biases towards easy queries, with frequent failures to generate any correct response for the most challenging queries**, in previous datasets.
 
-### Direct Use
+These biases are primarily caused by vanilla rejection sampling, where **the same number of responses is
+sampled for each query**, yet the likelihood of obtaining correct responses for difficult queries is significantly lower, sometimes even zero.
 
-<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+Please refer to [`DART-Math-Hard`](https://huggingface.co/datasets/hkust-nlp/dart-math-hard) / [`DART-Math-Uniform`](https://huggingface.co/datasets/hkust-nlp/dart-math-uniform) for more details.
 
-[More Information Needed]
+## Training Setup
 
-### Downstream Use [optional]
+We perform standard instruction tuning to several base models including Llama3-8B & Mistral-7B & Llama3-70B as representatives of general models and DeepSeekMath-
+7B as the representative of math-specialized model
+on our synthetic datasets [`DART-Math-Hard`](https://huggingface.co/datasets/hkust-nlp/dart-math-hard) & [`DART-Math-Uniform`](https://huggingface.co/datasets/hkust-nlp/dart-math-uniform),
+leading to `DART-Math (Prop2Diff)` & `DART-Math (Uniform)` respectively.
 
-<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+For simplicity, we keep most hyper-parameters the same across different models and datasets:
 
-[More Information Needed]
+- Model max length (of [packed](https://github.com/MeetKai/functionary/tree/main/functionary/train/packing) sequence): 4096
+- Batch size: 64
+- Warm-up ratio: 0.03
+- Learning rate scheduler: cosine
+- Prompt template: [Alpaca](https://github.com/tatsu-lab/stanford_alpaca)
 
-### Out-of-Scope Use
+Several other key hyper-parameters are tuned as follow:
 
-<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+| Base Model      | Max. L.R. | # of Epochs | # of Grad. Acc. Steps | # of A100 GPUs |
+|:--------------- | ---------:| -----------:| ---------------------:| --------------:|
+| Mistral-7B      |    `1e-5` |           3 |                     1 |              8 |
+| Llama3-8B       |    `5e-5` |           1 |                     2 |              8 |
+| Llama3-70B      |    `2e-5` |           1 |                     1 |             32 |
+| DeepSeekMath-7B |    `5e-5` |           3 |                     1 |              8 |
 
-[More Information Needed]
+- For **maximum learning rate**, we determine the values by **searching** through `1e-6,5e-6,1e-5,2e-5,5e-5,1e-4` according to the MATH performance after training on MMIQC for 1 epoch, except for Llama3-70B that is so expensive to search for that we derive from Llama3-8B’s learning rate in analogy to the relationship of (per-training) learning rates between [Llama2-7B](https://huggingface.co/meta-llama/Llama-2-7b-hf) and [Llama2-70B](https://huggingface.co/meta-llama/Llama-2-70b-hf) (\~2:1).
+- For **Llama3** models, preliminary experiments indicate that **training for 1 epoch consistently outperforms 3 epochs**.
 
-## Bias, Risks, and Limitations
+Please refer to [Appendix A.1 of our paper](https://tongyx361.github.io/assets/dart-math/paper-dart-math.pdf) for more details.
 
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+## Other Details
 
-[More Information Needed]
+- For Mistral-7B-based models, we disable `sliding_window` by default following [the newest Mistral-7B-Instruct](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.3/blob/main/config.json) (Flash Attention 2 does not support `sliding_window` and XFormer backend in vLLM has throughput \~10% lower in our experiments.)
 
-### Recommendations
+## Citation
 
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+If you find our data, model or code useful for your work, please kindly cite [our paper](https://arxiv.org/abs/2407.13690):
 
-Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
-
-## How to Get Started with the Model
-
-Use the code below to get started with the model.
-
-[More Information Needed]
-
-## Training Details
-
-### Training Data
-
-<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
-
-[More Information Needed]
-
-### Training Procedure
-
-<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
-
-#### Preprocessing [optional]
-
-[More Information Needed]
-
-
-#### Training Hyperparameters
-
-- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
-
-#### Speeds, Sizes, Times [optional]
-
-<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
-
-[More Information Needed]
-
-## Evaluation
-
-<!-- This section describes the evaluation protocols and provides the results. -->
-
-### Testing Data, Factors & Metrics
-
-#### Testing Data
-
-<!-- This should link to a Dataset Card if possible. -->
-
-[More Information Needed]
-
-#### Factors
-
-<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
-
-[More Information Needed]
-
-#### Metrics
-
-<!-- These are the evaluation metrics being used, ideally with a description of why. -->
-
-[More Information Needed]
-
-### Results
-
-[More Information Needed]
-
-#### Summary
-
-
-
-## Model Examination [optional]
-
-<!-- Relevant interpretability work for the model goes here -->
-
-[More Information Needed]
-
-## Environmental Impact
-
-<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
-
-Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
-
-- **Hardware Type:** [More Information Needed]
-- **Hours used:** [More Information Needed]
-- **Cloud Provider:** [More Information Needed]
-- **Compute Region:** [More Information Needed]
-- **Carbon Emitted:** [More Information Needed]
-
-## Technical Specifications [optional]
-
-### Model Architecture and Objective
-
-[More Information Needed]
-
-### Compute Infrastructure
-
-[More Information Needed]
-
-#### Hardware
-
-[More Information Needed]
-
-#### Software
-
-[More Information Needed]
-
-## Citation [optional]
-
-<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
-
-**BibTeX:**
-
-[More Information Needed]
-
-**APA:**
-
-[More Information Needed]
-
-## Glossary [optional]
-
-<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
-
-[More Information Needed]
-
-## More Information [optional]
-
-[More Information Needed]
-
-## Model Card Authors [optional]
-
-[More Information Needed]
-
-## Model Card Contact
-
-[More Information Needed]
+```latex
+@article{tong2024dartmath,
+  title={DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving},
+  author={Yuxuan Tong and Xiwen Zhang and Rui Wang and Ruidong Wu and Junxian He},
+  year={2024},
+  eprint={2407.13690},
+  archivePrefix={arXiv},
+  primaryClass={cs.CL},
+  url={https://arxiv.org/abs/2407.13690},
+}
+```