Update docs

Author: tongyx361

Diff for: .pre-commit-config.yaml

@@ -0,0 +1,6 @@
+repos:
+  - repo: https://github.com/fastai/nbdev
+    rev: 2.2.10
+    hooks:
+      - id: nbdev_clean
+      - id: nbdev_export

Diff for: LICENSE

@@ -1,4 +1,4 @@
-Copyright (c) 2012-2024 Scott Chacon and others
+Copyright (c) 2024 Yuxuan Tong
 
 Permission is hereby granted, free of charge, to any person obtaining
 a copy of this software and associated documentation files (the

Diff for: README.md

@@ -170,6 +170,7 @@ For intended contributors, we recommend installing the package with the
 
 ``` shell
 pip install -e ".[dev]"
+pre-commit install
 ```
 
 ### 🔨 Training
@@ -207,10 +208,7 @@ node of 8 A100 GPUs, please run the following command:
 bash scripts/train-single-node.sh \
     --data_path "hkust-nlp/dart-math-hard" \
     --model_path "meta-llama/Meta-Llama-3-8B" \
-    --lr "5e-5" \
-    --bs 64 \
-    --n_grad_acc_steps 1 \
-    --n_epochs 1 \
+    --lr "5e-5" --bs 64 --n_grad_acc_steps 1 --n_epochs 1 \
     --gpu_ids "0,1,2,3,4,5,6,7" \
     --output_dir "models/dart-math-llama3-8b-prop2diff"
 ```
@@ -229,10 +227,7 @@ your enviroment and then run the following command:
 bash scripts/train-multi-node.sh \
     --data_path "hkust-nlp/dart-math-hard" \
     --model_path "meta-llama/Meta-Llama-3-70B" \
-    --lr "2e-5" \
-    --bs 64 \
-    --n_grad_acc_steps 1 \
-    --n_epochs 1 \
+    --lr "2e-5" --bs 64 --n_grad_acc_steps 1 --n_epochs 1 \
     --n_nodes 4 \
     --output_dir "models/dart-math-llama3-70b-prop2diff"
 ```
@@ -249,10 +244,7 @@ The off-the-shelf command to train `DART-Math-Llama3-70B-Uniform`
 bash scripts/train-multi-node.sh \
     --data_path "hkust-nlp/dart-math-uniform" \
     --model_path "meta-llama/Meta-Llama-3-70B" \
-    --lr "2e-5" \
-    --bs 64 \
-    --n_grad_acc_steps 1 \
-    --n_epochs 1 \
+    --lr "2e-5" --bs 64 --n_grad_acc_steps 1 --n_epochs 1 \
     --n_nodes 4 \
     --output_dir "models/dart-math-llama3-70b-prop2diff"
 ```
@@ -277,17 +269,14 @@ For example, to reproduce one pass of greedy decoding with
 0, please run the following command:
 
 ``` shell
-CUDA_VISIBLE_DEVICES="0" RAY_DEDUP_LOGS=0 python pipeline/gen.py \
+CUDA_VISIBLE_DEVICES="0" python pipeline/gen.py \
     --gen_save_path "data/res/dart-math-mistral-7b-prop2diff.jsonl" \
     --model_name_or_path "hkust-nlp/dart-math-mistral-7b-prop2diff" \
     --datasets "math-test" "gsm8k-test" "mwpbench/college-math-test" "deepmind-mathematics" \
         "olympiadbench/OE_TO_maths_en_COMP" "theoremqa" \
-    --max_new_tokens 2048 \
-    --temperature 0 \
-    --prompt_template "auto" \
-    --n_shots -1 \
+    --max_new_tokens 2048 --temperature 0 --top_p 0.95 \
+    --prompt_template "cot" --n_shots -1 \
     --inf_seed -1 \
-    --do_eval \
     --max_n_trials 1
 ```
 
@@ -311,19 +300,14 @@ For example, to reproduce synthesis of `DART-Math-Uniform`, run the
 following command with different GPUs, please run the following command:
 
 ``` shell
-CUDA_VISIBLE_DEVICES="0" RAY_DEDUP_LOGS=0 python pipeline/gen.py \
+CUDA_VISIBLE_DEVICES="0" python pipeline/gen.py \
     --gen_save_path "data/res/dart-math-uniform.jsonl" \
     --model_name_or_path "deepseek-ai/deepseek-math-7b-rl" \
     --datasets "math-train" "gsm8k-train" \
-    --max_new_tokens 2048 \
-    --temperature 1.6 \
-    --top_p 0.95 \
-    --prompt_template "auto" \
-    --n_shots 0 \
+    --max_new_tokens 2048 --temperature 1.6 --top_p 0.95 \
+    --prompt_template "cot" --n_shots 0 \
     --inf_seed -1 \
-    --do_eval \
-    --min_n_corrects 40 \
-    --max_n_trials 0 # unlimited, should be killed manually
+    --min_n_corrects 40 --max_n_trials 0 # unlimited, should be killed manually
 ```
 
 ## [`dart-math` Package](https://hkust-nlp.github.io/dart-math): Efficient and Flexible Training & Inference & Evaluation Pipelines
@@ -336,7 +320,7 @@ website](https://hkust-nlp.github.io/dart-math/quick-start.html).
 
 Thanks to:
 
-- [`nbdev`](https://nbdev.fast.ai/) for genrating the [wonderful
+- [`nbdev`](https://nbdev.fast.ai/) for generating the [wonderful
   documentation website](https://hkust-nlp.github.io/dart-math),
 - [`stanford_alpaca`](https://github.com/tatsu-lab/stanford_alpaca) for
   reference code about training,
@@ -352,7 +336,7 @@ paper](https://tongyx361.github.io/assets/dart-math/paper-dart-math.pdf):
 
 ``` latex
 @article{tong2024dartmath,
-  author = {Yuxuan Tong, Xiwen Zhang, Rui Wang, Ruidong Wu, Junxian He },
+  author = {Yuxuan Tong, Xiwen Zhang, Rui Wang, Ruidong Wu, Junxian He},
   title = {DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving},
   year = {2024},
   publisher = {GitHub},

Diff for: nbs/00_quick-start.ipynb

@@ -40,6 +40,7 @@
     "\n",
     "```shell\n",
     "pip install -e \".[dev]\"\n",
+    "pre-commit install\n",
     "```\n"
    ]
   },
@@ -67,7 +68,7 @@
     "from dart_math.train import monkey_patch4pack, make_supervised_dset\n",
     "# ...\n",
     "monkey_patch4pack(model)\n",
-    "pack_dset = make_supervised_dset(tokenizer=tokenizer, data_path=data_args.data_path, pack_len=training_args.model_max_length, query_field=data_args.query_field,, resp_field=data_args.resp_field,, prompt_template=data_args.prompt_template)\n",
+    "pack_dset = make_supervised_dset(tokenizer=tokenizer, data_path=data_args.data_path, pack_len=training_args.model_max_length, query_field=data_args.query_field,, resp_field=data_args.resp_field, prompt_template=data_args.prompt_template)\n",
     "trainer = Trainer(model=model, tokenizer=tokenizer, train_dataset=pack_dset)\n",
     "# ...\n",
     "```\n",

Diff for: nbs/index.ipynb

@@ -206,6 +206,7 @@
     "\n",
     "```shell\n",
     "pip install -e \".[dev]\"\n",
+    "pre-commit install\n",
     "```\n"
    ]
   },
@@ -247,10 +248,7 @@
     "bash scripts/train-single-node.sh \\\n",
     "    --data_path \"hkust-nlp/dart-math-hard\" \\\n",
     "    --model_path \"meta-llama/Meta-Llama-3-8B\" \\\n",
-    "    --lr \"5e-5\" \\\n",
-    "    --bs 64 \\\n",
-    "    --n_grad_acc_steps 1 \\\n",
-    "    --n_epochs 1 \\\n",
+    "    --lr \"5e-5\" --bs 64 --n_grad_acc_steps 1 --n_epochs 1 \\\n",
     "    --gpu_ids \"0,1,2,3,4,5,6,7\" \\\n",
     "    --output_dir \"models/dart-math-llama3-8b-prop2diff\"\n",
     "```\n",
@@ -275,10 +273,7 @@
     "bash scripts/train-multi-node.sh \\\n",
     "    --data_path \"hkust-nlp/dart-math-hard\" \\\n",
     "    --model_path \"meta-llama/Meta-Llama-3-70B\" \\\n",
-    "    --lr \"2e-5\" \\\n",
-    "    --bs 64 \\\n",
-    "    --n_grad_acc_steps 1 \\\n",
-    "    --n_epochs 1 \\\n",
+    "    --lr \"2e-5\" --bs 64 --n_grad_acc_steps 1 --n_epochs 1 \\\n",
     "    --n_nodes 4 \\\n",
     "    --output_dir \"models/dart-math-llama3-70b-prop2diff\"\n",
     "```\n",
@@ -291,10 +286,7 @@
     "bash scripts/train-multi-node.sh \\\n",
     "    --data_path \"hkust-nlp/dart-math-uniform\" \\\n",
     "    --model_path \"meta-llama/Meta-Llama-3-70B\" \\\n",
-    "    --lr \"2e-5\" \\\n",
-    "    --bs 64 \\\n",
-    "    --n_grad_acc_steps 1 \\\n",
-    "    --n_epochs 1 \\\n",
+    "    --lr \"2e-5\" --bs 64 --n_grad_acc_steps 1 --n_epochs 1 \\\n",
     "    --n_nodes 4 \\\n",
     "    --output_dir \"models/dart-math-llama3-70b-prop2diff\"\n",
     "```\n",
@@ -321,17 +313,14 @@
     "For example, to reproduce one pass of greedy decoding with `DART-Math-Mistral-7B-Prop2Diff` on the 6 benchmarks in Table 2 on GPU 0, please run the following command:\n",
     "\n",
     "```shell\n",
-    "CUDA_VISIBLE_DEVICES=\"0\" RAY_DEDUP_LOGS=0 python pipeline/gen.py \\\n",
+    "CUDA_VISIBLE_DEVICES=\"0\" python pipeline/gen.py \\\n",
     "    --gen_save_path \"data/res/dart-math-mistral-7b-prop2diff.jsonl\" \\\n",
     "    --model_name_or_path \"hkust-nlp/dart-math-mistral-7b-prop2diff\" \\\n",
     "    --datasets \"math-test\" \"gsm8k-test\" \"mwpbench/college-math-test\" \"deepmind-mathematics\" \\\n",
     "        \"olympiadbench/OE_TO_maths_en_COMP\" \"theoremqa\" \\\n",
-    "    --max_new_tokens 2048 \\\n",
-    "    --temperature 0 \\\n",
-    "    --prompt_template \"auto\" \\\n",
-    "    --n_shots -1 \\\n",
+    "    --max_new_tokens 2048 --temperature 0 --top_p 0.95 \\\n",
+    "    --prompt_template \"cot\" --n_shots -1 \\\n",
     "    --inf_seed -1 \\\n",
-    "    --do_eval \\\n",
     "    --max_n_trials 1\n",
     "```\n",
     "\n",
@@ -358,19 +347,14 @@
     "For example, to reproduce synthesis of `DART-Math-Uniform`, run the following command with different GPUs, please run the following command:\n",
     "\n",
     "```shell\n",
-    "CUDA_VISIBLE_DEVICES=\"0\" RAY_DEDUP_LOGS=0 python pipeline/gen.py \\\n",
+    "CUDA_VISIBLE_DEVICES=\"0\" python pipeline/gen.py \\\n",
     "    --gen_save_path \"data/res/dart-math-uniform.jsonl\" \\\n",
     "    --model_name_or_path \"deepseek-ai/deepseek-math-7b-rl\" \\\n",
     "    --datasets \"math-train\" \"gsm8k-train\" \\\n",
-    "    --max_new_tokens 2048 \\\n",
-    "    --temperature 1.6 \\\n",
-    "    --top_p 0.95 \\\n",
-    "    --prompt_template \"auto\" \\\n",
-    "    --n_shots 0 \\\n",
+    "    --max_new_tokens 2048 --temperature 1.6 --top_p 0.95 \\\n",
+    "    --prompt_template \"cot\" --n_shots 0 \\\n",
     "    --inf_seed -1 \\\n",
-    "    --do_eval \\\n",
-    "    --min_n_corrects 40 \\\n",
-    "    --max_n_trials 0 # unlimited, should be killed manually\n",
+    "    --min_n_corrects 40 --max_n_trials 0 # unlimited, should be killed manually\n",
     "```\n"
    ]
   },
@@ -401,7 +385,7 @@
    "source": [
     "Thanks to:\n",
     "\n",
-    "- [`nbdev`](https://nbdev.fast.ai/) for genrating the [wonderful documentation website](https://hkust-nlp.github.io/dart-math),\n",
+    "- [`nbdev`](https://nbdev.fast.ai/) for generating the [wonderful documentation website](https://hkust-nlp.github.io/dart-math),\n",
     "- [`stanford_alpaca`](https://github.com/tatsu-lab/stanford_alpaca) for reference code about training,\n",
     "- [`functionary`](https://github.com/MeetKai/functionary/tree/main/functionary/train/packing) for reference code about [sequence packing](https://hkust-nlp.github.io/dart-math/train.html#sequence-packing).\n"
    ]
@@ -421,7 +405,7 @@
     "\n",
     "```latex\n",
     "@article{tong2024dartmath,\n",
-    "  author = {Yuxuan Tong, Xiwen Zhang, Rui Wang, Ruidong Wu, Junxian He },\n",
+    "  author = {Yuxuan Tong, Xiwen Zhang, Rui Wang, Ruidong Wu, Junxian He},\n",
     "  title = {DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving},\n",
     "  year = {2024},\n",
     "  publisher = {GitHub},\n",

Diff for: settings.ini

@@ -40,5 +40,5 @@ user = hkust-nlp
 ### Optional ###
 requirements = torch transformers datasets accelerate deepspeed triton vllm sympy==1.12 antlr4-python3-runtime==4.11.1 orjson tqdm
 # `flash-attn` should be built with `--no-build-isolation`
-dev_requirements = black flake8 flake8-bugbear flake8-nb flake8-mutable flake8-builtins flake8-no-implicit-concat flake8-comprehensions notebook ipywidgets nbdev nbconvert
+dev_requirements = black flake8 flake8-bugbear flake8-nb flake8-mutable flake8-builtins flake8-no-implicit-concat flake8-comprehensions pre-commit notebook ipywidgets nbdev nbconvert
 # console_scripts =