MaskGRPO
The official implementation of MaskGRPO: Consolidating Reinforcement Learning for Multimodal Discrete Diffusion Models. (ICLR 2026, arxiv:2510.02880)
File Explorer
- 1_gpu.yaml
- 1_node_8_gpus_deepspeed_zero2.yaml
- 1_node_8_gpus_deepspeed_zero3.yaml
- 8_node_8_gpus_deepspeed_zero2.yaml
- aesthetics.png
- banner.png
- portraits.png
- mmada_demo.yaml
- rl.yaml
- sft.yaml
- readme.txt
- dpg_bench.jsonl
- evaluation_metadata.jsonl
- evaluation_metadata_long.jsonl
- evaluation_metadata_merged.jsonl
- recaption.py
- rewrite.py
- test_metadata.jsonl
- train_metadata.jsonl
- train_metadata_filtered.jsonl
- HPSv2_eval.jsonl
- v3_eval_sample.jsonl
- benchmark-prompts.json
- mbpp.jsonl
- gsm8k.py
- math500.py
- parsers.py
- description.txt
- questions.jsonl
- ai2d.png
- clevr.jpg
- Decoupling Visual Encoding.png
- docvqa.png
- dog.png
- geo.jpg
- llava1.jpg
- llava2.jpg
- prompts.jsonl
- prompts_with_vqa.json
- sofa_under_water.jpg
- cot_questions.txt
- questions.txt
- countdown.py
- eval.py
- generate.py
- gsm8k.py
- math500.py
- parser_json.py
- parsers.py
- run_eval.sh
- sudoku.py
- vllm_eval.py
- mix-sft-old.txt
- mix-sft.txt
- quantative.txt
- test.txt
- text2image_prompts.txt
- get_math_accuracy.py
- parser_helper.py
- test_code_reward.py
- __init__.py
- code_providers.py
- e2b_router.py
- import_utils.py
- routed_morph.py
- routed_sandbox.py
- accelerate_a100.yaml
- countdown_base.sbatch
- countdown_sft.sbatch
- gsm_base.sbatch
- gsm_sft.sbatch
- math_base.sbatch
- math_sft.sbatch
- sudoku_base.sbatch
- sudoku_sft.sbatch
- train.yaml
- __init__.py
- accelerate.yaml
- data_utils.py
- grpo_config.py
- mask_grpo_train.py
- mask_grpo_trainer.py
- math500_utils.py
- reward_func.py
- __init__.py
- common_modules.py
- configuration_llada.py
- generate_llada.py
- logging.py
- lr_schedulers.py
- misc.py
- modeling_dimoo.py
- modeling_llada.py
- modeling_magvitv2.py
- modeling_mmada.py
- modeling_utils.py
- sampling.py
- training_utils.py
- __init__.py
- my_dataset.py
- DPG.sh
- GenEval.sh
- GSM8K.sh
- MATH500.sh
- MBPP.sh
- t2i_grpo_train.sh
- t2i_grpo_train_w_server.sh
- t2i_sft.sh
- t2t_grpo_train.sh
- run_e2b.sh
- __init__.py
- data.py
- optimizer.py
- prompting_utils.py
- train_mmada.py
- train_mmada_sft.py
- utils.py
- .gitignore
- distributed_inference_t2i.py
- distributed_inference_t2t.py
- generate.py
- inference_cot_local.py
- inference_mmu_local.py
- inference_t2i.py
- inference_t2i_local.py
- LICENSE
- MBPP_eval.py
- README.md
- requirements.txt
# Use via CDN
jsDelivrjsDelivr serves any public GitHub repository as a CDN with zero setup. Pick a version and a file to get a ready-to-paste link and snippet.
Command Glossary
Commands referenced in this DOCs, explained below.
conda activate
View Details ▼
conda activate
Activate a conda environment.
See also: `conda deactivate`.
conda activate myenv
Activate an existing environment named `myenv`:
conda activate {{path/to/myenv}}
Activate an existing environment located at custom path:
conda activate --stack myenv
Stack `myenv` environment on top of a previous environment making libraries/commands/variables from both accessible:
conda create
View Details ▼
conda create
Create new conda environments.
conda create {{[-y|--yes]}} {{[-n|--name]}} py39 python=3.9 "numpy>=1.11" scipy
Create a new environment named `py39`, install Python 3.9, NumPy v1.11 or above in it, and the latest stable version of SciPy. Say yes to all confirmations:
conda create {{[-n|--name]}} myenv --file {{file1.yml}} --file {{file2.yml}}
Create a new environment named `myenv` and install packages listed in files:
conda create {{[-p|--prefix]}} {{path/to/myenv}}
Create a new environment at a custom path (i.e. prefix):
pip install
View Details ▼
pip install
Install Python packages.
pip install {{package1 package2 ...}}
Install one or more packages:
pip install {{package1 package2 ...}} {{[-U|--upgrade]}}
Upgrade all specified packages to the latest version, installing any that are not already present:
pip install {{package}}=={{version}}
Install a specific version of a package:
