.dockerignore
.gitignore
.pre-commit-config.yaml
ATTRIBUTIONS.md
Dockerfile
LICENSE
README.md
pyproject.toml
ruff.toml
tools
.github/workflows/build-and-test.yaml
.github/workflows/doc_pages.yaml
.github/workflows/pre-commit.yaml
assets/nvidia-cosmos-header.png
assets/rl_infra.svg
configs/cosmos-policy/cosmos-policy-libero10-eval.toml
configs/cosmos-predict2-5/cosmos-predict2-5-14b-720-reason-embedding-ddrl.toml
configs/cosmos-predict2-5/cosmos-predict2-5-2b-480-grpo-mock-data.toml
configs/cosmos-predict2-5/cosmos-predict2-5-2b-480-grpo.toml
configs/cosmos-predict2-5/cosmos-predict2-5-2b-720-nft.toml
configs/cosmos-predict2-5/cosmos-predict2-5-2b-720-reason-embedding-ddrl.toml
configs/deepseek-v3/deepseek-v3-moe-670b-fsdp64-cp4-ep64-grpo.toml
configs/deepseek-v3/deepseek-v3-moe-670b-fsdp64-cp4-ep64-sft.toml
configs/deepseek-v3/deepseek-v3-moe-670b-fsdp64-pp4-ep64-sft.toml
configs/gemma/gemma3-12b-vlm-fsdp4-sft.toml
configs/gpt-oss/gpt-oss-20b-fsdp8-sft.toml
configs/gpt-oss/gpt-oss-20b-p-fsdp4-r-tp4-pp1-grpo.toml
configs/mistral/mistral-7b-fsdp4-sft.toml
configs/openvla-oft/openvla-oft-7b-fsdp2-8p8r-colocate-robotwin.toml
configs/openvla-oft/openvla-oft-7b-fsdp2-8p8r-colocate.toml
configs/openvla-oft/openvla-oft-7b-fsdp2-8p8r.toml
configs/openvla-oft/openvla-oft-libero10-eval.toml
configs/phi/phi4-14b-p-fsdp4-r-tp2-grpo.toml
configs/pi05/pi05-b1k-grpo-colocate.toml
configs/pi05/pi05-libero10-eval.toml
configs/pi05/pi05-libero10-grpo-colocate.toml
configs/pi05/pi05-libero10-grpo.toml
configs/pi05/pi05_behavior_pretrain.toml
configs/pi05/pi05_behavior_sft.toml
configs/qwen2-5/qwen2-5-1.5b-distill-countdown.toml
configs/qwen2-5/qwen2-5-32b-p-fsdp2-tp4-r-tp4-pp1-grpo.toml
configs/qwen2-5/qwen2-5-32b-tp2-fsdp4-sft.toml
configs/qwen2-5/qwen2-5-3b-p-fsdp1-tp1-r-tp1-pp1-grpo.toml
configs/qwen2-5/qwen2-5-3b-p-fsdp2-colocated-grpo.toml
configs/qwen2-5/qwen2-5-3b-p-fsdp2-colocated-separated-grpo.toml
configs/qwen2-5/qwen2-5-3b-tp1-sft.toml
configs/qwen2-5/qwen2-5-7b-p-fsdp2-cp2-r-tp2-pp1-grpo.toml
configs/qwen2-5/qwen2-5-7b-p-fsdp2-tp2-r-tp2-pp1-fp8-grpo.toml
configs/qwen2-5/qwen2-5-7b-p-fsdp4-r-tp2-grpo-trtllm.toml
configs/qwen2-5/qwen2-5-7b-p-fsdp4-r-tp2-pp1-grpo.toml
configs/qwen2-5/qwen2-5-7b-pp2-cp2-sft.toml
configs/qwen2-5/qwen2-5-multiturn-3b-p-fsdp1-tp1-r-tp1-pp1-grpo.toml
configs/qwen3/qwen3-32b-p-fsdp1-tp8-r-tp4-pp1-grpo.toml
configs/qwen3/qwen3-32b-tp8-sft.toml
configs/qwen3/qwen3-8b-distill-deepmath.toml
configs/qwen3/qwen3-8b-distill.toml
configs/qwen3/qwen3-8b-p-fsdp1-tp4-fp8-r-tp4-pp1-fp8-grpo.toml
configs/qwen3/qwen3-8b-p-fsdp1-tp4-r-tp4-pp1-grpo-fp8.toml
configs/qwen3/qwen3-8b-p-tp4-r-tp2-pp1-grpo.toml
configs/qwen3-moe/qwen3-moe-30b-fsdp2-tp4-sft.toml
configs/qwen3-moe/qwen3-moe-30b-p-tp4-r-tp2-grpo-trtllm.toml
configs/qwen3-moe/qwen3-moe-30b-tp4-sft.toml
configs/sana/sana-image-nft.toml
configs/sana/sana-image-sft-lora.toml
configs/sana/sana-image-sft.toml
configs/sana/sana-video-nft.toml
configs/sana/sana-video-sft-lora.toml
configs/sana/sana-video-sft.toml
configs/stable-diffusion-3-5/stable-diffusion-3-5-image-sft-lora.toml
configs/stable-diffusion-3-5/stable-diffusion-3-5-image-sft.toml
configs/stable-diffusion-3-5/stable-diffusion-3-5-medium-nft.toml
cosmos_rl/__init__.py
cosmos_rl/_version.py
cosmos_rl.egg-info/PKG-INFO
cosmos_rl.egg-info/SOURCES.txt
cosmos_rl.egg-info/dependency_links.txt
cosmos_rl.egg-info/entry_points.txt
cosmos_rl.egg-info/requires.txt
cosmos_rl.egg-info/top_level.txt
cosmos_rl/cli/__init__.py
cosmos_rl/cli/algo.py
cosmos_rl/cli/cli.py
cosmos_rl/cli/custom_group.py
cosmos_rl/cli/nccl.py
cosmos_rl/cli/profiler.py
cosmos_rl/cli/replica.py
cosmos_rl/cli/utils.py
cosmos_rl/collective/__init__.py
cosmos_rl/collective/collective.py
cosmos_rl/colocated/__init__.py
cosmos_rl/colocated/api_client.py
cosmos_rl/colocated/controller.py
cosmos_rl/colocated/rl_worker.py
cosmos_rl/colocated/utils.py
cosmos_rl/comm/__init__.py
cosmos_rl/comm/base.py
cosmos_rl/dispatcher/__init__.py
cosmos_rl/dispatcher/command.py
cosmos_rl/dispatcher/controller.py
cosmos_rl/dispatcher/protocol.py
cosmos_rl/dispatcher/replica.py
cosmos_rl/dispatcher/run_web_panel.py
cosmos_rl/dispatcher/status.py
cosmos_rl/dispatcher/algo/__init__.py
cosmos_rl/dispatcher/algo/base.py
cosmos_rl/dispatcher/algo/grpo.py
cosmos_rl/dispatcher/algo/reward.py
cosmos_rl/dispatcher/api/client.py
cosmos_rl/dispatcher/config/__init__.py
cosmos_rl/dispatcher/config/frontend/__init__.py
cosmos_rl/dispatcher/config/frontend/configure_template.html
cosmos_rl/dispatcher/config/frontend/dispatcher_status.html
cosmos_rl/dispatcher/data/__init__.py
cosmos_rl/dispatcher/data/data_fetcher.py
cosmos_rl/dispatcher/data/load_balanced_dataset.py
cosmos_rl/dispatcher/data/schema.py
cosmos_rl/dispatcher/data/packer/__init__.py
cosmos_rl/dispatcher/data/packer/base.py
cosmos_rl/dispatcher/data/packer/decoder_only_llm_data_packer.py
cosmos_rl/dispatcher/data/packer/deepseek_data_packer.py
cosmos_rl/dispatcher/data/packer/hf_vlm_data_packer.py
cosmos_rl/dispatcher/data/packer/multi_turn.py
cosmos_rl/dispatcher/data/packer/pi05_data_packer.py
cosmos_rl/dispatcher/data/packer/qwen2_5_vlm_data_packer.py
cosmos_rl/dispatcher/data/packer/qwen3_vl_data_packer.py
cosmos_rl/dispatcher/data/packer/vla_data_packer.py
cosmos_rl/launcher/__init__.py
cosmos_rl/launcher/launch_all.py
cosmos_rl/launcher/launch_controller.sh
cosmos_rl/launcher/launch_replica.sh
cosmos_rl/launcher/launch_vision.py
cosmos_rl/launcher/utility.py
cosmos_rl/launcher/worker_entry.py
cosmos_rl/patch/__init__.py
cosmos_rl/patch/vllm_patch.py
cosmos_rl/policy/__init__.py
cosmos_rl/policy/policy_entry.py
cosmos_rl/policy/train.py
cosmos_rl/policy/config/__init__.py
cosmos_rl/policy/config/wfm/__init__.py
cosmos_rl/policy/config/wfm/qwen_config.py
cosmos_rl/policy/config/wfm/reason1_model_config.py
cosmos_rl/policy/kernel/__init__.py
cosmos_rl/policy/kernel/fused.py
cosmos_rl/policy/kernel/group_gemms.py
cosmos_rl/policy/kernel/loss.py
cosmos_rl/policy/kernel/modeling_utils.py
cosmos_rl/policy/kernel/norm.py
cosmos_rl/policy/kernel/rope.py
cosmos_rl/policy/kernel/megatron_moe/README.md
cosmos_rl/policy/kernel/megatron_moe/__init__.py
cosmos_rl/policy/kernel/megatron_moe/fused_a2a.py
cosmos_rl/policy/kernel/megatron_moe/fused_indices_converter.py
cosmos_rl/policy/kernel/megatron_moe/moe_utils.py
cosmos_rl/policy/kernel/megatron_moe/permutation.py
cosmos_rl/policy/kernel/megatron_moe/te_permutation.py
cosmos_rl/policy/kernel/megatron_moe/token_dispatcher.py
cosmos_rl/policy/kernel/moe/__init__.py
cosmos_rl/policy/kernel/moe/grouped_gemm.py
cosmos_rl/policy/kernel/moe/indices.py
cosmos_rl/policy/kernel/moe/moe.py
cosmos_rl/policy/kernel/symm_mem_recipes/__init__.py
cosmos_rl/policy/kernel/symm_mem_recipes/triton_barrier.py
cosmos_rl/policy/kernel/symm_mem_recipes/triton_on_device_all_to_all_v.py
cosmos_rl/policy/kernel/symm_mem_recipes/triton_utils.py
cosmos_rl/policy/lora/__init__.py
cosmos_rl/policy/lora/plugin.py
cosmos_rl/policy/model/__init__.py
cosmos_rl/policy/model/base.py
cosmos_rl/policy/model/deepseek_v3/__init__.py
cosmos_rl/policy/model/deepseek_v3/deepseekv3_mapped.py
cosmos_rl/policy/model/deepseek_v3/grpo_launcher.py
cosmos_rl/policy/model/deepseek_v3/parallelize.py
cosmos_rl/policy/model/deepseek_v3/weight_mapper.py
cosmos_rl/policy/model/deepseek_v3/pipeline_parallelism/__init__.py
cosmos_rl/policy/model/deepseek_v3/pipeline_parallelism/pipeline_model.py
cosmos_rl/policy/model/diffusers/__init__.py
cosmos_rl/policy/model/diffusers/cosmos_predict2_5_model.py
cosmos_rl/policy/model/diffusers/parallelize.py
cosmos_rl/policy/model/diffusers/sana_model.py
cosmos_rl/policy/model/diffusers/sd3_model.py
cosmos_rl/policy/model/diffusers/weight_mapper.py
cosmos_rl/policy/model/gpt/__init__.py
cosmos_rl/policy/model/gpt/parallelize.py
cosmos_rl/policy/model/gpt/weight_converter.py
cosmos_rl/policy/model/gpt/weight_mapper.py
cosmos_rl/policy/model/hf_models/__init__.py
cosmos_rl/policy/model/hf_models/parallelize.py
cosmos_rl/policy/model/hf_models/patch.py
cosmos_rl/policy/model/hf_models/tp_plans.py
cosmos_rl/policy/model/hf_models/weight_converter.py
cosmos_rl/policy/model/hf_models/weight_mapper.py
cosmos_rl/policy/model/internvl/__init__.py
cosmos_rl/policy/model/internvl/parallelize.py
cosmos_rl/policy/model/internvl/weight_converter.py
cosmos_rl/policy/model/internvl/weight_mapper.py
cosmos_rl/policy/model/pi05/__init__.py
cosmos_rl/policy/model/pi05/explore_noise_net.py
cosmos_rl/policy/model/pi05/model_utils.py
cosmos_rl/policy/model/pi05/parallelize.py
cosmos_rl/policy/model/pi05/weight_mapper.py
cosmos_rl/policy/model/pi05/transformers_replace/__init__.py
cosmos_rl/policy/model/pi05/transformers_replace/configuration_gemma.py
cosmos_rl/policy/model/pi05/transformers_replace/configuration_paligemma.py
cosmos_rl/policy/model/pi05/transformers_replace/configuration_siglip.py
cosmos_rl/policy/model/pi05/transformers_replace/modeling_gemma.py
cosmos_rl/policy/model/pi05/transformers_replace/modeling_paligamma.py
cosmos_rl/policy/model/pi05/transformers_replace/modeling_siglip.py
cosmos_rl/policy/model/qwen2_5_vl/__init__.py
cosmos_rl/policy/model/qwen2_5_vl/parallelize.py
cosmos_rl/policy/model/qwen2_5_vl/weight_converter.py
cosmos_rl/policy/model/qwen2_5_vl/weight_mapper.py
cosmos_rl/policy/model/qwen3_moe/__init__.py
cosmos_rl/policy/model/qwen3_moe/parallelize.py
cosmos_rl/policy/model/qwen3_moe/weight_converter.py
cosmos_rl/policy/model/qwen3_moe/weight_mapper.py
cosmos_rl/policy/model/qwen3_vl_moe/__init__.py
cosmos_rl/policy/model/qwen3_vl_moe/parallelize.py
cosmos_rl/policy/model/qwen3_vl_moe/weight_converter.py
cosmos_rl/policy/model/qwen3_vl_moe/weight_mapper.py
cosmos_rl/policy/model/vision_encoder/internvl.py
cosmos_rl/policy/model/vision_encoder/qwen2_5_vl.py
cosmos_rl/policy/model/vision_encoder/qwen3_vl_moe.py
cosmos_rl/policy/model/vla/__init__.py
cosmos_rl/policy/model/vla/parallelize.py
cosmos_rl/policy/model/vla/processing_utils.py
cosmos_rl/policy/model/vla/weight_converter.py
cosmos_rl/policy/model/vla/weight_mapper.py
cosmos_rl/policy/model/vla/openvla/__init__.py
cosmos_rl/policy/model/vla/openvla/configuration_prismatic.py
cosmos_rl/policy/model/vla/openvla/modeling_prismatic.py
cosmos_rl/policy/model/vla/openvla/processing_prismatic.py
cosmos_rl/policy/model/vla/openvla_oft/__init__.py
cosmos_rl/policy/model/vla/openvla_oft/configuration_prismatic.py
cosmos_rl/policy/model/vla/openvla_oft/constants.py
cosmos_rl/policy/model/vla/openvla_oft/modeling_prismatic.py
cosmos_rl/policy/model/vla/openvla_oft/processing_prismatic.py
cosmos_rl/policy/model/vla/openvla_oft/train_utils.py
cosmos_rl/policy/model/wfm/__init__.py
cosmos_rl/policy/model/wfm/conditioner/__init__.py
cosmos_rl/policy/model/wfm/conditioner/condition.py
cosmos_rl/policy/model/wfm/conditioner/emb_models.py
cosmos_rl/policy/model/wfm/conditioner/video2world.py
cosmos_rl/policy/model/wfm/models/__init__.py
cosmos_rl/policy/model/wfm/models/cosmos_policy.py
cosmos_rl/policy/model/wfm/models/t2v_model.py
cosmos_rl/policy/model/wfm/models/v2v_model.py
cosmos_rl/policy/model/wfm/modules/__init__.py
cosmos_rl/policy/model/wfm/modules/attention.py
cosmos_rl/policy/model/wfm/modules/neighborhood_attn.py
cosmos_rl/policy/model/wfm/modules/rope.py
cosmos_rl/policy/model/wfm/networks/__init__.py
cosmos_rl/policy/model/wfm/networks/a2a_cp.py
cosmos_rl/policy/model/wfm/networks/minimal_v1_lvg_dit.py
cosmos_rl/policy/model/wfm/networks/minimal_v4_dit.py
cosmos_rl/policy/model/wfm/networks/model_weights_stats.py
cosmos_rl/policy/model/wfm/networks/vlm_base.py
cosmos_rl/policy/model/wfm/networks/vlm_qwen/__init__.py
cosmos_rl/policy/model/wfm/networks/vlm_qwen/parallelize.py
cosmos_rl/policy/model/wfm/networks/vlm_qwen/processor.py
cosmos_rl/policy/model/wfm/networks/vlm_qwen/qwen2_5_vl.py
cosmos_rl/policy/model/wfm/networks/vlm_qwen/qwen2_vl.py
cosmos_rl/policy/model/wfm/networks/vlm_qwen/qwen_omni.py
cosmos_rl/policy/model/wfm/rewards/__init__.py
cosmos_rl/policy/model/wfm/rewards/reward_models.py
cosmos_rl/policy/model/wfm/sampler/__init__.py
cosmos_rl/policy/model/wfm/sampler/cosmos_policy_sampler.py
cosmos_rl/policy/model/wfm/tokenizer/__init__.py
cosmos_rl/policy/model/wfm/tokenizer/base.py
cosmos_rl/policy/model/wfm/tokenizer/wan2pt1.py
cosmos_rl/policy/model/wfm/tokenizer/wan2pt2.py
cosmos_rl/policy/trainer/__init__.py
cosmos_rl/policy/trainer/base.py
cosmos_rl/policy/trainer/cosmos_policy_trainer.py
cosmos_rl/policy/trainer/sampler.py
cosmos_rl/policy/trainer/wfm_trainer.py
cosmos_rl/policy/trainer/diffusers_trainer/__init__.py
cosmos_rl/policy/trainer/diffusers_trainer/diffusers_trainer.py
cosmos_rl/policy/trainer/diffusers_trainer/nft_trainer.py
cosmos_rl/policy/trainer/diffusers_trainer/sft_trainer.py
cosmos_rl/policy/trainer/llm_trainer/__init__.py
cosmos_rl/policy/trainer/llm_trainer/dpo_trainer.py
cosmos_rl/policy/trainer/llm_trainer/grpo_trainer.py
cosmos_rl/policy/trainer/llm_trainer/llm_trainer.py
cosmos_rl/policy/trainer/llm_trainer/sft_trainer.py
cosmos_rl/policy/trainer/optm/__init__.py
cosmos_rl/policy/trainer/vla_trainer/__init__.py
cosmos_rl/policy/trainer/vla_trainer/pi_grpo_trainer.py
cosmos_rl/policy/trainer/vla_trainer/pi_sft_trainer.py
cosmos_rl/policy/trainer/vla_trainer/vla_trainer.py
cosmos_rl/policy/worker/__init__.py
cosmos_rl/policy/worker/base.py
cosmos_rl/policy/worker/multi_replica_sft_worker.py
cosmos_rl/policy/worker/rl_worker.py
cosmos_rl/policy/worker/sft_worker.py
cosmos_rl/policy/worker/wfm_worker.py
cosmos_rl/policy/worker/colocated/__init__.py
cosmos_rl/policy/worker/colocated/policy_control.py
cosmos_rl/reference/__init__.py
cosmos_rl/reference/reference_entry.py
cosmos_rl/reference/engine/__init__.py
cosmos_rl/reference/engine/torch_engine.py
cosmos_rl/reference/worker/__init__.py
cosmos_rl/reference/worker/teacher_worker.py
cosmos_rl/reward/__init__.py
cosmos_rl/reward/base.py
cosmos_rl/reward/dispatcher.py
cosmos_rl/reward/local_calculator.py
cosmos_rl/reward/remote_calculator.py
cosmos_rl/rollout/__init__.py
cosmos_rl/rollout/rollout_base.py
cosmos_rl/rollout/rollout_entry.py
cosmos_rl/rollout/schema.py
cosmos_rl/rollout/diffuers_rollout/__init__.py
cosmos_rl/rollout/diffuers_rollout/nft_rollout.py
cosmos_rl/rollout/example_custom_rollout/__init__.py
cosmos_rl/rollout/example_custom_rollout/example_custom_rollout.py
cosmos_rl/rollout/trtllm_rollout/__init__.py
cosmos_rl/rollout/trtllm_rollout/patch_trtllm.py
cosmos_rl/rollout/trtllm_rollout/trtllm_common.py
cosmos_rl/rollout/trtllm_rollout/trtllm_rollout.py
cosmos_rl/rollout/trtllm_rollout/trtllm_rollout_wrapper.py
cosmos_rl/rollout/trtllm_rollout/trtllm_worker.py
cosmos_rl/rollout/trtllm_rollout/exec_loops/__init__.py
cosmos_rl/rollout/trtllm_rollout/exec_loops/trtllm_1_0_0_rc6.py
cosmos_rl/rollout/trtllm_rollout/patches_for_trtllm/__init__.py
cosmos_rl/rollout/trtllm_rollout/patches_for_trtllm/trtllm_1_0_0_rc6.py
cosmos_rl/rollout/vla_rollout/__init__.py
cosmos_rl/rollout/vla_rollout/trace_utils.py
cosmos_rl/rollout/vla_rollout/vla_rollout.py
cosmos_rl/rollout/vllm_rollout/__init__.py
cosmos_rl/rollout/vllm_rollout/monkey_patch_for_fp8.py
cosmos_rl/rollout/vllm_rollout/monkey_patch_for_mxfp4.py
cosmos_rl/rollout/vllm_rollout/vllm_rollout.py
cosmos_rl/rollout/vllm_rollout/vllm_rollout_async.py
cosmos_rl/rollout/wfm_rollout/__init__.py
cosmos_rl/rollout/wfm_rollout/wfm_rollout.py
cosmos_rl/rollout/worker/__init__.py
cosmos_rl/rollout/worker/llm_worker.py
cosmos_rl/rollout/worker/rollout_control.py
cosmos_rl/rollout/worker/weight_sync.py
cosmos_rl/rollout/worker/asynchronous/__init__.py
cosmos_rl/rollout/worker/asynchronous/rollout_task_scheduler.py
cosmos_rl/rollout/worker/colocated/__init__.py
cosmos_rl/rollout/worker/colocated/rollout_control.py
cosmos_rl/simulators/__init__.py
cosmos_rl/simulators/env_manager.py
cosmos_rl/simulators/utils.py
cosmos_rl/simulators/b1k/__init__.py
cosmos_rl/simulators/b1k/behavior_task.jsonl
cosmos_rl/simulators/b1k/env_wrapper.py
cosmos_rl/simulators/b1k/utils.py
cosmos_rl/simulators/b1k/venv.py
cosmos_rl/simulators/libero/__init__.py
cosmos_rl/simulators/libero/env_wrapper.py
cosmos_rl/simulators/libero/utils.py
cosmos_rl/simulators/libero/venv.py
cosmos_rl/simulators/maniskill/__init__.py
cosmos_rl/simulators/maniskill/env_wrapper.py
cosmos_rl/simulators/robotwin/__init__.py
cosmos_rl/simulators/robotwin/env_wrapper.py
cosmos_rl/simulators/robotwin/venv.py
cosmos_rl/tools/__init__.py
cosmos_rl/tools/chat_templates/gsm8k_tool_chat_template.jinja2
cosmos_rl/tools/custom_example/README.md
cosmos_rl/tools/custom_example/__init__.py
cosmos_rl/tools/custom_example/custom_grpo_trainer.py
cosmos_rl/tools/dataset/__init__.py
cosmos_rl/tools/dataset/b1k_grpo.py
cosmos_rl/tools/dataset/b1k_sft.py
cosmos_rl/tools/dataset/countdown_distill.py
cosmos_rl/tools/dataset/deepmath_distill.py
cosmos_rl/tools/dataset/diffusers_dataset.py
cosmos_rl/tools/dataset/diffusion_nft.py
cosmos_rl/tools/dataset/dummy_diffusers_sft.py
cosmos_rl/tools/dataset/dummy_sft.py
cosmos_rl/tools/dataset/gsm8k_grpo.py
cosmos_rl/tools/dataset/hf_vlm_grpo.py
cosmos_rl/tools/dataset/hf_vlm_sft.py
cosmos_rl/tools/dataset/internvl_sft.py
cosmos_rl/tools/dataset/libero_grpo.py
cosmos_rl/tools/dataset/math_dapo.py
cosmos_rl/tools/dataset/math_grpo.py
cosmos_rl/tools/dataset/post_completion.py
cosmos_rl/tools/dataset/robotwin2_grpo.py
cosmos_rl/tools/dataset/wfm_rl.py
cosmos_rl/tools/dataset/wfm/__init__.py
cosmos_rl/tools/dataset/wfm/augmentor_provider.py
cosmos_rl/tools/dataset/wfm/cached_replay_dataloader.py
cosmos_rl/tools/dataset/wfm/dataset_provider.py
cosmos_rl/tools/dataset/wfm/joint_dataloader.py
cosmos_rl/tools/dataset/wfm/local_dataloader.py
cosmos_rl/tools/dataset/wfm/utils.py
cosmos_rl/tools/dataset/wfm/video_decoder.py
cosmos_rl/tools/dataset/wfm/data_sources/__init__.py
cosmos_rl/tools/dataset/wfm/data_sources/data_registration.py
cosmos_rl/tools/dataset/wfm/data_sources/item_datasets_for_validation.py
cosmos_rl/tools/dataset/wfm/data_sources/mock_data.py
cosmos_rl/tools/dataset/wfm/local_datasets/__init__.py
cosmos_rl/tools/dataset/wfm/local_datasets/dataset_image.py
cosmos_rl/tools/dataset/wfm/local_datasets/dataset_utils.py
cosmos_rl/tools/dataset/wfm/local_datasets/dataset_video.py
cosmos_rl/tools/dataset/wfm/webdataset/__init__.py
cosmos_rl/tools/dataset/wfm/webdataset/dataloader.py
cosmos_rl/tools/dataset/wfm/webdataset/webdataset.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/__init__.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/augmentor.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/merge_datadict.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/image/__init__.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/image/append_fps_frames_for_image.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/image/cropping.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/image/flip.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/image/misc.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/image/normalize.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/image/padding.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/image/resize.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/image/text_transform_for_image.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/video/__init__.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/video/text_transforms_for_video.py
cosmos_rl/tools/dataset/wfm/webdataset/augmentors/video/video_parsing.py
cosmos_rl/tools/dataset/wfm/webdataset/config/schema.py
cosmos_rl/tools/dataset/wfm/webdataset/decoders/__init__.py
cosmos_rl/tools/dataset/wfm/webdataset/decoders/image.py
cosmos_rl/tools/dataset/wfm/webdataset/decoders/pickle.py
cosmos_rl/tools/dataset/wfm/webdataset/distributors/__init__.py
cosmos_rl/tools/dataset/wfm/webdataset/distributors/basic.py
cosmos_rl/tools/dataset/wfm/webdataset/distributors/multi_aspect_ratio.py
cosmos_rl/tools/dataset/wfm/webdataset/distributors/multi_aspect_ratio_v2.py
cosmos_rl/tools/dataset/wfm/webdataset/distributors/parallel_sync_multi_aspect_ratio.py
cosmos_rl/tools/dataset/wfm/webdataset/utils/iterators.py
cosmos_rl/tools/dataset/wfm/webdataset/utils/misc.py
cosmos_rl/tools/dataset/wfm/webdataset/utils/stream.py
cosmos_rl/tools/eval_utils/code_verifier_utils.py
cosmos_rl/tools/eval_utils/grader.py
cosmos_rl/tools/scripts/gb200_setup_torch2.10.sh
cosmos_rl/tools/scripts/gb200_setup_torch2.9.sh
cosmos_rl/tools/scripts/install_flash_attn.sh
cosmos_rl/tools/scripts/setup_vla.sh
cosmos_rl/tools/scripts/sync_repo_filtered.sh
cosmos_rl/tools/slurm/README.md
cosmos_rl/tools/slurm/__init__.py
cosmos_rl/tools/slurm/cosmos_rl_job_multi_node.sh
cosmos_rl/tools/slurm/cosmos_rl_slurm_launch.py
cosmos_rl/tools/slurm/dispatch_job.py
cosmos_rl/tools/slurm/util.py
cosmos_rl/utils/__init__.py
cosmos_rl/utils/activation_offloading.py
cosmos_rl/utils/api_suffix.py
cosmos_rl/utils/arithmetic.py
cosmos_rl/utils/async_utils.py
cosmos_rl/utils/attn_util.py
cosmos_rl/utils/balance_seqlen.py
cosmos_rl/utils/cache.py
cosmos_rl/utils/checkpoint.py
cosmos_rl/utils/constant.py
cosmos_rl/utils/diffusers_utils.py
cosmos_rl/utils/dim_slice_info.py
cosmos_rl/utils/dist_signal_handler.py
cosmos_rl/utils/distributed.py
cosmos_rl/utils/ema.py
cosmos_rl/utils/logging.py
cosmos_rl/utils/model_converter.py
cosmos_rl/utils/modelscope.py
cosmos_rl/utils/mpi_distributed.py
cosmos_rl/utils/multi_rank_weight_loader.py
cosmos_rl/utils/nccl_transfer_protocol.py
cosmos_rl/utils/network_util.py
cosmos_rl/utils/parallelism.py
cosmos_rl/utils/parallelism_map.py
cosmos_rl/utils/parallelism_registry.py
cosmos_rl/utils/payload.py
cosmos_rl/utils/profiler.py
cosmos_rl/utils/pynccl.py
cosmos_rl/utils/pynccl_wrapper.py
cosmos_rl/utils/redis_stream.py
cosmos_rl/utils/replay_buffer.py
cosmos_rl/utils/s3_utils.py
cosmos_rl/utils/sequence_packing.py
cosmos_rl/utils/ulysses.py
cosmos_rl/utils/util.py
cosmos_rl/utils/diffusers/__init__.py
cosmos_rl/utils/diffusers/solver.py
cosmos_rl/utils/fp4/__init__.py
cosmos_rl/utils/fp4/config.py
cosmos_rl/utils/fp4/float4_linear.py
cosmos_rl/utils/fp4/float4_linear_utils.py
cosmos_rl/utils/fp4/fp4_util.py
cosmos_rl/utils/fp8/__init__.py
cosmos_rl/utils/fp8/fp8_util.py
cosmos_rl/utils/ipc/__init__.py
cosmos_rl/utils/ipc/adapters.py
cosmos_rl/utils/ipc/tensor_util.py
cosmos_rl/utils/mxfp4/__init__.py
cosmos_rl/utils/mxfp4/quantizer.py
cosmos_rl/utils/pipelining/__init__.py
cosmos_rl/utils/pipelining/pipelining_utils.py
cosmos_rl/utils/report/metrics_collection.py
cosmos_rl/utils/report/wandb_logger.py
cosmos_rl/utils/tools_use/__init__.py
cosmos_rl/utils/tools_use/base_tool.py
cosmos_rl/utils/tools_use/base_tool_parser.py
cosmos_rl/utils/tools_use/hermes_tool_parser.py
cosmos_rl/utils/tools_use/schema.py
cosmos_rl/utils/tools_use/tool_agent.py
cosmos_rl/utils/transformers_utils/__init__.py
cosmos_rl/utils/transformers_utils/modeling_rope_utils.py
cosmos_rl/utils/wfm/__init__.py
cosmos_rl/utils/wfm/callback.py
cosmos_rl/utils/wfm/checkpointer.py
cosmos_rl/utils/wfm/context_parallel.py
cosmos_rl/utils/wfm/distributed.py
cosmos_rl/utils/wfm/ema.py
cosmos_rl/utils/wfm/env_parser.py
cosmos_rl/utils/wfm/lr_scheduler.py
cosmos_rl/utils/wfm/model_loader.py
cosmos_rl/utils/wfm/optimizer.py
cosmos_rl/utils/wfm/ref.py
cosmos_rl/utils/wfm/selective_activation_checkpoint.py
cosmos_rl/utils/wfm/torch_future.py
cosmos_rl/utils/wfm/torchtitan_utilts.py
cosmos_rl/utils/wfm/utils.py
cosmos_rl/utils/wfm/validator.py
cosmos_rl/utils/wfm/io/__init__.py
cosmos_rl/utils/wfm/io/cred_env_parser.py
cosmos_rl/utils/wfm/io/object_store.py
cosmos_rl/utils/wfm/io/s3_filesystem.py
cosmos_rl/utils/wfm/io/easy_io/__init__.py
cosmos_rl/utils/wfm/io/easy_io/easy_io.py
cosmos_rl/utils/wfm/io/easy_io/file_client.py
cosmos_rl/utils/wfm/io/easy_io/backends/__init__.py
cosmos_rl/utils/wfm/io/easy_io/backends/auto_auth.py
cosmos_rl/utils/wfm/io/easy_io/backends/base_backend.py
cosmos_rl/utils/wfm/io/easy_io/backends/boto3_backend.py
cosmos_rl/utils/wfm/io/easy_io/backends/boto3_client.py
cosmos_rl/utils/wfm/io/easy_io/backends/local_backend.py
cosmos_rl/utils/wfm/io/easy_io/backends/registry_utils.py
cosmos_rl/utils/wfm/io/easy_io/handlers/__init__.py
cosmos_rl/utils/wfm/io/easy_io/handlers/base.py
cosmos_rl/utils/wfm/io/easy_io/handlers/handlers.py
cosmos_rl/utils/wfm/io/easy_io/handlers/registry_utils.py
cosmos_rl/utils/wfm/visualize/__init__.py
cosmos_rl/utils/wfm/visualize/video.py
docker/Dockerfile.arm64
docker/Dockerfile.trtllm
docs/Makefile
docs/README.md
docs/conf.py
docs/index.rst
docs/make.bat
docs/assets/data_flow.png
docs/assets/old_rollout.png
docs/assets/policy.png
docs/assets/rollout.png
docs/assets/weight_p2p.png
docs/async/overview.rst
docs/distillation/on_policy_distillation.rst
docs/elastic/overview.rst
docs/multi_turn/multi-turn rollout example.rst
docs/multinodes/dgxc_lepton.rst
docs/multinodes/overview.rst
docs/multinodes/slurm.rst
docs/parallelism/overview.rst
docs/profiler/overview.rst
docs/quantization/fp8.rst
docs/quickstart/configuration.rst
docs/quickstart/customization.rst
docs/quickstart/dataflow.rst
docs/quickstart/hf_models_support.rst
docs/quickstart/installation.rst
docs/quickstart/load_balanced_batching.rst
docs/quickstart/single_node_example.rst
docs/rollout/async_rollout.rst
docs/rollout/overview.rst
docs/rollout/trtllm.rst
docs/rollout/vllm.rst
docs/vla/overview.rst
docs/wfm/overview.rst
examples/ddrl.md
examples/on_policy_distillation.md
examples/profiler.md
examples/sana.md
reward_service/README.md
reward_service/pyproject.toml
reward_service/cosmos_rl_reward/__init__.py
reward_service/cosmos_rl_reward/configs/rewards.toml
reward_service/cosmos_rl_reward/example/__init__.py
reward_service/cosmos_rl_reward/example/image_client.py
reward_service/cosmos_rl_reward/example/latency_benchmark.py
reward_service/cosmos_rl_reward/example/video_client.py
reward_service/cosmos_rl_reward/handler/__init__.py
reward_service/cosmos_rl_reward/handler/decode.py
reward_service/cosmos_rl_reward/handler/process.py
reward_service/cosmos_rl_reward/handler/registry.py
reward_service/cosmos_rl_reward/handler/reward_base.py
reward_service/cosmos_rl_reward/handler/score_kv.py
reward_service/cosmos_rl_reward/launcher/__init__.py
reward_service/cosmos_rl_reward/launcher/config.py
reward_service/cosmos_rl_reward/launcher/redis_serve.py
reward_service/cosmos_rl_reward/launcher/reward_compute.py
reward_service/cosmos_rl_reward/launcher/reward_serve.py
reward_service/cosmos_rl_reward/model/__init__.py
reward_service/cosmos_rl_reward/model/cosmos_reason1.py
reward_service/cosmos_rl_reward/model/dance_grpo.py
reward_service/cosmos_rl_reward/model/gen_eval.py
reward_service/cosmos_rl_reward/model/hpsv2.py
reward_service/cosmos_rl_reward/model/hpsv3.py
reward_service/cosmos_rl_reward/model/image_reward.py
reward_service/cosmos_rl_reward/model/ocr.py
reward_service/cosmos_rl_reward/model/pickscore.py
reward_service/cosmos_rl_reward/model/unified_reward.py
reward_service/cosmos_rl_reward/setup/__init__.py
reward_service/cosmos_rl_reward/setup/cosmos_reason1.sh
reward_service/cosmos_rl_reward/setup/dance_grpo.sh
reward_service/cosmos_rl_reward/setup/env_prepare.py
reward_service/cosmos_rl_reward/setup/gen_eval.sh
reward_service/cosmos_rl_reward/setup/hpsv2.sh
reward_service/cosmos_rl_reward/setup/hpsv3.sh
reward_service/cosmos_rl_reward/setup/image_reward.sh
reward_service/cosmos_rl_reward/setup/ocr.sh
reward_service/cosmos_rl_reward/setup/pickscore.sh
reward_service/cosmos_rl_reward/setup/unified_reward.sh
reward_service/cosmos_rl_reward/utils/__init__.py
reward_service/cosmos_rl_reward/utils/client.py
reward_service/cosmos_rl_reward/utils/kv_store.py
reward_service/cosmos_rl_reward/utils/logging.py
reward_service/cosmos_rl_reward/utils/protocol.py
reward_service/cosmos_rl_reward/utils/redis.py
reward_service/cosmos_rl_reward/utils/shmem.py
reward_service/cosmos_rl_reward/utils/util.py
tests/launch_test_worker.py
tests/run_test.sh
tests/test_activation_offload.py
tests/test_apex.py
tests/test_cache.py
tests/test_checkpoint.py
tests/test_colocated.py
tests/test_colocated_separated.py
tests/test_comm.py
tests/test_context_parallel.py
tests/test_cosmos_hf_precision.py
tests/test_custom_args.py
tests/test_custom_class.py
tests/test_data_loader.py
tests/test_data_packer.py
tests/test_dataset_signature.py
tests/test_deepep.py
tests/test_diffusion_rl_e2e.py
tests/test_dpo_direct.py
tests/test_env_manager_b1k.py
tests/test_env_manager_libero.py
tests/test_env_manager_maniskill.py
tests/test_env_manager_robotwin.py
tests/test_fa_switch.py
tests/test_fp8.py
tests/test_freeze_pattern.py
tests/test_grad_allreduce.py
tests/test_hf_models.py
tests/test_hf_models_tp.py
tests/test_high_availability_nccl.py
tests/test_integration.py
tests/test_load_balanced_dataset.py
tests/test_lora.py
tests/test_math_verify.py
tests/test_mmpr_dpo_dataset.py
tests/test_nccl_collectives.py
tests/test_nccl_timeout.py
tests/test_nemotron_vl.py
tests/test_parallel_map.py
tests/test_policy_overfit.py
tests/test_policy_to_policy.py
tests/test_policy_to_rollout.py
tests/test_policy_variant.py
tests/test_process_flow.py
tests/test_qwen3_vl_moe.py
tests/test_resume_data_index.py
tests/test_sequence_packing.py
tests/test_sft_direct.py
tests/test_teacher_model.py
tests/test_vla_libero.py
tests/test_vllm_rollout_async.py
tests/test_wan2pt2_reward_service.py
tests/test_weight_sync.py
tests/benchmark/launch_rollout_bench.sh
tests/benchmark/rollout_benchmark.py
tests/configs/diffusion_rl_e2e_test.toml
tests/configs/grpo_integration_test.toml
tests/configs/sft_integration_deepseek_simple.toml
tests/configs/sft_integration_test.toml
tests/configs/test_simple_grpo.toml
tests/configs/test_simple_sft.toml
tests/custom_dataset/custom_gsm8k_grpo.py
tests/data/test_data_packer.mp4
tests/data/test_hf_model.jpg
tests/data/test_policy_extract_pp_1_fsdp_1_tp_1.npy
tests/data/test_policy_extract_pp_1_fsdp_2_tp_2.npy
tests/data/test_policy_extract_pp_2_fsdp_2_tp_1.npy
tests/data/test_rollout_extract_pp_1_fsdp_1_tp_1.npy
tests/data/test_rollout_extract_pp_1_fsdp_1_tp_4.npy
tests/data_fixtures/sharegpt52k_small/dataset_dict.json
tests/data_fixtures/sharegpt52k_small/train/data-00000-of-00001.arrow
tests/data_fixtures/sharegpt52k_small/train/dataset_info.json
tests/data_fixtures/sharegpt52k_small/train/state.json
tests/data_fixtures/test_dataset/dataset_dict.json
tests/data_fixtures/test_dataset/train/data-00000-of-00001.arrow
tests/data_fixtures/test_dataset/train/dataset_info.json
tests/data_fixtures/test_dataset/train/state.json
tests/utils/mock_policy_entrance.py
tests/utils/mock_rollout_entrance.py
tests/utils/test_async_utils.py
tests/utils/test_tensor_util_ipc.py