Skip to content

Actions: verl-project/verl

Actions

e2e_ppo_trainer_megatron_vllm_2

Actions

Loading...
Loading

Show workflow options

Create status badge

Loading
e2e_ppo_trainer_megatron_vllm_2.yml
3,399 workflow runs
3,399 workflow runs

Filter by Event

Filter by Status

Filter by Branch

Filter by Actor

[BREAKING][megatron] feat: support Megatron-FSDP as a new training backend e2e_ppo_trainer_megatron_vllm_2 #4587: Pull request #5423 opened by conver334
Action required conver334:megatron-fsdp
conver334:megatron-fsdp
Action required
[WIP][algo] Migrate and implement the GDPO algorithm into the existing framework. e2e_ppo_trainer_megatron_vllm_2 #4586: Pull request #5422 synchronize by Rhetee
Action required Rhetee:GDPO
Rhetee:GDPO
Action required
[megatron, vllm] feat: NVFP4 (W4A16) QAT training support via ModelOpt e2e_ppo_trainer_megatron_vllm_2 #4585: Pull request #5254 synchronize by jQizhang
Action required jQizhang:nvfp4_cleanup
jQizhang:nvfp4_cleanup
Action required
[WIP][algo] Migrate and implement the GDPO algorithm into the existing framework. e2e_ppo_trainer_megatron_vllm_2 #4584: Pull request #5422 opened by Rhetee
Action required Rhetee:GDPO
Rhetee:GDPO
Action required
[tool] feature: scheduling analysis based on profiling data for torch profiler e2e_ppo_trainer_megatron_vllm_2 #4583: Pull request #5367 synchronize by Rhetee
Action required Rhetee:main
Rhetee:main
Action required
[rollout] feat: Fix partial load problem, Add vlm support for trtllm rollout e2e_ppo_trainer_megatron_vllm_2 #4582: Pull request #5149 synchronize by SchumiDing
Action required SchumiDing:vlm_trtllm_support
SchumiDing:vlm_trtllm_support
Action required
[megatron, vllm] feat: NVFP4 (W4A16) QAT training support via ModelOpt e2e_ppo_trainer_megatron_vllm_2 #4581: Pull request #5254 synchronize by jQizhang
Action required jQizhang:nvfp4_cleanup
jQizhang:nvfp4_cleanup
Action required
[perf, trtllm] feat: Add Nsight support for rollout server mode (trtllm) e2e_ppo_trainer_megatron_vllm_2 #4580: Pull request #5391 synchronize by davidmlw
1h 47m 42s joyang-nv:liweim/nsys
joyang-nv:liweim/nsys
1h 47m 42s
[perf, trtllm] feat: Add Nsight support for rollout server mode (trtllm) e2e_ppo_trainer_megatron_vllm_2 #4579: Pull request #5391 synchronize by davidmlw
4m 8s joyang-nv:liweim/nsys
joyang-nv:liweim/nsys
4m 8s
[rollout] refactor: bucketed transfer utils e2e_ppo_trainer_megatron_vllm_2 #4578: Pull request #5309 synchronize by pengwu22
1h 51m 38s pengwu22:pw/test-wt-utils
pengwu22:pw/test-wt-utils
1h 51m 38s
[trainer] feat: add padding for tensor alignment in preprocess_thd_no… e2e_ppo_trainer_megatron_vllm_2 #4577: Commit 32705dc pushed by vermouth1992
1h 30m 16s main
main
1h 30m 16s
why are there multiple settings for actor_rollout_ref.model.enable_gradient_checkpointing? Is this a deliberate design choice? e2e_ppo_trainer_megatron_vllm_2 #4576: Pull request #4263 synchronize by khazic
Action required khazic:main
khazic:main
Action required
[megatron, vllm] feat: NVFP4 (W4A16) QAT training support via ModelOpt e2e_ppo_trainer_megatron_vllm_2 #4575: Pull request #5254 synchronize by jQizhang
Action required jQizhang:nvfp4_cleanup
jQizhang:nvfp4_cleanup
Action required
[perf] fix: the overwritten of Torch_profile with multi steps. (#5395) e2e_ppo_trainer_megatron_vllm_2 #4574: Commit 5f7c345 pushed by wuxibin89
39m 9s main
main
39m 9s
[rollout] refactor: bucketed transfer utils e2e_ppo_trainer_megatron_vllm_2 #4573: Pull request #5309 synchronize by pengwu22
42m 43s pengwu22:pw/test-wt-utils
pengwu22:pw/test-wt-utils
42m 43s
[Megatron] feat: Support routing replay on NPU with performance and compatibility enhancements e2e_ppo_trainer_megatron_vllm_2 #4572: Pull request #5298 synchronize by 755651978
1h 25m 40s 755651978:main-0212
755651978:main-0212
1h 25m 40s
[fsdp,vllm,trainer,algo] feat: On-Policy Distillation e2e_ppo_trainer_megatron_vllm_2 #4571: Pull request #4897 synchronize by JacobHelwig
21m 57s JacobHelwig:jhelwig/onPolicyDistillation
JacobHelwig:jhelwig/onPolicyDistillation
21m 57s
[fsdp,vllm,trainer,algo] feat: On-Policy Distillation e2e_ppo_trainer_megatron_vllm_2 #4570: Pull request #4897 synchronize by JacobHelwig
3m 3s JacobHelwig:jhelwig/onPolicyDistillation
JacobHelwig:jhelwig/onPolicyDistillation
3m 3s
[fsdp,vllm,trainer,algo] feat: On-Policy Distillation e2e_ppo_trainer_megatron_vllm_2 #4569: Pull request #4897 synchronize by JacobHelwig
18m 55s JacobHelwig:jhelwig/onPolicyDistillation
JacobHelwig:jhelwig/onPolicyDistillation
18m 55s
[rollout] refactor: bucketed transfer utils e2e_ppo_trainer_megatron_vllm_2 #4568: Pull request #5309 synchronize by pengwu22
27m 7s pengwu22:pw/test-wt-utils
pengwu22:pw/test-wt-utils
27m 7s
[tool] fix: handle empty image inputs in ToolAgentLoop e2e_ppo_trainer_megatron_vllm_2 #4567: Pull request #5420 opened by denismegerle
2h 24m 22s denismegerle:public/tool-agent-downstream-image-processing-compat
denismegerle:public/tool-agent-downstream-image-processing-compat
2h 24m 22s
[fsdp, model]{fix} Allow flash attention 2 to be used for NemotronH model on FSDP e2e_ppo_trainer_megatron_vllm_2 #4566: Pull request #5419 opened by thvasilo
Action required thvasilo:enable-nemotron-h-fa2
thvasilo:enable-nemotron-h-fa2
Action required
[megatron] fix: patch support newer mcore version e2e_ppo_trainer_megatron_vllm_2 #4565: Pull request #5372 synchronize by HollowMan6
22m 51s HollowMan6:mcore_patch
HollowMan6:mcore_patch
22m 51s
[fsdp,vllm,trainer,algo] feat: On-Policy Distillation e2e_ppo_trainer_megatron_vllm_2 #4564: Pull request #4897 synchronize by JacobHelwig
1h 7m 17s JacobHelwig:jhelwig/onPolicyDistillation
JacobHelwig:jhelwig/onPolicyDistillation
1h 7m 17s
[BREAKING][rollout,cfg] refactor: get rid of actor_rollout_ref config from rollout e2e_ppo_trainer_megatron_vllm_2 #4563: Pull request #5418 synchronize by wuxibin89
43m 49s wuxibin89:wuxibin/refactor_config
wuxibin89:wuxibin/refactor_config
43m 49s