Skip to content

Actions: verl-project/verl

Actions

e2e_one_step_off_policy_ascend

Actions

Loading...
Loading

Show workflow options

Create status badge

Loading
e2e_one_step_off_policy_ascend.yml
540 workflow runs
540 workflow runs

Filter by Event

Filter by Status

Filter by Branch

Filter by Actor

[fsdp,vllm,trainer,algo] feat: On-Policy Distillation e2e_one_step_off_policy_ascend #502: Pull request #4897 synchronize by JacobHelwig
14m 24s JacobHelwig:jhelwig/onPolicyDistillation
JacobHelwig:jhelwig/onPolicyDistillation
14m 24s
[fsdp,vllm,trainer,algo] feat: On-Policy Distillation e2e_one_step_off_policy_ascend #501: Pull request #4897 synchronize by JacobHelwig
22m 27s JacobHelwig:jhelwig/onPolicyDistillation
JacobHelwig:jhelwig/onPolicyDistillation
22m 27s
[fsdp,vllm,trainer,algo] feat: On-Policy Distillation e2e_one_step_off_policy_ascend #500: Pull request #4897 synchronize by JacobHelwig
7m 50s JacobHelwig:jhelwig/onPolicyDistillation
JacobHelwig:jhelwig/onPolicyDistillation
7m 50s
[fsdp,vllm,trainer,algo] feat: On-Policy Distillation e2e_one_step_off_policy_ascend #499: Pull request #4897 synchronize by JacobHelwig
24m 9s JacobHelwig:jhelwig/onPolicyDistillation
JacobHelwig:jhelwig/onPolicyDistillation
24m 9s
[megatron] feat: enhance model offloading and loading for frozen para… e2e_one_step_off_policy_ascend #498: Commit b5979db pushed by vermouth1992
58m 0s main
main
58m 0s
[megatron] fix: missing model offload to CPU for forward_only mode (#… e2e_one_step_off_policy_ascend #497: Commit 6b0bff3 pushed by vermouth1992
14m 27s main
main
14m 27s
[algo] fix: seq mean and default scale factor loss_mask.shape[-1] a… e2e_one_step_off_policy_ascend #495: Commit b8d91ef pushed by tongyx361
1h 0m 56s main
main
1h 0m 56s
[trainer] feat: add padding for tensor alignment in preprocess_thd_no_padding function e2e_one_step_off_policy_ascend #494: Pull request #5410 synchronize by RobotGF
47m 53s RobotGF:fix_mcore_cp
RobotGF:fix_mcore_cp
47m 53s
[megatron] feat: enhance model offloading and loading for frozen parameters e2e_one_step_off_policy_ascend #493: Pull request #5412 synchronize by RobotGF
1h 1m 37s RobotGF:fix_lora_offload
RobotGF:fix_lora_offload
1h 1m 37s
[ckpt] feat: implement large tensor slicing in vllm rollout and CheckpointEngine for weight updating e2e_one_step_off_policy_ascend #492: Pull request #5378 synchronize by jianjunzhong
54m 46s jianjunzhong:feat/chunked_weight_update
jianjunzhong:feat/chunked_weight_update
54m 46s
[algo] fix: seq mean and default scale factor loss_mask.shape[-1] as in seq-mean-token-sum-norm e2e_one_step_off_policy_ascend #491: Pull request #5417 opened by tongyx361
1h 1m 12s tongyx361:tyx/fix/seq-mean-in-seq-mean-token-sum-norm
tongyx361:tyx/fix/seq-mean-in-seq-mean-token-sum-norm
1h 1m 12s
[tool] feature: scheduling analysis based on profiling data for torch profiler e2e_one_step_off_policy_ascend #490: Pull request #5367 synchronize by Rhetee
Action required Rhetee:main
Rhetee:main
Action required
[perf, trtllm] feat: Add Nsight support for rollout server mode (trtllm) e2e_one_step_off_policy_ascend #489: Pull request #5391 synchronize by davidmlw
15m 13s joyang-nv:liweim/nsys
joyang-nv:liweim/nsys
15m 13s
[Megatron] feat: Support routing replay on NPU with performance and compatibility enhancements e2e_one_step_off_policy_ascend #488: Pull request #5298 synchronize by 755651978
1h 0m 46s 755651978:main-0212
755651978:main-0212
1h 0m 46s
why are there multiple settings for actor_rollout_ref.model.enable_gradient_checkpointing? Is this a deliberate design choice? e2e_one_step_off_policy_ascend #487: Pull request #4263 synchronize by khazic
Action required khazic:main
khazic:main
Action required
[algo] feat: add DPPO with binary TV or binary KL implementation (#5397) e2e_one_step_off_policy_ascend #486: Commit 182383b pushed by tongyx361
1h 0m 45s main
main
1h 0m 45s
[Megatron] feat: Support routing replay on NPU with performance and compatibility enhancements e2e_one_step_off_policy_ascend #485: Pull request #5298 synchronize by 755651978
1h 0m 33s 755651978:main-0212
755651978:main-0212
1h 0m 33s
[Megatron] feat: Support routing replay on NPU with performance and compatibility enhancements e2e_one_step_off_policy_ascend #483: Pull request #5298 synchronize by 755651978
14m 24s 755651978:main-0212
755651978:main-0212
14m 24s
why are there multiple settings for actor_rollout_ref.model.enable_gradient_checkpointing? Is this a deliberate design choice? e2e_one_step_off_policy_ascend #482: Pull request #4263 synchronize by khazic
Action required khazic:main
khazic:main
Action required
[fsdp,algo] feat: Support QAT (NVFP4) in FSDPEngine for the unified engine_workers architecture e2e_one_step_off_policy_ascend #481: Pull request #5411 synchronize by zhangyimi
Action required zhangyimi:qat-core-v2
zhangyimi:qat-core-v2
Action required
[fsdp,algo] feat: Support QAT (NVFP4) in FSDPEngine for the unified engine_workers architecture e2e_one_step_off_policy_ascend #480: Pull request #5411 synchronize by zhangyimi
Action required zhangyimi:qat-core-v2
zhangyimi:qat-core-v2
Action required
[fsdp,algo] feat: Support QAT (NVFP4) in FSDPEngine for the unified engine_workers architecture e2e_one_step_off_policy_ascend #479: Pull request #5411 opened by zhangyimi
Action required zhangyimi:qat-core-v2
zhangyimi:qat-core-v2
Action required
[misc,trainer,rollout] feat: add Prometheus metrics logging to experiment tracking e2e_one_step_off_policy_ascend #477: Pull request #5291 synchronize by guillemgt
Action required guillemgt:guillem.tarrach/upstream-prometheus-metrics
guillemgt:guillem.tarrach/upstream-prometheus-metrics
Action required
[misc,trainer,rollout] feat: add Prometheus metrics logging to experiment tracking e2e_one_step_off_policy_ascend #476: Pull request #5291 synchronize by guillemgt
Action required guillemgt:guillem.tarrach/upstream-prometheus-metrics
guillemgt:guillem.tarrach/upstream-prometheus-metrics
Action required
[trainer] feat: add support for the GDPO algorithm e2e_one_step_off_policy_ascend #475: Pull request #5409 opened by yue-zeng-yue
Action required yue-zeng-yue:feat-gdpo
yue-zeng-yue:feat-gdpo
Action required