Skip to content

Actions: verl-project/verl

Actions

cpu_unit_tests

Actions

Loading...
Loading

Show workflow options

Create status badge

Loading
cpu_unit_tests.yml
6,809 workflow runs
6,809 workflow runs

Filter by Event

Filter by Status

Filter by Branch

Filter by Actor

[fsdp] fix: pass dp_group to prepare_dynamic_batch to fix potential hang cpu_unit_tests #9267: Pull request #5511 synchronize by dubin555
Action required dubin555:fix/fsdp-dp-group-dynamic-batch
dubin555:fix/fsdp-dp-group-dynamic-batch
Action required
why are there multiple settings for actor_rollout_ref.model.enable_gradient_checkpointing? Is this a deliberate design choice? cpu_unit_tests #9266: Pull request #4263 synchronize by khazic
Action required khazic:main
khazic:main
Action required
[megatron] feat: model engine support mtp cpu_unit_tests #9265: Pull request #5561 synchronize by ArronHZG
1h 4m 50s meituan-search:model_engine_support_mtp
meituan-search:model_engine_support_mtp
1h 4m 50s
why are there multiple settings for actor_rollout_ref.model.enable_gradient_checkpointing? Is this a deliberate design choice? cpu_unit_tests #9264: Pull request #4263 synchronize by khazic
Action required khazic:main
khazic:main
Action required
why are there multiple settings for actor_rollout_ref.model.enable_gradient_checkpointing? Is this a deliberate design choice? cpu_unit_tests #9263: Pull request #4263 synchronize by khazic
Action required khazic:main
khazic:main
Action required
[megatron] feat: model engine support mtp cpu_unit_tests #9262: Pull request #5561 synchronize by ArronHZG
18m 53s meituan-search:model_engine_support_mtp
meituan-search:model_engine_support_mtp
18m 53s
why are there multiple settings for actor_rollout_ref.model.enable_gradient_checkpointing? Is this a deliberate design choice? cpu_unit_tests #9261: Pull request #4263 synchronize by khazic
Action required khazic:main
khazic:main
Action required
[megatron, model] feat: qwen3.5 example (#5381) cpu_unit_tests #9260: Commit ef072ac pushed by wuxibin89
1h 26m 33s main
main
1h 26m 33s
[megatron, model] feat: qwen3.5 example cpu_unit_tests #9259: Pull request #5381 synchronize by wuxibin89
13m 37s ISEEKYAN:mcore_qwen35
ISEEKYAN:mcore_qwen35
13m 37s
why are there multiple settings for actor_rollout_ref.model.enable_gradient_checkpointing? Is this a deliberate design choice? cpu_unit_tests #9258: Pull request #4263 synchronize by khazic
Action required khazic:main
khazic:main
Action required
[training_utils] fix: Fix total train_time in sft_trainer.py and add 1) step time 2) seq len stats for training data. cpu_unit_tests #9257: Pull request #5579 opened by sheilaliuxl
Action required sheilaliuxl:dev-sft-metrics
sheilaliuxl:dev-sft-metrics
Action required
[fsdp,vllm,trainer,algo] feat: On-Policy Distillation cpu_unit_tests #9256: Pull request #4897 synchronize by JacobHelwig
4m 37s JacobHelwig:jhelwig/onPolicyDistillation
JacobHelwig:jhelwig/onPolicyDistillation
4m 37s
[fsdp,vllm,trainer,algo] feat: On-Policy Distillation cpu_unit_tests #9255: Pull request #4897 synchronize by JacobHelwig
3m 45s JacobHelwig:jhelwig/onPolicyDistillation
JacobHelwig:jhelwig/onPolicyDistillation
3m 45s
[trainer, algo] feat: Atropos RL environment integration with verl GRPO cpu_unit_tests #9254: Pull request #5520 synchronize by 1HazyOne707
Action required 1HazyOne707:atropos-integration
1HazyOne707:atropos-integration
Action required
[megatron] feat: checkpoint save as HF PEFT format cpu_unit_tests #9253: Pull request #5575 synchronize by HollowMan6
6m 25s HollowMan6:lora_megatron_hf
HollowMan6:lora_megatron_hf
6m 25s
[sglang,fsdp] feat: LoRA support for SGLang rollouts (merge + native adapter paths) cpu_unit_tests #9252: Pull request #5564 synchronize by cavities12
25m 35s cavities12:feat/native-lora-sglang
cavities12:feat/native-lora-sglang
25m 35s
[megatron] feat: checkpoint save as HF PEFT format cpu_unit_tests #9251: Pull request #5575 opened by HollowMan6
4m 29s HollowMan6:lora_megatron_hf
HollowMan6:lora_megatron_hf
4m 29s
[sglang,fsdp] feat: LoRA support for SGLang rollouts (merge + native adapter paths) cpu_unit_tests #9250: Pull request #5564 synchronize by cavities12
12m 13s cavities12:feat/native-lora-sglang
cavities12:feat/native-lora-sglang
12m 13s
[tool] refactor: Unify the profiler tools in one place cpu_unit_tests #9249: Pull request #5573 synchronize by zhink
15m 11s zhink:main
zhink:main
15m 11s
[megatron, model] feat: qwen3.5 example cpu_unit_tests #9248: Pull request #5381 synchronize by wuxibin89
14m 14s ISEEKYAN:mcore_qwen35
ISEEKYAN:mcore_qwen35
14m 14s
[megatron, model] feat: qwen3.5 example cpu_unit_tests #9247: Pull request #5381 synchronize by wuxibin89
17m 31s ISEEKYAN:mcore_qwen35
ISEEKYAN:mcore_qwen35
17m 31s
[megatron] fix: avoid mindspeed rope and attention synchronization cpu_unit_tests #9246: Pull request #5574 opened by Shangwei-Li
Action required Shangwei-Li:cu_seqlens_fix
Shangwei-Li:cu_seqlens_fix
Action required
feat(vision): add Vision DP for parallel ViT computation across SP ranks cpu_unit_tests #9245: Pull request #5230 synchronize by aoshen524
Action required aoshen524:feat/vision-dp-parallel
aoshen524:feat/vision-dp-parallel
Action required
feat(vision): add Vision DP for parallel ViT computation across SP ranks cpu_unit_tests #9244: Pull request #5230 synchronize by aoshen524
Action required aoshen524:feat/vision-dp-parallel
aoshen524:feat/vision-dp-parallel
Action required
[tool] refactor: Unify the profiler tools in one place cpu_unit_tests #9243: Pull request #5573 opened by zhink
Action required zhink:main
zhink:main
Action required