Skip to content

Actions: verl-project/verl

Actions

sgl

Actions

Loading...
Loading

Show workflow options

Create status badge

Loading
sgl.yml
6,206 workflow runs
6,206 workflow runs

Filter by Event

Filter by Status

Filter by Branch

Filter by Actor

[misc] fix: add missing __init__.py files to package directories sgl #8412: Pull request #5209 opened by guillemgt
Action required guillemgt:add-missing-init-files
guillemgt:add-missing-init-files
Action required
[trtllm] fix: add synchronization before resume kv_cache to prevent oom in non-leader ranks sgl #8411: Pull request #5208 opened by shuyixiong
Action required shuyixiong:shuyix/optimize_memory_usage
shuyixiong:shuyix/optimize_memory_usage
Action required
[reward] fix: preserve input non_tensor_batch in AgentLoopManager when reward_loop_worker_handles is None sgl #8410: Pull request #5195 synchronize by none0663
21m 24s none0663:agent-loop-preserve-non-tensor-batch
none0663:agent-loop-preserve-non-tensor-batch
21m 24s
[recipe] refactor: refactor ray trainer for separate recipe use. (fully async / one step off) sgl #8409: Pull request #5184 synchronize by ArronHZG
25m 45s meituan-search:refactor_fully_async_ray_trainer
meituan-search:refactor_fully_async_ray_trainer
25m 45s
[recipe] refactor: refactor ray trainer for separate recipe use. (fully async / one step off) sgl #8408: Pull request #5184 synchronize by ArronHZG
18m 44s meituan-search:refactor_fully_async_ray_trainer
meituan-search:refactor_fully_async_ray_trainer
18m 44s
[recipe] refactor: refactor ray trainer for separate recipe use. (fully async / one step off) sgl #8407: Pull request #5184 synchronize by ArronHZG
17m 58s meituan-search:refactor_fully_async_ray_trainer
meituan-search:refactor_fully_async_ray_trainer
17m 58s
[recipe] refactor: refactor ray trainer for separate recipe use. (fully async / one step off) sgl #8406: Pull request #5184 synchronize by ArronHZG
10m 27s meituan-search:refactor_fully_async_ray_trainer
meituan-search:refactor_fully_async_ray_trainer
10m 27s
[perf] feat: add images_seqlens on mfu calculation for engine_worker sgl #8405: Pull request #5207 synchronize by tardis-key
20m 22s alwaysyiyu:engineworker_mfu
alwaysyiyu:engineworker_mfu
20m 22s
[perf] feat: add images_seqlens on mfu calculation for engine_worker sgl #8404: Pull request #5207 opened by alwaysyiyu
Action required alwaysyiyu:engineworker_mfu
alwaysyiyu:engineworker_mfu
Action required
[rollout] feat: Fix partial load problem, Add vlm support for trtllm rollout sgl #8403: Pull request #5149 synchronize by SchumiDing
Action required SchumiDing:vlm_trtllm_support
SchumiDing:vlm_trtllm_support
Action required
[algo] feat: support rollout router replay in MegatronEngine sgl #8402: Pull request #5185 synchronize by xhx1022
15m 20s xhx1022:xhx/r3_engine
xhx1022:xhx/r3_engine
15m 20s
[algo] feat: support rollout router replay in MegatronEngine sgl #8401: Pull request #5185 synchronize by xhx1022
5m 10s xhx1022:xhx/r3_engine
xhx1022:xhx/r3_engine
5m 10s
[recipe] refactor: refactor ray trainer for separate recipe use. (fully async / one step off) sgl #8400: Pull request #5184 synchronize by ArronHZG
22m 51s meituan-search:refactor_fully_async_ray_trainer
meituan-search:refactor_fully_async_ray_trainer
22m 51s
[algo] feat: support rollout router replay in MegatronEngine sgl #8399: Pull request #5185 synchronize by xhx1022
9m 14s xhx1022:xhx/r3_engine
xhx1022:xhx/r3_engine
9m 14s
why are there multiple settings for actor_rollout_ref.model.enable_gradient_checkpointing? Is this a deliberate design choice? sgl #8398: 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? sgl #8397: Pull request #4263 synchronize by khazic
Action required khazic:main
khazic:main
Action required
[algo] feat: support rollout router replay in MegatronEngine sgl #8396: Pull request #5185 synchronize by xhx1022
9m 56s xhx1022:xhx/r3_engine
xhx1022:xhx/r3_engine
9m 56s
why are there multiple settings for actor_rollout_ref.model.enable_gradient_checkpointing? Is this a deliberate design choice? sgl #8395: Pull request #4263 synchronize by khazic
Action required khazic:main
khazic:main
Action required
[megatron] fix: add protections for logits_processor_args.pop("loss_m… sgl #8394: Commit ad404cb pushed by ISEEKYAN
19m 29s main
main
19m 29s
why are there multiple settings for actor_rollout_ref.model.enable_gradient_checkpointing? Is this a deliberate design choice? sgl #8393: 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? sgl #8391: Pull request #4263 synchronize by khazic
Action required khazic:main
khazic:main
Action required
[worker, training_utils] fix: Metric Aggregation Across DP Ranks (#5203) sgl #8390: Commit 2320603 pushed by wuxibin89
18m 26s main
main
18m 26s
[reward] refactor: remove reward model worker code and invocation sgl #8389: Pull request #5194 synchronize by yyDing1
18m 30s yyDing1:rm_rm_config
yyDing1:rm_rm_config
18m 30s
[worker, training_utils] fix: Metric Aggregation Across DP Ranks sgl #8388: Pull request #5203 synchronize by JacobHelwig
9m 10s JacobHelwig:jhelwig/fixDPMetrics
JacobHelwig:jhelwig/fixDPMetrics
9m 10s
[worker, training_utils] fix: Metric Aggregation Across DP Ranks sgl #8387: Pull request #5203 synchronize by JacobHelwig
3m 28s JacobHelwig:jhelwig/fixDPMetrics
JacobHelwig:jhelwig/fixDPMetrics
3m 28s