Skip to content

Change services to actors except Policy and drop apps/rl #231

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 17 commits into from
Sep 29, 2025
Merged

Change services to actors except Policy and drop apps/rl #231

Show file tree
Hide file tree
Changes from 8 commits
Commits
Show all changes
17 commits
Select commit Hold shift + click to select a range
c91649b
grpo main
DNXie Sep 24, 2025
e7ba81e
move service config to actor section
DNXie Sep 24, 2025
df2e04c
update toyrl
DNXie Sep 24, 2025
e9fb976
app/rl
DNXie Sep 25, 2025
17c7b5b
fix rl/main
DNXie Sep 25, 2025
5e79423
fix rl.main
DNXie Sep 25, 2025
fca8156
fix toy_rl, drop rl
DNXie Sep 25, 2025
2a9870a
Merge remote-tracking branch 'origin/main' into app_service2actor
DNXie Sep 25, 2025
5aa4029
only trainer and rb are actors
DNXie Sep 26, 2025
ff7b435
for easier comparison
DNXie Sep 26, 2025
01c5b64
Merge branch 'main' into app_service2actor
DNXie Sep 26, 2025
5e521d0
change dataloader etc to actor
DNXie Sep 26, 2025
6b00fcb
Merge branch 'app_service2actor' of github.com:DNXie/forge into app_s…
DNXie Sep 26, 2025
bc95d9c
fix lint
DNXie Sep 26, 2025
15dbedd
Merge branch 'main' into app_service2actor
DNXie Sep 29, 2025
1d07ede
use call_one instead of choose
DNXie Sep 29, 2025
de7a47a
fix toy_rl errors.
DNXie Sep 29, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 6 additions & 6 deletions apps/grpo/main.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -259,17 +259,17 @@ async def main(cfg: DictConfig):
ref_model,
reward_actor,
) = await asyncio.gather(
DatasetActor.options(**cfg.services.dataset).as_service(**cfg.dataset),
DatasetActor.options(**cfg.actors.dataset).as_actor(**cfg.dataset),
Policy.options(**cfg.services.policy).as_service(**cfg.policy),
RLTrainer.options(**cfg.services.trainer).as_service(
RLTrainer.options(**cfg.actors.trainer).as_actor(
**cfg.trainer, loss=simple_grpo_loss
),
ReplayBuffer.options(**cfg.services.replay_buffer).as_service(
ReplayBuffer.options(**cfg.actors.replay_buffer).as_actor(
**cfg.replay_buffer, collate=collate
),
ComputeAdvantages.options(**cfg.services.compute_advantages).as_service(),
ReferenceModel.options(**cfg.services.ref_model).as_service(**cfg.ref_model),
RewardActor.options(**cfg.services.reward_actor).as_service(
ComputeAdvantages.options(**cfg.actors.compute_advantages).as_actor(),
ReferenceModel.options(**cfg.actors.ref_model).as_actor(**cfg.ref_model),
RewardActor.options(**cfg.actors.reward_actor).as_actor(
reward_functions=[MathReward(), ThinkingReward()]
),
)
Expand Down
14 changes: 5 additions & 9 deletions apps/grpo/qwen3_1_7b.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -100,31 +100,27 @@ ref_model:

# All resource allocations
services:
dataset:
procs: 1
num_replicas: 1
with_gpus: false
policy:
procs: ${policy.engine_config.tensor_parallel_size}
num_replicas: 1
with_gpus: true

actors:
dataset:
procs: 1
with_gpus: false
trainer:
procs: 1
num_replicas: 1
with_gpus: true
replay_buffer:
procs: 1
num_replicas: 1
with_gpus: false
ref_model:
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

actually I was wrong - I think we only have trainer and replay buffer be actors, the rest are ok to keep as services

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It makes sense. Why are dataset, compute_advantages, and reward_actor also services? Do they need replicas?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

If num_replica=1, what's the difference between actor and service?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Done!

Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

sorry Danning, Dataset is an actor, compute_advantages is an actor, reward_actor is a service

procs: 1
num_replicas: 1
with_gpus: true
compute_advantages:
procs: 1
num_replicas: 1
with_gpus: false
reward_actor:
procs: 1
num_replicas: 1
with_gpus: false
14 changes: 5 additions & 9 deletions apps/grpo/qwen3_8b.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -101,31 +101,27 @@ ref_model:

# All resource allocations
services:
dataset:
procs: 1
num_replicas: 1
with_gpus: false
policy:
procs: ${policy.engine_config.tensor_parallel_size}
num_replicas: 1
with_gpus: true

actors:
dataset:
procs: 1
with_gpus: false
trainer:
procs: 2
num_replicas: 1
with_gpus: true
replay_buffer:
procs: 1
num_replicas: 1
with_gpus: false
ref_model:
procs: 1
num_replicas: 1
with_gpus: true
compute_advantages:
procs: 1
num_replicas: 1
with_gpus: false
reward_actor:
procs: 1
num_replicas: 1
with_gpus: false
14 changes: 5 additions & 9 deletions apps/grpo/qwen3_multinode.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -46,33 +46,29 @@ ref_model:
model_name: ${model}

services:
dataset:
procs: 1
num_replicas: 1
with_gpus: false
policy:
procs: 1
hosts: 1
num_replicas: 1
with_gpus: true

actors:
dataset:
procs: 1
with_gpus: false
trainer:
procs: 1
hosts: 1
num_replicas: 1
with_gpus: true
replay_buffer:
procs: 1
num_replicas: 1
with_gpus: false
compute_advantages:
procs: 1
num_replicas: 1
with_gpus: false
ref_model:
procs: 1
num_replicas: 1
with_gpus: true
reward_actor:
procs: 1
num_replicas: 1
with_gpus: false
5 changes: 0 additions & 5 deletions apps/rl/__init__.py
View file
Open in desktop

This file was deleted.

62 changes: 0 additions & 62 deletions apps/rl/llama3_8b.yaml
View file
Open in desktop

This file was deleted.

182 changes: 0 additions & 182 deletions apps/rl/main.py
View file
Open in desktop

This file was deleted.

Loading
Loading