Skip to content

split experience to send #11

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 1 commit into from
Apr 21, 2023
Merged

split experience to send #11

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
17 changes: 16 additions & 1 deletion applications/Chat/coati/ray/src/experience_maker_holder.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -10,6 +10,7 @@
import torch.nn as nn
from coati.experience_maker import Experience, ExperienceMaker, NaiveExperienceMaker
from coati.models.base import Actor, Critic, RewardModel
from coati.replay_buffer.utils import split_experience_batch, make_experience_batch, BufferItem
from coati.trainer.callbacks import Callback
from coati.trainer.callbacks.performance_evaluator import ExperienceMakerPerformanceEvaluator
from coati.trainer.strategies import Strategy
Expand Down Expand Up @@ -47,6 +48,7 @@ def __init__(self,
callbacks: List[Callback] = [],
eval_performance: bool = False,
debug: bool = False,
send_grain_size: int = 4,
**generate_kwargs):
# set environment variables
if env_info:
Expand All @@ -63,6 +65,7 @@ def __init__(self,
self.experience_maker = NaiveExperienceMaker(actor, critic, reward_model, initial_model, self.kl_coef)
self.callbacks = callbacks
self.eval_performance = eval_performance
self.send_grain_size = send_grain_size

self._model_visit_lock = Lock()
self._initial_model_initialized = False
Expand Down Expand Up @@ -165,7 +168,19 @@ def workingloop(self, dataset, tokenizer: Optional[Callable[[Any], dict]] = None
experience = self._make_experience(inputs=inputs)
self._on_make_experience_end(experience)
self._model_visit_lock.release()
self._send_experience(experience=experience)
# split experience for smoother handover
items = split_experience_batch(experience)
temp_buffer = []
for item in items:
temp_buffer.append(item)
if len(temp_buffer) >= self.send_grain_size:
experience_fragment = make_experience_batch(temp_buffer)
self._send_experience(experience=experience_fragment)
temp_buffer = []
# remain
if len(temp_buffer) > 0:
experience_fragment = make_experience_batch(temp_buffer)
self._send_experience(experience=experience_fragment)
self._on_finish()

@ray.method(concurrency_group="model_io")
Expand Down