Skip to content

Fix SAC checkpoint restore device axis mismatch (#659) #669

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

Open
SAY-5 wants to merge 1 commit into
base: main
Choose a base branch
from
Open

Fix SAC checkpoint restore device axis mismatch (#659) #669

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
28 changes: 15 additions & 13 deletions brax/training/agents/sac/train.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -77,7 +77,6 @@ def _unpmap(v):
def _init_training_state(
key: PRNGKey,
obs_size: int,
local_devices_to_use: int,
sac_network: sac_networks.SACNetworks,
alpha_optimizer: optax.GradientTransformation,
policy_optimizer: optax.GradientTransformation,
Expand Down Expand Up @@ -109,16 +108,7 @@ def _init_training_state(
alpha_params=log_alpha,
normalizer_params=normalizer_params,
)
devices = jax.local_devices()[:local_devices_to_use]
mesh = jax.sharding.Mesh(np.array(devices), ('_device_put_sharded',))
sharding = jax.NamedSharding(mesh, jax.P('_device_put_sharded'))

def _replicate(x):
if isinstance(x, jax.Array):
return jax.device_put(jnp.stack([x] * len(devices)), sharding)
return jax.device_put(np.stack([x] * len(devices)), sharding)

return jax.tree_util.tree_map(_replicate, training_state)
return training_state


def train(
Expand Down Expand Up @@ -491,7 +481,6 @@ def training_epoch_with_timing(
training_state = _init_training_state(
key=global_key,
obs_size=obs_size,
local_devices_to_use=local_devices_to_use,
sac_network=sac_network,
alpha_optimizer=alpha_optimizer,
policy_optimizer=policy_optimizer,
Expand All @@ -506,6 +495,19 @@ def training_epoch_with_timing(
policy_params=params[1],
)

# Replicate training state across devices AFTER checkpoint restoration
# so that restored params have the correct per-device shape. Fixes #659.
devices = jax.local_devices()[:local_devices_to_use]
mesh = jax.sharding.Mesh(np.array(devices), ('_device_put_sharded',))
sharding = jax.NamedSharding(mesh, jax.P('_device_put_sharded'))

def _replicate(x):
if isinstance(x, jax.Array):
return jax.device_put(jnp.stack([x] * len(devices)), sharding)
return jax.device_put(np.stack([x] * len(devices)), sharding)

training_state = jax.tree_util.tree_map(_replicate, training_state)

local_key, rb_key, env_key, eval_key = jax.random.split(local_key, 4)

# Env init
Expand Down Expand Up @@ -624,4 +626,4 @@ def training_epoch_with_timing(
pmap.assert_is_replicated(training_state)
logging.info('total steps: %s', total_steps)
pmap.synchronize_hosts()
return (make_policy, params, metrics)
return (make_policy, params, metrics)