Skip to content

fix(nnx): fix gpt3 paxml checkpoint conversion state map and layout #4555

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

Draft
hsuan-lun-chiang wants to merge 1 commit into
base: main
Choose a base branch
from
+15 −0
Draft

fix(nnx): fix gpt3 paxml checkpoint conversion state map and layout #4555

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
15 changes: 15 additions & 0 deletions src/maxtext/checkpoint_conversion/standalone_scripts/convert_gpt3_ckpt_from_paxml.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -213,6 +213,7 @@ def init_state_fn():
state_map = {
"['optimizer']['step'].value": ("step", None),
"['optimizer']['opt_state'][0]['count'].value": ("opt_states_0.no_prefix_0.count", None),
"['optimizer']['opt_state']['count'].value": ("opt_states_0.no_prefix_0.count", None),
}
else:
state_map = {
Expand All @@ -237,10 +238,18 @@ def get_layer_prefix(keystr_pax):
f"opt_states_0.{prefix_pax_opt_state}.m{keystr_pax}",
transform_fn,
)
state_map[f"['optimizer']['opt_state']['mu']{keystr_maxtext}.value"] = (
f"opt_states_0.{prefix_pax_opt_state}.m{keystr_pax}",
transform_fn,
)
state_map[f"['optimizer']['opt_state'][0]['nu']{keystr_maxtext}.value"] = (
f"opt_states_0.{prefix_pax_opt_state}.v{keystr_pax}",
transform_fn,
)
state_map[f"['optimizer']['opt_state']['nu']{keystr_maxtext}.value"] = (
f"opt_states_0.{prefix_pax_opt_state}.v{keystr_pax}",
transform_fn,
)
else:
state_map[f".params['params']{keystr_maxtext}"] = (f"mdl_vars{keystr_pax}", transform_fn)
state_map[f".opt_state.mu['params']{keystr_maxtext}"] = (
Expand All @@ -254,6 +263,8 @@ def get_layer_prefix(keystr_pax):

def verify_fn(key_path, _):
keystr = jax.tree_util.keystr(key_path)
if "['rngs']" in keystr:
return
assert keystr in state_map, f"{keystr} not found"

jax.tree_util.tree_map_with_path(verify_fn, state)
Expand All @@ -264,6 +275,8 @@ def verify_fn(key_path, _):

def map_fn(key_path, value):
key_path_str = jax.tree_util.keystr(key_path)
if "['rngs']" in key_path_str:
return value
file_path, transform_fn = state_map[key_path_str]
full_path = os.path.join(paxml_ckpt_prefix, file_path)
spec = {"driver": "zarr", "metadata_key": ".zarray", "kvstore": {}}
Expand Down Expand Up @@ -303,6 +316,8 @@ def map_fn(key_path, value):
max_utils.print_mem_stats("converted state finished")

step_value = int(converted_state.optimizer.step.value) if cfg.pure_nnx else converted_state.step
if cfg.pure_nnx:
converted_state = train_state_nnx.to_checkpoint_dict(converted_state)
if checkpointing.save_checkpoint(checkpoint_manager, step_value, converted_state):
max_logging.log(f"saved a checkpoint at step {step_value}")
# Upon preemption, exit when and only when all ongoing saves are complete.
Expand Down
Loading