File size: 8,251 Bytes
3671737
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
"""Checkpoint management with FAST tokenizer saving.

Reference: https://github.com/Physical-Intelligence/openpi
"""

from __future__ import annotations

import asyncio
import concurrent.futures as futures
import dataclasses
import logging
import shutil
from typing import Protocol

from etils import epath
import flax.traverse_util
import jax
import flax.nnx as nnx

import orbax.checkpoint as ocp
import orbax.checkpoint.future as future

from openpi.shared import array_typing as at
import openpi.training.data_loader as _data_loader
import openpi.training.utils as training_utils

# Use our custom normalize (has per_timestamp fields and can save/load them)
from b1k.shared import normalize as _normalize


def initialize_checkpoint_dir(
    checkpoint_dir: epath.Path | str, *, keep_period: int | None, overwrite: bool, resume: bool
) -> tuple[ocp.CheckpointManager, bool]:
    checkpoint_dir = epath.Path(checkpoint_dir).resolve()
    resuming = False
    if checkpoint_dir.exists():
        if overwrite:
            checkpoint_dir.rmtree()
            checkpoint_dir.mkdir(parents=True, exist_ok=True)
            logging.info(f"Wiped checkpoint directory {checkpoint_dir}")
        elif resume:
            resuming = True
        else:
            raise FileExistsError(
                f"Checkpoint directory {checkpoint_dir} already exists. Use --overwrite or --resume "
                "to indicate how to handle it."
            )

    checkpoint_dir.mkdir(parents=True, exist_ok=True)

    mngr = ocp.CheckpointManager(
        checkpoint_dir,
        item_handlers={
            "assets": CallbackHandler(),
            "train_state": ocp.PyTreeCheckpointHandler(),
            "params": ocp.PyTreeCheckpointHandler(),
        },
        options=ocp.CheckpointManagerOptions(
            max_to_keep=1,
            keep_period=keep_period,
            create=False,
            async_options=ocp.AsyncOptions(timeout_secs=7200),
        ),
    )

    # Special case: the checkpoint directory exists and the user requests to resume training, but the training run did
    # not get to the first checkpoint saved. In this case, we don't actually want the train script to try and restore a
    # checkpoint, since it will fail.
    if resuming and tuple(mngr.all_steps()) in [(), (0,)]:
        logging.info("Checkpoint directory exists, but does not contain any checkpoints. Aborting resume.")
        resuming = False

    return mngr, resuming


def save_state(
    checkpoint_manager: ocp.CheckpointManager,
    state: training_utils.TrainState,
    data_loader: _data_loader.DataLoader,
    step: int,
):
    def save_assets(directory: epath.Path):
        # Save the normalization stats.
        data_config = data_loader.data_config()
        norm_stats = data_config.norm_stats
        if norm_stats is not None and data_config.asset_id is not None:
            _normalize.save(directory / data_config.asset_id, norm_stats)
        
        # Save FAST tokenizer if it exists
        model = nnx.merge(state.model_def, state.params)
        # Check if model has FAST enabled (indicated by having fast_token_embedding)
        if hasattr(model, 'fast_token_embedding'):
            # Get the checkpoint directory to determine assets base dir
            # Directory structure: checkpoint_dir/step/assets/asset_id/fast_tokenizer
            # We need to copy from: assets_base_dir/asset_id/fast_tokenizer
            # The assets_base_dir is stored in the config, but we can infer it from checkpoint_dir
            checkpoint_dir = checkpoint_manager.directory
            
            # Infer assets_base_dir from checkpoint structure
            # checkpoint_dir is like: ./outputs/checkpoints/config_name/exp_name
            # assets_base_dir is like: ./outputs/assets/config_name
            parts = checkpoint_dir.parts
            if 'checkpoints' in parts:
                idx = parts.index('checkpoints')
                assets_base_parts = parts[:idx] + ('assets',) + (parts[idx + 1],)  # Keep config_name
                assets_base_dir = epath.Path(*assets_base_parts)
            else:
                # Fallback: try relative to checkpoint dir
                assets_base_dir = checkpoint_dir.parent.parent / 'assets' / checkpoint_dir.parent.name
            
            fast_tokenizer_source = assets_base_dir / data_config.asset_id / "fast_tokenizer"
            fast_tokenizer_dest = directory / data_config.asset_id / "fast_tokenizer"
            
            if fast_tokenizer_source.exists():
                # Create parent directory
                fast_tokenizer_dest.parent.mkdir(parents=True, exist_ok=True)
                # Copy tokenizer directory
                shutil.copytree(fast_tokenizer_source, fast_tokenizer_dest, dirs_exist_ok=True)
                logging.info(f"Saved FAST tokenizer to checkpoint: {fast_tokenizer_dest}")

    # Split params that can be used for inference into a separate item.
    with at.disable_typechecking():
        train_state, params = _split_params(state)
    items = {
        "assets": save_assets,
        "train_state": train_state,
        "params": {"params": params},
    }
    checkpoint_manager.save(step, items)


def restore_state(
    checkpoint_manager: ocp.CheckpointManager,
    state: training_utils.TrainState,
    data_loader: _data_loader.DataLoader,
    step: int | None = None,
) -> training_utils.TrainState:
    del data_loader

    with at.disable_typechecking():
        # Split params that can be used for inference into a separate item.
        train_state, params = _split_params(state)
        
        restored = checkpoint_manager.restore(
            step,
            items={
                "train_state": train_state,
                "params": {"params": params},
            },
        )
                
    return _merge_params(restored["train_state"], restored["params"])


def load_norm_stats(assets_dir: epath.Path | str, asset_id: str) -> dict[str, _normalize.NormStats] | None:
    norm_stats_dir = epath.Path(assets_dir) / asset_id
    norm_stats = _normalize.load(norm_stats_dir)
    logging.info(f"Loaded norm stats from {norm_stats_dir}")
    return norm_stats


class Callback(Protocol):
    def __call__(self, directory: epath.Path) -> None: ...


class CallbackHandler(ocp.AsyncCheckpointHandler):
    """A CheckpointHandler for calling an arbitrary function asynchronously. Only for saving, not for restoring."""

    def save(self, directory: epath.Path, args: CallbackSave):
        if jax.process_index() == 0:
            args.callback(directory)

    async def async_save(self, directory: epath.Path, args: CallbackSave) -> list[futures.Future]:
        return [future.CommitFutureAwaitingContractedSignals(asyncio.to_thread(self.save, directory, args))]

    def restore(self, *args, **kwargs):
        raise NotImplementedError("CallbackHandler does not support restore")


@ocp.args.register_with_handler(CallbackHandler, for_save=True)
@dataclasses.dataclass
class CallbackSave(ocp.args.CheckpointArgs):
    callback: Callback


@ocp.args.register_with_handler(CallbackHandler, for_restore=True)
class CallbackRestore(ocp.args.CheckpointArgs): ...


def _split_params(state: training_utils.TrainState) -> tuple[training_utils.TrainState, at.Params]:
    if state.ema_params is not None:
        params = state.ema_params
        train_state = dataclasses.replace(state, ema_params=None)
    else:
        params = state.params
        train_state = dataclasses.replace(state, params={})
    return train_state, params


def _merge_params(train_state: training_utils.TrainState, params: dict[str, at.Params]) -> training_utils.TrainState:
    # Revert the logic inside `_split_params`.
    # If the restored train_state has an empty `params` dict, it means we're in the non-EMA case
    # and we should restore the main training params.
    if not train_state.params:
        # Non-EMA case: The saved 'params' are the training weights.
        return dataclasses.replace(train_state, params=params["params"])
    else:
        # EMA case: The saved 'params' are the ema_params, and train_state already has the training params.
        return dataclasses.replace(train_state, ema_params=params["params"])