File size: 16,435 Bytes
74e8f2f
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
# Copyright 2022 Big Vision Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

"""Train loop for training the stage-I model."""
# pylint: disable=consider-using-from-import
import functools
import importlib
import multiprocessing.pool
import os

from absl import app
from absl import flags
from absl import logging
from big_vision import input_pipeline
import big_vision.datasets.core as ds_core
import big_vision.evaluators.common as eval_common
import big_vision.optax as bv_optax
import big_vision.pp.builder as pp_builder
import big_vision.utils as u
from clu import parameter_overview
import flax
import jax
import jax.numpy as jnp
from ml_collections import config_flags
import numpy as np
import optax

import tensorflow.io.gfile as gfile


SG = jax.lax.stop_gradient
partial = functools.partial

config_flags.DEFINE_config_file(
    "config", None, "Training configuration.", lock_config=True)

flags.DEFINE_string("workdir", default=None, help="Work unit directory.")
flags.DEFINE_boolean("cleanup", default=False,
                     help="Delete workdir (only) after successful completion.")

# Adds jax flags to the program.
jax.config.parse_flags_with_absl()


def main(argv):
  del argv

  config = flags.FLAGS.config
  workdir = flags.FLAGS.workdir
  logging.info("Workdir: %s", workdir)

  logging.info("\u001b[33mHello from process %i holding %i/%i devices and "
               "writing to workdir %s.\u001b[0m", jax.process_index(),
               jax.local_device_count(), jax.device_count(), workdir)

  # Define task input, loss and predict functions.
  task_module = importlib.import_module(f"big_vision.trainers.{config.task}")
  input_pp_fn = partial(task_module.input_pp, config=config)
  task_loss_fn = partial(task_module.loss_fn, config=config)
  predict_outputs_fn = partial(task_module.predict_outputs, config=config)

  save_ckpt_path = None
  if workdir:  # Always create if requested, even if we may not write into it.
    gfile.makedirs(workdir)
    save_ckpt_path = os.path.join(workdir, "checkpoint.npz")

  # The pool is used to perform misc operations such as logging in async way.
  pool = multiprocessing.pool.ThreadPool()

  # Here we register preprocessing ops from modules listed on `pp_modules`.
  for m in config.get("pp_modules",
                      ["ops_general", "ops_image", "proj.uvim.pp_ops"]):
    importlib.import_module(f"big_vision.pp.{m}")

  # This seed makes the Jax part of things (like model init) deterministic.
  # However, full training still won't be deterministic, for example due to the
  # tf.data pipeline not being deterministic even if we would set TF seed.
  # See (internal link) for a fun read on what it takes.
  rng = jax.random.PRNGKey(config.get("seed", 0))

  # These functions do more stuff internally, for OSS release we mock them by
  # trivial alternatives in order to minize disruptions in the code.
  xid, wid = -1, -1
  fillin = lambda s: s
  def info(s, *a):
    logging.info("\u001b[33mNOTE\u001b[0m: " + s, *a)
  def write_note(note):
    if jax.process_index() == 0:
      info("%s", note)

  write_note("Initializing...")

  batch_size = config.input.batch_size
  if batch_size % jax.device_count() != 0:
    raise ValueError(f"Batch size ({batch_size}) must "
                     f"be divisible by device number ({jax.device_count()})")
  info("Global batch size %d on %d hosts results in %d local batch size. With "
       "%d dev per host (%d dev total), that's a %d per-device batch size.",
       batch_size, jax.process_count(), batch_size // jax.process_count(),
       jax.local_device_count(), jax.device_count(),
       batch_size // jax.device_count())

  # First thing after above sanity checks, so we can log "start" ticks.
  mw = u.BigVisionMetricWriter(xid, wid, workdir, config)
  chrono = u.Chrono()

  write_note("Initializing train dataset...")
  train_data = ds_core.get(**config.input.data)
  train_ds = input_pipeline.make_for_train(
      data=train_data.get_tfdata(ordered=False),
      batch_size=batch_size,
      preprocess_fn=pp_builder.get_preprocess_fn(config.input.get("pp")),
      shuffle_buffer_size=config.input.get("shuffle_buffer_size"),
      cache_raw=config.input.get("cache_raw", False),
      filter_fn=config.input.get("filter_fn"),
  )

  # Start prefetching already.
  n_prefetch = config.get("prefetch_to_device", 1)
  train_iter = input_pipeline.start_input_pipeline(train_ds, n_prefetch)
  ntrain_img = train_data.total_examples

  def get_steps(name, default=ValueError):  # partial doesn't work well here.
    return u.steps(name, config, ntrain_img, batch_size, default)
  total_steps = get_steps("total")

  info("Running for %d steps, that means %f epochs",
       total_steps, total_steps * batch_size / ntrain_img)

  write_note(f"Initializing {config.model_name} model...")
  model_mod = importlib.import_module(f"big_vision.models.{config.model_name}")
  model = model_mod.Model(**config.model)

  # We want all parameters to be created in host RAM, not on any device, they'll
  # be sent there later as needed, otherwise we already encountered two
  # situations where we allocate them twice.
  @partial(jax.jit, backend="cpu")
  def init(rng):
    batch = jax.tree_map(
        lambda x: jnp.zeros(x.shape, x.dtype.as_numpy_dtype),
        train_ds.element_spec)
    init_res = flax.core.unfreeze(model.init(rng, **input_pp_fn(batch)))
    params, state = init_res["params"], init_res["state"]

    # Set bias in the heads to a low value, such that loss is small initially.
    for key in config.model.outputs:
      params[f"head_{key}"]["bias"] = jnp.full_like(
          params[f"head_{key}"]["bias"], config.get("init_head_bias", 0))

    return params, state

  rng, rng_init = jax.random.split(rng)

  rng_init_params, rng_init_state = jax.random.split(rng_init)
  params_cpu, state_cpu = init({"params": rng_init_params,
                                "state": rng_init_state})

  if jax.process_index() == 0:
    num_params = sum(p.size for p in jax.tree_leaves(params_cpu))
    parameter_overview.log_parameter_overview(params_cpu, msg="init params")
    mw.measure("num_params", num_params)

  write_note(f"Initializing {config.optax_name} optimizer...")
  tx, sched_fns = bv_optax.make(config, params_cpu, sched_kw=dict(
      total_steps=total_steps, batch_size=batch_size, data_size=ntrain_img))

  # We jit this, such that the arrays are created on the CPU, not device[0].
  opt_cpu = jax.jit(tx.init, backend="cpu")(params_cpu)
  sched_fns_cpu = [jax.jit(sched_fn, backend="cpu") for sched_fn in sched_fns]

  @partial(jax.pmap, axis_name="batch", donate_argnums=(0, 1, 2),
           static_broadcasted_argnums=(5,))
  def update_fn(params, opt, state, batch, rng, update_dict=True):
    """Update step."""
    measurements = {}

    # Get device-specific loss rng.
    rng, rng_model = jax.random.split(rng, 2)
    rng_model_local = jax.random.fold_in(rng_model, jax.lax.axis_index("batch"))

    def loss_fn(params, state, batch):
      (logits, out), mutated_col = model.apply(
          {"params": params, "state": state},
          **input_pp_fn(batch),
          train=True, update_dict=update_dict,
          rngs={"dropout": rng_model_local, "vqvae": rng_model},
          mutable=["state"])
      btlneck = out["bottleneck"]
      btlneck_q = out["bottleneck_q"]

      loss_rec, logs = jax.tree_map(jnp.mean, task_loss_fn(logits, batch))
      loss_commitment = jnp.mean(jnp.square(btlneck - SG(btlneck_q)))
      loss = loss_rec + config.get("w_commitment", 0.25) * loss_commitment
      aux = {
          "loss_rec": jax.lax.pmean(loss_rec, axis_name="batch"),
          "loss_commitment": jax.lax.pmean(loss_commitment, axis_name="batch"),
          "codebook_zeros_ratio": out["codebook_zeros_ratio"],
          "codebook_max_ratio": out["codebook_max_ratio"],
          "state": mutated_col["state"],
          **jax.tree_map(partial(jax.lax.pmean, axis_name="batch"), logs),
      }
      return loss, aux

    (l, aux), grads = jax.value_and_grad(loss_fn, has_aux=True)(
        params, state, batch)
    l, grads = jax.lax.pmean((l, grads), axis_name="batch")
    updates, opt = tx.update(grads, opt, params)
    params = optax.apply_updates(params, updates)
    state = aux.pop("state")
    measurements = {**measurements, **aux}

    gs = jax.tree_leaves(bv_optax.replace_frozen(config.schedule, grads, 0.))
    measurements["l2_grads"] = jnp.sqrt(sum([jnp.vdot(g, g) for g in gs]))
    ps = jax.tree_leaves(params)
    measurements["l2_params"] = jnp.sqrt(sum([jnp.vdot(p, p) for p in ps]))
    us = jax.tree_leaves(updates)
    measurements["l2_updates"] = jnp.sqrt(sum([jnp.vdot(u, u) for u in us]))

    return params, opt, state, l, rng, measurements

  # Define evaluators.
  def validation_fn(params, batch):
    """Compute per-example metrics."""
    logits, out = model.apply(params, **input_pp_fn(batch))
    _, losses = task_loss_fn(logits, batch)
    btlneck = out["bottleneck"]
    btlneck_q = out["bottleneck_q"]
    losses["loss_commitment"] = jnp.square(btlneck - btlneck_q)
    return jax.tree_map(
        lambda x: jnp.mean(x, axis=tuple(range(1, x.ndim))),
        losses)

  def predict_fn(params, batch):
    logits, _ = model.apply(params, **input_pp_fn(batch))
    outputs = predict_outputs_fn(logits)
    return outputs

  # Only initialize evaluators when they are first needed.
  @functools.lru_cache(maxsize=None)
  def evaluators():
    return eval_common.from_config(
        config, {"predict": predict_fn, "validation": validation_fn},
        lambda s: write_note(f"Initializing evaluator: {s}...\n{chrono.note}")
    )

  # Decide how to initialize training. The order is important.
  # 1. Always resumes from the existing checkpoint, e.g. resumes a finetune job.
  # 2. Resume from a previous checkpoint, e.g. start a cooldown training job.
  # 3. Initialize model from something, e,g, start a fine-tuning job.
  # 4. Train from scratch.
  resume_ckpt_path = None
  if save_ckpt_path and gfile.exists(save_ckpt_path):
    resume_ckpt_path = save_ckpt_path
  elif config.get("resume"):
    resume_ckpt_path = fillin(config.resume)
  if resume_ckpt_path:
    write_note("Resume training from checkpoint...")
    checkpoint = {
        "params": params_cpu,
        "state": state_cpu,
        "opt": opt_cpu,
        "chrono": chrono.save(),
    }
    checkpoint_tree = jax.tree_structure(checkpoint)
    loaded = u.load_checkpoint(checkpoint_tree, resume_ckpt_path)
    # bfloat16 type gets lost when data is saved to disk, so we recover it.
    checkpoint = jax.tree_map(u.recover_dtype, loaded)
    params_cpu = checkpoint["params"]
    state_cpu = checkpoint["state"]
    opt_cpu = checkpoint["opt"]
    chrono.load(checkpoint["chrono"])
  elif config.get("model_init"):
    write_note(f"Initialize model from {config.model_init}...")
    params_cpu, state_cpu = model_mod.load(
        {"params": params_cpu, "state": state_cpu},
        config.model_init, config.model,
        **config.get("model_load", {}))
    if jax.process_index() == 0:
      parameter_overview.log_parameter_overview(
          params_cpu, msg="restored params")

  write_note("Kicking off misc stuff...")
  first_step = bv_optax.get_count(opt_cpu)
  chrono.inform(first_step, total_steps, batch_size, ntrain_img / batch_size)
  prof = None  # Keeps track of start/stop of profiler state.

  write_note(f"Replicating...\n{chrono.note}")
  params_repl = flax.jax_utils.replicate(params_cpu)
  opt_repl = flax.jax_utils.replicate(opt_cpu)
  state_repl = flax.jax_utils.replicate(state_cpu)

  rng, rng_loop = jax.random.split(rng, 2)
  rngs_loop = flax.jax_utils.replicate(rng_loop)
  ckpt_writer = None

  write_note(f"First step compilations...\n{chrono.note}")
  error = None  # For exiting with an error after cleanup. Avoids indentation.

  # Using a python integer for step here, because opt.state.step is allocated
  # on TPU during replication.
  for step, batch in zip(range(first_step + 1, total_steps + 1), train_iter):
    mw.step_start(step)

    with jax.profiler.StepTraceAnnotation("train_step", step_num=step):
      params_repl, opt_repl, state_repl, loss_value, rngs_loop, measurements = (
          update_fn(
              params_repl,
              opt_repl,
              state_repl,
              batch,
              rngs_loop,
              not config.get("freeze_dict", True)))

    # On the first host, let's always profile a handful of early steps.
    if jax.process_index() == 0:
      prof = u.startstop_prof(prof, step, first_step, get_steps("log_training"))

    # Report training progress
    if (u.itstime(step, get_steps("log_training"), total_steps, host=0)
        or chrono.warmup and jax.process_index() == 0):
      for i, sched_fn_cpu in enumerate(sched_fns_cpu):
        mw.measure(f"global_schedule{i if i else ''}", sched_fn_cpu(step - 1))
      l = mw.measure("training_loss", loss_value[0])
      for name, value in measurements.items():
        mw.measure(name, value[0])
      chrono.tick(step, mw.measure, write_note)
      if not np.isfinite(l):
        error = (f"The loss became nan or inf somewhere within steps "
                 f"[{step - get_steps('log_training')}, {step}]")
        break

    # Checkpoint saving
    if (save_ckpt_path and
        (u.itstime(step, get_steps("ckpt", None), total_steps, host=0) or
         u.itstime(step, get_steps("keep_ckpt", None), total_steps, host=0))):
      chrono.pause(wait_for=(params_repl, opt_repl, state_repl))
      u.checkpointing_timeout(ckpt_writer, config.get("ckpt_timeout", 1))
      # We need to transfer the weights over now or else we risk keeping them
      # alive while they'll be updated in a future step, creating hard to debug
      # memory errors (see (internal link)). Also, takes device 0's params only.
      params_cpu, opt_cpu, state_cpu = jax.tree_map(
          lambda x: np.array(x[0]), (params_repl, opt_repl, state_repl))

      # Check whether we want to keep a copy of the current checkpoint.
      copy_step = None
      if u.itstime(step, get_steps("keep_ckpt", None), total_steps):
        copy_step = step

      ckpt = {
          "params": params_cpu,
          "state": state_cpu,
          "opt": opt_cpu,
          "chrono": chrono.save(),
      }
      ckpt_writer = pool.apply_async(
          u.save_checkpoint, (ckpt, save_ckpt_path, copy_step))
      chrono.resume()

    for (name, evaluator, log_steps, prefix) in evaluators():
      if u.itstime(step, log_steps, total_steps):
        chrono.pause(wait_for=(params_repl, state_repl))
        write_note(f"{name} evaluation...\n{chrono.note}")
        for key, value in evaluator.run(
            {"params": params_repl, "state": state_repl}):
          mw.measure(f"{prefix}{key}", value)
        chrono.resume()
    mw.step_end()

  # Always give a chance to stop the profiler, no matter how things ended.
  # TODO: can we also do this when dying of an exception like OOM?
  if jax.process_index() == 0 and prof is not None:
    u.startstop_prof(prof)

  # Support eval only runs: run evaluation if total_steps (or num_epochs) is 0.
  if total_steps == 0:
    for (name, evaluator, _, prefix) in evaluators():
      write_note(f"{name} evaluation...\n{chrono.note}")
      for key, value in evaluator.run(
          {"params": params_repl, "state": state_repl}):
        mw.measure(f"{prefix}{key}", value)

  # Last note needs to happen before the pool's closed =)
  if not error:
    write_note(f"Done!\n{chrono.note}")
  else:
    write_note(f"Failed!\n{error}\n{chrono.note}")

  pool.close()
  pool.join()
  mw.close()

  # Make sure all hosts stay up until the end of main.
  u.sync()

  # Before cleanup, as cleanup should only run for successful jobs.
  if error is not None:
    raise RuntimeError(error)

  u.maybe_cleanup_workdir(workdir, flags.FLAGS.cleanup, info)


if __name__ == "__main__":
  app.run(main)