[Re-submit] Compute true loss Flax examples (#19504)

* Compute true loss

* fixup

* final

* final

* final

* Update examples/flax/language-modeling/run_bart_dlm_flax.py

Co-authored-by: Sanchit Gandhi <93869735+sanchit-gandhi@users.noreply.github.com>

* jax.tree_map => jax.tree_util.tree_map

* Compute true loss

* final

* fixup

* final

* final

* Update examples/flax/language-modeling/run_bart_dlm_flax.py

Co-authored-by: Sanchit Gandhi <93869735+sanchit-gandhi@users.noreply.github.com>

* jax.tree_map => jax.tree_util.tree_map

Co-authored-by: Sanchit Gandhi <93869735+sanchit-gandhi@users.noreply.github.com>

examples/flax/image-captioning/run_image_captioning_flax.py

@@ -335,7 +335,6 @@ def data_loader(rng: jax.random.PRNGKey, dataset: Dataset, batch_size: int, shuf
         batch_idx = np.arange(len(dataset))
 
     for idx in range(steps):
-
         start_idx = batch_size * idx
         end_idx = batch_size * (idx + 1)
 
@@ -347,7 +346,6 @@ def data_loader(rng: jax.random.PRNGKey, dataset: Dataset, batch_size: int, shuf
 
 
 def write_metric(summary_writer, metrics, train_time, step, metric_key_prefix="train"):
-
     if train_time:
         summary_writer.scalar("train_time", train_time, step)
 
@@ -782,11 +780,9 @@ def main():
         num_splits = steps // steps_per_block + int(steps % steps_per_block > 0)
 
         for idx in range(num_splits):
-
             if not block_size:
                 _ds = ds
             else:
-
                 start_idx = block_size * idx
                 end_idx = block_size * (idx + 1)
 
@@ -926,8 +922,9 @@ def main():
 
         # ignore padded tokens from loss
         loss = loss * padding_mask
-        loss = loss.sum() / padding_mask.sum()
+        loss = loss.sum()
-        return loss
+        num_labels = padding_mask.sum()
+        return loss, num_labels
 
     # Define gradient update step fn
     def train_step(state, batch, label_smoothing_factor=0.0):
@@ -936,29 +933,38 @@ def main():
         def compute_loss(params):
             labels = batch.pop("labels")
             logits = state.apply_fn(**batch, params=params, dropout_rng=dropout_rng, train=True)[0]
-            loss = loss_fn(logits, labels, batch["decoder_attention_mask"], label_smoothing_factor)
+            loss, num_labels = loss_fn(logits, labels, batch["decoder_attention_mask"], label_smoothing_factor)
-            return loss
+            return loss, num_labels
 
-        grad_fn = jax.value_and_grad(compute_loss)
+        grad_fn = jax.value_and_grad(compute_loss, has_aux=True)
-        loss, grad = grad_fn(state.params)
+        (loss, num_labels), grad = grad_fn(state.params)
-        grad = jax.lax.pmean(grad, "batch")
+        num_labels = jax.lax.psum(num_labels, "batch")
 
+        # true loss = total loss / total samples
+        loss = jax.lax.psum(loss, "batch")
+        loss = jax.tree_util.tree_map(lambda x: x / num_labels, loss)
+
+        # true grad = total grad / total samples
+        grad = jax.lax.psum(grad, "batch")
+        grad = jax.tree_util.tree_map(lambda x: x / num_labels, grad)
         new_state = state.apply_gradients(grads=grad, dropout_rng=new_dropout_rng)
 
         metrics = {"loss": loss, "learning_rate": linear_decay_lr_schedule_fn(state.step)}
-        metrics = jax.lax.pmean(metrics, axis_name="batch")
-
         return new_state, metrics
 
     # Define eval fn
     def eval_step(params, batch, label_smoothing_factor=0.0):
         labels = batch.pop("labels")
         logits = model(**batch, params=params, train=False)[0]
-        loss = loss_fn(logits, labels, batch["decoder_attention_mask"], label_smoothing_factor)
 
-        # summarize metrics
+        loss, num_labels = loss_fn(logits, labels, batch["decoder_attention_mask"], label_smoothing_factor)
+        num_labels = jax.lax.psum(num_labels, "batch")
+
+        # true loss = total loss / total samples
+        loss = jax.lax.psum(loss, "batch")
+        loss = jax.tree_util.tree_map(lambda x: x / num_labels, loss)
+
         metrics = {"loss": loss}
-        metrics = jax.lax.pmean(metrics, axis_name="batch")
         return metrics
 
     # Define generation function
@@ -1024,7 +1030,6 @@ def main():
         ckpt_dir: str = "",
         is_prediction=False,
     ):
-
         logger.info(f"*** {'Predict' if is_prediction else 'Evaluate'} ***")
 
         metrics = []
@@ -1103,12 +1108,10 @@ def main():
             logger.info(desc)
 
         if jax.process_index() == 0:
-
             if not os.path.isdir(os.path.join(training_args.output_dir, ckpt_dir)):
                 os.makedirs(os.path.join(training_args.output_dir, ckpt_dir), exist_ok=True)
 
             if metrics:
-
                 # Save metrics (only for the evaluation/prediction being done along with training)
                 if has_tensorboard and training_args.do_train:
                     write_metric(
@@ -1143,7 +1146,6 @@ def main():
     input_rng = None
 
     if training_args.do_train:
-
         cur_step = 0
         train_time = 0
         epochs = tqdm(range(num_epochs), desc=f"Epoch ... (1/{num_epochs})", position=0)
@@ -1166,7 +1168,6 @@ def main():
 
             # train
             for batch_idx, _ in enumerate(tqdm(range(steps_per_epoch), desc="Training...", position=1, leave=False)):
-
                 cur_step += 1
                 batch = next(train_batches)
                 batch_start = time.time()
@@ -1177,7 +1178,6 @@ def main():
 
                 # log and save info
                 if training_args.logging_steps > 0 and cur_step % training_args.logging_steps == 0:
-
                     _train_metric = unreplicate(train_metric)
                     desc = (
                         f"Epoch... ({epoch + 1}/{num_epochs} | Step: {cur_step} | Loss: {_train_metric['loss']} |"
@@ -1217,7 +1217,6 @@ def main():
 
             # log and save info
             if training_args.logging_steps <= 0:
-
                 logger.info(desc)
 
                 with open(os.path.join(training_args.output_dir, "log"), "a", encoding="UTF-8") as fp:

examples/flax/language-modeling/README.md

@@ -351,7 +351,7 @@ The example script uses the 🤗 Datasets library. You can easily customize them
 To setup all relevant files for training, let's create a directory.
 
 ```bash
-mkdir ./norwegian-roberta-base
+mkdir ./norwegian-bart-base
 ```
 
 ### Train tokenizer

examples/flax/language-modeling/run_bart_dlm_flax.py

@@ -799,19 +799,25 @@ def main():
             loss = optax.softmax_cross_entropy(logits, onehot(labels, logits.shape[-1])) * label_mask
 
             # take average
-            loss = loss.sum() / label_mask.sum()
+            loss = loss.sum()
+            num_labels = label_mask.sum()
 
-            return loss
+            return loss, num_labels
 
-        grad_fn = jax.value_and_grad(loss_fn)
+        grad_fn = jax.value_and_grad(loss_fn, has_aux=True)
-        loss, grad = grad_fn(state.params)
+        (loss, num_labels), grad = grad_fn(state.params)
-        grad = jax.lax.pmean(grad, "batch")
+        num_labels = jax.lax.psum(num_labels, "batch")
+
+        # true loss = total loss / total samples
+        loss = jax.lax.psum(loss, "batch")
+        loss = jax.tree_util.tree_map(lambda x: x / num_labels, loss)
+
+        # true grad = total grad / total samples
+        grad = jax.lax.psum(grad, "batch")
+        grad = jax.tree_util.tree_map(lambda x: x / num_labels, grad)
         new_state = state.apply_gradients(grads=grad)
 
-        metrics = jax.lax.pmean(
+        metrics = {"loss": loss, "learning_rate": linear_decay_lr_schedule_fn(state.step)}
-            {"loss": loss, "learning_rate": linear_decay_lr_schedule_fn(state.step)}, axis_name="batch"
-        )
-
         return new_state, metrics, new_dropout_rng
 
     # Create parallel version of the train step
@@ -888,7 +894,7 @@ def main():
                 num_eval_samples = len(tokenized_datasets["validation"])
                 # Avoid using jax.numpy here in case of TPU training
                 eval_samples_idx = np.arange(num_eval_samples)
-                eval_batch_idx = generate_batch_splits(eval_samples_idx, eval_batch_size, drop_last=False)
+                eval_batch_idx = generate_batch_splits(eval_samples_idx, eval_batch_size)
 
                 eval_metrics = []
                 for i, batch_idx in enumerate(tqdm(eval_batch_idx, desc="Evaluating ...", position=2)):
@@ -903,9 +909,9 @@ def main():
 
                 # normalize eval metrics
                 eval_metrics = get_metrics(eval_metrics)
-                eval_metrics = jax.tree_map(jnp.sum, eval_metrics)
+                eval_metrics = jax.tree_util.tree_map(jnp.sum, eval_metrics)
                 eval_normalizer = eval_metrics.pop("normalizer")
-                eval_metrics = jax.tree_map(lambda x: x / eval_normalizer, eval_metrics)
+                eval_metrics = jax.tree_util.tree_map(lambda x: x / eval_normalizer, eval_metrics)
 
                 # Update progress bar
                 epochs.desc = f"Step... ({cur_step} | Loss: {eval_metrics['loss']}, Acc: {eval_metrics['accuracy']})"
@@ -917,7 +923,7 @@ def main():
             if cur_step % training_args.save_steps == 0 and cur_step > 0:
                 # save checkpoint after each epoch and push checkpoint to the hub
                 if jax.process_index() == 0:
-                    params = jax.device_get(jax.tree_map(lambda x: x[0], state.params))
+                    params = jax.device_get(jax.tree_util.tree_map(lambda x: x[0], state.params))
                     model.save_pretrained(training_args.output_dir, params=params)
                     tokenizer.save_pretrained(training_args.output_dir)
                     if training_args.push_to_hub:
@@ -928,7 +934,7 @@ def main():
         num_eval_samples = len(tokenized_datasets["validation"])
         # Avoid using jax.numpy here in case of TPU training
         eval_samples_idx = np.arange(num_eval_samples)
-        eval_batch_idx = generate_batch_splits(eval_samples_idx, eval_batch_size, drop_last=False)
+        eval_batch_idx = generate_batch_splits(eval_samples_idx, eval_batch_size)
 
         eval_metrics = []
         for _, batch_idx in enumerate(tqdm(eval_batch_idx, desc="Evaluating ...", position=2)):
@@ -943,9 +949,9 @@ def main():
 
         # normalize eval metrics
         eval_metrics = get_metrics(eval_metrics)
-        eval_metrics = jax.tree_map(lambda metric: jnp.sum(metric).item(), eval_metrics)
+        eval_metrics = jax.tree_util.tree_map(lambda metric: jnp.sum(metric).item(), eval_metrics)
         eval_normalizer = eval_metrics.pop("normalizer")
-        eval_metrics = jax.tree_map(lambda x: x / eval_normalizer, eval_metrics)
+        eval_metrics = jax.tree_util.tree_map(lambda x: x / eval_normalizer, eval_metrics)
 
         try:
             perplexity = math.exp(eval_metrics["loss"])

examples/flax/language-modeling/run_mlm_flax.py

@@ -723,18 +723,25 @@ def main():
             loss = optax.softmax_cross_entropy(logits, onehot(labels, logits.shape[-1])) * label_mask
 
             # take average
-            loss = loss.sum() / label_mask.sum()
+            loss = loss.sum()
+            num_labels = label_mask.sum()
 
-            return loss
+            return loss, num_labels
 
-        grad_fn = jax.value_and_grad(loss_fn)
+        grad_fn = jax.value_and_grad(loss_fn, has_aux=True)
-        loss, grad = grad_fn(state.params)
+        (loss, num_labels), grad = grad_fn(state.params)
-        grad = jax.lax.pmean(grad, "batch")
+        num_labels = jax.lax.psum(num_labels, "batch")
+
+        # true loss = total loss / total samples
+        loss = jax.lax.psum(loss, "batch")
+        loss = jax.tree_util.tree_map(lambda x: x / num_labels, loss)
+
+        # true grad = total grad / total samples
+        grad = jax.lax.psum(grad, "batch")
+        grad = jax.tree_util.tree_map(lambda x: x / num_labels, grad)
         new_state = state.apply_gradients(grads=grad)
 
-        metrics = jax.lax.pmean(
+        metrics = {"loss": loss, "learning_rate": linear_decay_lr_schedule_fn(state.step)}
-            {"loss": loss, "learning_rate": linear_decay_lr_schedule_fn(state.step)}, axis_name="batch"
-        )
 
         return new_state, metrics, new_dropout_rng
 

examples/flax/language-modeling/run_t5_mlm_flax.py

@@ -328,7 +328,6 @@ class FlaxDataCollatorForT5MLM:
     decoder_start_token_id: int
 
     def __call__(self, examples: List[Dict[str, np.ndarray]]) -> BatchEncoding:
-
         # convert list to dict and tensorize input
         batch = BatchEncoding(
             {k: np.array([examples[i][k] for i in range(len(examples))]) for k, v in examples[0].items()}
@@ -397,7 +396,6 @@ class FlaxDataCollatorForT5MLM:
         return input_ids
 
     def random_spans_noise_mask(self, length):
-
         """This function is copy of `random_spans_helper <https://github.com/google-research/text-to-text-transfer-transformer/blob/84f8bcc14b5f2c03de51bd3587609ba8f6bbd1cd/t5/data/preprocessors.py#L2682>`__ .
 
         Noise mask consisting of random spans of noise tokens.

examples/flax/summarization/run_summarization_flax.py

@@ -784,8 +784,9 @@ def main():
 
         # ignore padded tokens from loss
         loss = loss * padding_mask
-        loss = loss.sum() / padding_mask.sum()
+        loss = loss.sum()
-        return loss
+        num_labels = padding_mask.sum()
+        return loss, num_labels
 
     # Define gradient update step fn
     def train_step(state, batch, label_smoothing_factor=0.0):
@@ -794,29 +795,38 @@ def main():
         def compute_loss(params):
             labels = batch.pop("labels")
             logits = state.apply_fn(**batch, params=params, dropout_rng=dropout_rng, train=True)[0]
-            loss = loss_fn(logits, labels, batch["decoder_attention_mask"], label_smoothing_factor)
+            loss, num_labels = loss_fn(logits, labels, batch["decoder_attention_mask"], label_smoothing_factor)
-            return loss
+            return loss, num_labels
 
-        grad_fn = jax.value_and_grad(compute_loss)
+        grad_fn = jax.value_and_grad(compute_loss, has_aux=True)
-        loss, grad = grad_fn(state.params)
+        (loss, num_labels), grad = grad_fn(state.params)
-        grad = jax.lax.pmean(grad, "batch")
+        num_labels = jax.lax.psum(num_labels, "batch")
 
+        # true loss = total loss / total samples
+        loss = jax.lax.psum(loss, "batch")
+        loss = jax.tree_util.tree_map(lambda x: x / num_labels, loss)
+
+        # true grad = total grad / total samples
+        grad = jax.lax.psum(grad, "batch")
+        grad = jax.tree_util.tree_map(lambda x: x / num_labels, grad)
         new_state = state.apply_gradients(grads=grad, dropout_rng=new_dropout_rng)
 
         metrics = {"loss": loss, "learning_rate": linear_decay_lr_schedule_fn(state.step)}
-        metrics = jax.lax.pmean(metrics, axis_name="batch")
-
         return new_state, metrics
 
     # Define eval fn
     def eval_step(params, batch, label_smoothing_factor=0.0):
         labels = batch.pop("labels")
         logits = model(**batch, params=params, train=False)[0]
-        loss = loss_fn(logits, labels, batch["decoder_attention_mask"], label_smoothing_factor)
 
-        # summarize metrics
+        loss, num_labels = loss_fn(logits, labels, batch["decoder_attention_mask"], label_smoothing_factor)
+        num_labels = jax.lax.psum(num_labels, "batch")
+
+        # true loss = total loss / total samples
+        loss = jax.lax.psum(loss, "batch")
+        loss = jax.tree_util.tree_map(lambda x: x / num_labels, loss)
+
         metrics = {"loss": loss}
-        metrics = jax.lax.pmean(metrics, axis_name="batch")
         return metrics
 
     # Define generation function