Tacotron2+Waveglow/PyT * AMP support * Data preprocessing for Tacotron 2 training * Fixed dropouts on LSTMCells SSD/PyT * script and notebook for inference * AMP support * README update * updates to examples/* BERT/PyT * initial release GNMT/PyT * Default container updated to NGC PyTorch 19.05-py3 * Mixed precision training implemented using APEX AMP * Added inference throughput and latency results on NVIDIA Tesla V100 16G * Added option to run inference on user-provided raw input text from command line NCF/PyT * Updated performance tables. * Default container changed to PyTorch 19.06-py3. * Caching validation negatives between runs Transformer/PyT * new README * jit support added UNet Medical/TF * inference example scripts added * inference benchmark measuring latency added * TRT/TF-TRT support added * README updated GNMT/TF * Performance improvements Small updates (mostly README) for other models.
242 lines
8.8 KiB
Python
242 lines
8.8 KiB
Python
# Copyright (c) 2018. All rights reserved.
|
|
#
|
|
# 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.
|
|
#
|
|
# -----------------------------------------------------------------------
|
|
#
|
|
# Copyright (c) 2018, NVIDIA CORPORATION. All rights reserved.
|
|
#
|
|
# 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.
|
|
|
|
import tensorflow as tf
|
|
import horovod.tensorflow as hvd
|
|
|
|
def float32_variable_storage_getter(getter, name, shape=None, dtype=None,
|
|
initializer=None, regularizer=None,
|
|
trainable=True,
|
|
*args, **kwargs):
|
|
"""
|
|
Custom variable getter that forces trainable variables to be stored in
|
|
float32 precision and then casts them to the half-precision
|
|
"""
|
|
storage_dtype = tf.float32 if trainable else dtype
|
|
variable = getter(name, shape, dtype=storage_dtype,
|
|
initializer=initializer, regularizer=regularizer,
|
|
trainable=trainable,
|
|
*args, **kwargs)
|
|
if trainable and dtype != tf.float32:
|
|
variable = tf.cast(variable, dtype)
|
|
return variable
|
|
|
|
def neural_mf(users,
|
|
items,
|
|
model_dtype,
|
|
nb_users,
|
|
nb_items,
|
|
mf_dim,
|
|
mf_reg,
|
|
mlp_layer_sizes,
|
|
mlp_layer_regs,
|
|
dropout_rate,
|
|
sigmoid=False):
|
|
"""
|
|
Constructs the model graph
|
|
"""
|
|
# Check params
|
|
if len(mlp_layer_sizes) != len(mlp_layer_regs):
|
|
raise RuntimeError('u dummy, layer_sized != layer_regs')
|
|
if mlp_layer_sizes[0] % 2 != 0:
|
|
raise RuntimeError('u dummy, mlp_layer_sizes[0] % 2 != 0')
|
|
nb_mlp_layers = len(mlp_layer_sizes)
|
|
|
|
# Embeddings
|
|
user_embed = tf.get_variable(
|
|
"user_embeddings",
|
|
shape=[nb_users, mf_dim + mlp_layer_sizes[0] // 2],
|
|
initializer=tf.initializers.random_normal(mean=0.0, stddev=0.01))
|
|
item_embed = tf.get_variable(
|
|
"item_embeddings",
|
|
shape=[nb_items, mf_dim + mlp_layer_sizes[0] // 2],
|
|
initializer=tf.initializers.random_normal(mean=0.0, stddev=0.01))
|
|
# Matrix Factorization Embeddings
|
|
xmfu = tf.nn.embedding_lookup(user_embed[:, :mf_dim], users, partition_strategy='div')
|
|
xmfi = tf.nn.embedding_lookup(item_embed[:, :mf_dim], items, partition_strategy='div')
|
|
# MLP Network Embeddings
|
|
xmlpu = tf.nn.embedding_lookup(user_embed[:, mf_dim:], users, partition_strategy='div')
|
|
xmlpi = tf.nn.embedding_lookup(item_embed[:, mf_dim:], items, partition_strategy='div')
|
|
# Enforce model to use fp16 data types when manually enabling mixed precision
|
|
# (Tensorfow ops will use automatically use the data type of the first input)
|
|
if model_dtype == tf.float16:
|
|
xmfu = tf.cast(xmfu, model_dtype)
|
|
xmfi = tf.cast(xmfi, model_dtype)
|
|
xmlpu = tf.cast(xmlpu, model_dtype)
|
|
xmlpi = tf.cast(xmlpi, model_dtype)
|
|
|
|
# Matrix Factorization
|
|
xmf = tf.math.multiply(xmfu, xmfi)
|
|
|
|
# MLP Layers
|
|
xmlp = tf.concat((xmlpu, xmlpi), 1)
|
|
for i in range(1, nb_mlp_layers):
|
|
xmlp = tf.layers.Dense(
|
|
mlp_layer_sizes[i],
|
|
activation=tf.nn.relu,
|
|
kernel_initializer=tf.glorot_uniform_initializer()
|
|
).apply(xmlp)
|
|
xmlp = tf.layers.Dropout(rate=dropout_rate).apply(xmlp)
|
|
|
|
# Final fully-connected layer
|
|
logits = tf.concat((xmf, xmlp), 1)
|
|
logits = tf.layers.Dense(
|
|
1,
|
|
kernel_initializer=tf.keras.initializers.lecun_uniform()
|
|
).apply(logits)
|
|
|
|
if sigmoid:
|
|
logits = tf.math.sigmoid(logits)
|
|
|
|
# Cast model outputs back to float32 if manually enabling mixed precision for loss calculation
|
|
if model_dtype == tf.float16:
|
|
logits = tf.cast(logits, tf.float32)
|
|
|
|
return logits
|
|
|
|
def compute_eval_metrics(logits, dup_mask, val_batch_size, K):
|
|
"""
|
|
Constructs the graph to compute Hit Rate and NDCG
|
|
"""
|
|
# Replace duplicate (uid, iid) pairs with -inf
|
|
logits = logits * (1. - dup_mask)
|
|
logits = logits + (dup_mask * logits.dtype.min)
|
|
# Reshape tensors so that each row corresponds with a user
|
|
logits_by_user = tf.reshape(logits, [-1, val_batch_size])
|
|
dup_mask_by_user = tf.cast(tf.reshape(logits, [-1, val_batch_size]), tf.bool)
|
|
# Get the topk items for each user
|
|
top_item_indices = tf.math.top_k(logits_by_user, K)[1]
|
|
# Check that the positive sample (last index) is in the top K
|
|
is_positive = tf.cast(tf.equal(top_item_indices, val_batch_size-1), tf.int32)
|
|
found_positive = tf.reduce_sum(is_positive, axis=1)
|
|
# Extract the rankings of the positive samples
|
|
positive_ranks = tf.reduce_sum(is_positive * tf.expand_dims(tf.range(K), 0), axis=1)
|
|
dcg = tf.log(2.) / tf.log(tf.cast(positive_ranks, tf.float32) + 2)
|
|
dcg *= tf.cast(found_positive, dcg.dtype)
|
|
|
|
return found_positive, dcg
|
|
|
|
def ncf_model_ops(users,
|
|
items,
|
|
labels,
|
|
dup_mask,
|
|
params,
|
|
mode='TRAIN'):
|
|
"""
|
|
Constructs the training and evaluation graphs
|
|
"""
|
|
# Validation params
|
|
val_batch_size = params['val_batch_size']
|
|
K = params['top_k']
|
|
# Training params
|
|
learning_rate = params['learning_rate']
|
|
beta_1 = params['beta_1']
|
|
beta_2 = params['beta_2']
|
|
epsilon = params['epsilon']
|
|
# Model params
|
|
fp16 = params['fp16']
|
|
nb_users = params['num_users']
|
|
nb_items = params['num_items']
|
|
mf_dim = params['num_factors']
|
|
mf_reg = params['mf_reg']
|
|
mlp_layer_sizes = params['layer_sizes']
|
|
mlp_layer_regs = params['layer_regs']
|
|
dropout = params['dropout']
|
|
sigmoid = False #params['sigmoid']
|
|
loss_scale = params['loss_scale']
|
|
|
|
model_dtype = tf.float16 if fp16 else tf.float32
|
|
|
|
# If manually enabling mixed precision, use the custom variable getter
|
|
custom_getter = None if not fp16 else float32_variable_storage_getter
|
|
# Allow soft device placement
|
|
with tf.device(None), \
|
|
tf.variable_scope('neumf', custom_getter=custom_getter):
|
|
# Model graph
|
|
logits = neural_mf(
|
|
users,
|
|
items,
|
|
model_dtype,
|
|
nb_users,
|
|
nb_items,
|
|
mf_dim,
|
|
mf_reg,
|
|
mlp_layer_sizes,
|
|
mlp_layer_regs,
|
|
dropout,
|
|
sigmoid
|
|
)
|
|
logits = tf.squeeze(logits)
|
|
|
|
if mode == 'INFERENCE':
|
|
return logits
|
|
|
|
# Evaluation Ops
|
|
found_positive, dcg = compute_eval_metrics(logits, dup_mask, val_batch_size, K)
|
|
# Metrics
|
|
hit_rate = tf.metrics.mean(found_positive, name='hit_rate')
|
|
ndcg = tf.metrics.mean(dcg, name='ndcg')
|
|
|
|
eval_op = tf.group(hit_rate[1], ndcg[1])
|
|
|
|
if mode == 'EVAL':
|
|
return hit_rate[0], ndcg[0], eval_op, None
|
|
|
|
# Labels
|
|
labels = tf.reshape(labels, [-1, 1])
|
|
logits = tf.reshape(logits, [-1, 1])
|
|
|
|
# Use adaptive momentum optimizer
|
|
optimizer = tf.train.AdamOptimizer(
|
|
learning_rate=learning_rate,
|
|
beta1=beta_1, beta2=beta_2,
|
|
epsilon=epsilon)
|
|
|
|
loss = tf.losses.sigmoid_cross_entropy(
|
|
labels,
|
|
logits,
|
|
reduction=tf.losses.Reduction.MEAN)
|
|
|
|
# Apply loss scaling if manually enabling mixed precision
|
|
if fp16:
|
|
if loss_scale is None:
|
|
loss_scale_manager = tf.contrib.mixed_precision.ExponentialUpdateLossScaleManager(2**32, 1000)
|
|
else:
|
|
loss_scale_manager = tf.contrib.mixed_precision.FixedLossScaleManager(loss_scale)
|
|
optimizer = tf.contrib.mixed_precision.LossScaleOptimizer(optimizer, loss_scale_manager)
|
|
|
|
# Horovod wrapper for distributed training
|
|
optimizer = hvd.DistributedOptimizer(optimizer)
|
|
|
|
# Update ops
|
|
global_step = tf.train.get_global_step()
|
|
train_op = optimizer.minimize(loss, global_step=global_step)
|
|
|
|
return hit_rate[0], ndcg[0], eval_op, train_op
|