"git@developer.sourcefind.cn:norm/vllm.git" did not exist on "017d9f15151ce571a5f4fd381699c72a872636ec"
Commit ee0b6a6d authored by Hongkun Yu's avatar Hongkun Yu Committed by A. Unique TensorFlower
Browse files

Clean up configs.

PiperOrigin-RevId: 306883491
parent 1ff3c319
...@@ -3,8 +3,6 @@ ...@@ -3,8 +3,6 @@
# Reaches ~76.1% within 350 epochs. # Reaches ~76.1% within 350 epochs.
# Note: This configuration uses a scaled per-replica batch size based on the number of devices. # Note: This configuration uses a scaled per-replica batch size based on the number of devices.
runtime: runtime:
model_dir: null
mode: 'train_and_eval'
distribution_strategy: 'mirrored' distribution_strategy: 'mirrored'
num_gpus: 1 num_gpus: 1
train_dataset: train_dataset:
......
...@@ -3,8 +3,6 @@ ...@@ -3,8 +3,6 @@
# Reaches ~76.1% within 350 epochs. # Reaches ~76.1% within 350 epochs.
# Note: This configuration uses a scaled per-replica batch size based on the number of devices. # Note: This configuration uses a scaled per-replica batch size based on the number of devices.
runtime: runtime:
model_dir: null
mode: 'train_and_eval'
distribution_strategy: 'tpu' distribution_strategy: 'tpu'
train_dataset: train_dataset:
name: 'imagenet2012' name: 'imagenet2012'
......
# Note: This configuration uses a scaled per-replica batch size based on the number of devices. # Note: This configuration uses a scaled per-replica batch size based on the number of devices.
runtime: runtime:
model_dir: null
mode: 'train_and_eval'
distribution_strategy: 'mirrored' distribution_strategy: 'mirrored'
num_gpus: 1 num_gpus: 1
train_dataset: train_dataset:
......
...@@ -2,8 +2,6 @@ ...@@ -2,8 +2,6 @@
# Takes ~3 minutes, 15 seconds per epoch for v3-32. # Takes ~3 minutes, 15 seconds per epoch for v3-32.
# Note: This configuration uses a scaled per-replica batch size based on the number of devices. # Note: This configuration uses a scaled per-replica batch size based on the number of devices.
runtime: runtime:
model_dir: null
mode: 'train_and_eval'
distribution_strategy: 'tpu' distribution_strategy: 'tpu'
train_dataset: train_dataset:
name: 'imagenet2012' name: 'imagenet2012'
......
...@@ -2,8 +2,6 @@ ...@@ -2,8 +2,6 @@
# Reaches > 76.1% within 90 epochs. # Reaches > 76.1% within 90 epochs.
# Note: This configuration uses a scaled per-replica batch size based on the number of devices. # Note: This configuration uses a scaled per-replica batch size based on the number of devices.
runtime: runtime:
model_dir: null
mode: 'train_and_eval'
distribution_strategy: 'mirrored' distribution_strategy: 'mirrored'
num_gpus: 1 num_gpus: 1
train_dataset: train_dataset:
...@@ -33,7 +31,7 @@ validation_dataset: ...@@ -33,7 +31,7 @@ validation_dataset:
mean_subtract: True mean_subtract: True
standardize: True standardize: True
model: model:
model_name: 'resnet' name: 'resnet'
model_params: model_params:
rescale_inputs: False rescale_inputs: False
optimizer: optimizer:
......
...@@ -3,8 +3,6 @@ ...@@ -3,8 +3,6 @@
# Reaches > 76.1% within 90 epochs. # Reaches > 76.1% within 90 epochs.
# Note: This configuration uses a scaled per-replica batch size based on the number of devices. # Note: This configuration uses a scaled per-replica batch size based on the number of devices.
runtime: runtime:
model_dir: null
mode: 'train_and_eval'
distribution_strategy: 'tpu' distribution_strategy: 'tpu'
train_dataset: train_dataset:
name: 'imagenet2012' name: 'imagenet2012'
...@@ -35,7 +33,7 @@ validation_dataset: ...@@ -35,7 +33,7 @@ validation_dataset:
standardize: True standardize: True
dtype: 'bfloat16' dtype: 'bfloat16'
model: model:
model_name: 'resnet' name: 'resnet'
model_params: model_params:
rescale_inputs: False rescale_inputs: False
optimizer: optimizer:
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment