updated.

Author: tanyuqian

examples/classification_regression/README.md

@@ -7,7 +7,7 @@ This example implements classification tasks language models, supporting sequenc
 
 ```shell
 # Install RedCoast
-pip install redco==0.4.20
+pip install redco==0.4.21
 # Install torchvision/torch (cpu version)
 pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
 ```

examples/classification_regression/main.py

@@ -75,7 +75,7 @@ def main(dataset_name='sst2',
          per_device_batch_size=4,
          n_epochs=2,
          learning_rate=2e-5,
-         warmup_rate=0.1,
+         warmup_ratio=0.1,
          lr_schedule_type='linear',
          grad_norm_clip=1.,
          weight_decay=0.,
@@ -117,7 +117,7 @@ def main(dataset_name='sst2',
         n_epochs=n_epochs,
         learning_rate=learning_rate,
         schedule_type=lr_schedule_type,
-        warmup_rate=warmup_rate)
+        warmup_ratio=warmup_ratio)
     optimizer = optax.MultiSteps(optax.chain(
         optax.clip_by_global_norm(grad_norm_clip),
         optax.adamw(learning_rate=lr_schedule_fn, weight_decay=weight_decay)

examples/image_to_text/README.md

@@ -6,7 +6,7 @@ This example implements image-to-text with Redco.
 
 Install Redco
 ```shell
-pip install redco==0.4.8
+pip install redco==0.4.21
 ```
 
 ### Usage

examples/image_to_text/main.py

@@ -92,7 +92,7 @@ def main(data_dir='./mscoco_data',
          per_device_batch_size=8,
          learning_rate=2e-5,
          lr_schedule_type='linear',
-         warmup_rate=0.1,
+         warmup_ratio=0.1,
          grad_norm_clip=1.,
          weight_decay=0.01,
          workdir='./workdir',
@@ -147,7 +147,7 @@ def main(data_dir='./mscoco_data',
         n_epochs=n_epochs,
         learning_rate=learning_rate,
         schedule_type=lr_schedule_type,
-        warmup_rate=warmup_rate)
+        warmup_ratio=warmup_ratio)
     optimizer = optax.MultiSteps(optax.chain(
         optax.clip_by_global_norm(grad_norm_clip),
         optax.adamw(learning_rate=lr_schedule_fn, weight_decay=weight_decay)

examples/language_modeling/README.md

@@ -7,7 +7,7 @@ This example implements training causal language models, supporting causal langu
 
 ```shell
 # Install RedCoast
-pip install redco==0.4.20
+pip install redco==0.4.21
 # Install torchvision/torch (cpu version)
 pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cpu
 ```

examples/language_modeling/main.py

@@ -42,7 +42,7 @@ def main(dataset_name='alexgshaw/llama-13b-tokenized-wikitext-2-v1',
          per_device_batch_size=4,
          learning_rate=2e-5,
          lr_schedule_type='linear',
-         warmup_rate=0.1,
+         warmup_ratio=0.1,
          grad_norm_clip=1.,
          weight_decay=0.01,
          workdir='./workdir',
@@ -81,7 +81,7 @@ def main(dataset_name='alexgshaw/llama-13b-tokenized-wikitext-2-v1',
         n_epochs=n_epochs,
         learning_rate=learning_rate,
         schedule_type=lr_schedule_type,
-        warmup_rate=warmup_rate)
+        warmup_ratio=warmup_ratio)
     optimizer = optax.MultiSteps(optax.chain(
         optax.clip_by_global_norm(grad_norm_clip),
         optax.adamw(learning_rate=lr_schedule_fn, weight_decay=weight_decay)

examples/text_to_text/README.md

@@ -11,7 +11,7 @@ It supports
 
 ```shell
 # Install RedCoast
-pip install redco==0.4.20
+pip install redco==0.4.21
 # Install nltk, rouge for evaluation
 pip install nltk rouge-score
 # Install torchvision/torch (cpu version) for loading PyTorch initializations

examples/text_to_text/main.py

@@ -87,7 +87,7 @@ def main(dataset_name='EdinburghNLP/xsum',
          per_device_batch_size=32,
          learning_rate=2e-5,
          lr_schedule_type='linear',
-         warmup_rate=0.1,
+         warmup_ratio=0.1,
          grad_norm_clip=1.,
          weight_decay=0.01,
          workdir='./workdir',
@@ -133,7 +133,7 @@ def main(dataset_name='EdinburghNLP/xsum',
         n_epochs=n_epochs,
         learning_rate=learning_rate,
         schedule_type=lr_schedule_type,
-        warmup_rate=warmup_rate)
+        warmup_ratio=warmup_ratio)
     optimizer = optax.MultiSteps(optax.chain(
         optax.clip_by_global_norm(grad_norm_clip),
         optax.adamw(learning_rate=lr_schedule_fn, weight_decay=weight_decay)

redco/__init__.py

@@ -12,7 +12,7 @@
 #  See the License for the specific language governing permissions and
 #  limitations under the License.
 
-__version__ = '0.4.19'
+__version__ = '0.4.21'
 
 from .deployers import *
 from .trainers import *

setup.py

@@ -17,10 +17,10 @@
 
 setup(
     name="redco",
-    version="0.4.20",
+    version="0.4.21",
     author="Bowen Tan",
     packages=find_packages(),
-    install_requires=['jax', 'flax', 'optax', 'numpy'],
+    install_requires=['jax', 'flax', 'optax'],
     include_package_data=True,
     python_requires=">=3.8",
     long_description=open('README.md').read(),