Skip jax distributed initialize if specified to skip (#172)

Author: coolkp

.github/workflows/UploadDockerImages.yml

@@ -44,7 +44,7 @@ jobs:
       run: docker system prune --all --force
     - name: build maxdiffusion jax stable stack gpu image
       run: |
-        bash .github/workflows/build_and_upload_images.sh CLOUD_IMAGE_NAME=maxdiffusion_jax_stable_stack_gpu MODE=stable_stack PROJECT=tpu-prod-env-multipod LOCAL_IMAGE_NAME=maxdiffusion_jax_stable_stack_gpu BASEIMAGE=us-docker.pkg.dev/cloud-tpu-images/jax-stable-stack/tpu:latest DEVICE=gpu
+        bash .github/workflows/build_and_upload_images.sh CLOUD_IMAGE_NAME=maxdiffusion_jax_stable_gpu MODE=stable PROJECT=tpu-prod-env-multipod LOCAL_IMAGE_NAME=maxdiffusion_jax_stable_gpu DEVICE=gpu
     - name: build maxdiffusion jax nightly image
       run: |
          bash .github/workflows/build_and_upload_images.sh CLOUD_IMAGE_NAME=maxdiffusion_jax_nightly_gpu MODE=nightly PROJECT=tpu-prod-env-multipod LOCAL_IMAGE_NAME=maxdiffusion_jax_nightly DEVICE=gpu

src/maxdiffusion/configs/base14.yml

@@ -92,6 +92,7 @@ diffusion_scheduler_config: {
 
 # Hardware
 hardware: 'tpu' # Supported hardware types are 'tpu', 'gpu'
+skip_jax_distributed_system: False
 
 base_output_directory: ""
 

src/maxdiffusion/configs/base21.yml

@@ -91,6 +91,7 @@ diffusion_scheduler_config: {
 
 # Hardware
 hardware: 'tpu' # Supported hardware types are 'tpu', 'gpu'
+skip_jax_distributed_system: False
 
 # Output directory
 # Create a GCS bucket, e.g. my-maxtext-outputs and set this to "gs://my-maxtext-outputs/"

src/maxdiffusion/configs/base_2_base.yml

@@ -104,6 +104,7 @@ diffusion_scheduler_config: {
 
 # Hardware
 hardware: 'tpu' # Supported hardware types are 'tpu', 'gpu'
+skip_jax_distributed_system: False
 
 # Output directory
 # Create a GCS bucket, e.g. my-maxtext-outputs and set this to "gs://my-maxtext-outputs/"

src/maxdiffusion/configs/base_flux_dev.yml

@@ -120,6 +120,7 @@ base_output_directory: ""
 
 # Hardware
 hardware: 'tpu' # Supported hardware types are 'tpu', 'gpu'
+skip_jax_distributed_system: False
 
 # Parallelism
 mesh_axes: ['data', 'fsdp', 'tensor']

src/maxdiffusion/configs/base_flux_dev_multi_res.yml

@@ -120,6 +120,7 @@ base_output_directory: ""
 
 # Hardware
 hardware: 'tpu' # Supported hardware types are 'tpu', 'gpu'
+skip_jax_distributed_system: False
 
 # Parallelism
 mesh_axes: ['data', 'fsdp', 'tensor']

src/maxdiffusion/configs/base_flux_schnell.yml

@@ -128,6 +128,7 @@ base_output_directory: ""
 
 # Hardware
 hardware: 'tpu' # Supported hardware types are 'tpu', 'gpu'
+skip_jax_distributed_system: False
 
 # Parallelism
 mesh_axes: ['data', 'fsdp', 'tensor']

src/maxdiffusion/configs/base_wan_t2v.yml

@@ -114,6 +114,7 @@ base_output_directory: ""
 
 # Hardware
 hardware: 'tpu' # Supported hardware types are 'tpu', 'gpu'
+skip_jax_distributed_system: False
 
 # Parallelism
 mesh_axes: ['data', 'fsdp', 'tensor']

src/maxdiffusion/configs/base_xl.yml

@@ -95,7 +95,7 @@ base_output_directory: ""
 
 # Hardware
 hardware: 'tpu' # Supported hardware types are 'tpu', 'gpu'
-
+skip_jax_distributed_system: False
 # Parallelism
 mesh_axes: ['data', 'fsdp', 'tensor']
 

src/maxdiffusion/configs/base_xl_lightning.yml

@@ -71,7 +71,7 @@ diffusion_scheduler_config: {
 
 # Hardware
 hardware: 'tpu' # Supported hardware types are 'tpu', 'gpu'
-
+skip_jax_distributed_system: False
 # Output directory
 # Create a GCS bucket, e.g. my-maxtext-outputs and set this to "gs://my-maxtext-outputs/"
 base_output_directory: ""