Merge pull request #60 from AI-Hypercomputer/raymondzou-collectives

Update jax stable stack image to 0.5.2 in collectives recipe

Author: raymondzouu

microbenchmarks/trillium/collectives/README.md

@@ -16,7 +16,7 @@ python3 ~/xpk/xpk.py workload create \
     --device-type=v6e-256 \
     --command="git clone https://github.com/AI-Hypercomputer/accelerator-microbenchmarks.git && cd accelerator-microbenchmarks && git checkout trillium-collectives && pip install -r requirements.txt && echo '4096 41943040 314572800' > /proc/sys/net/ipv4/tcp_rmem && export LIBTPU_INIT_ARGS='--megascale_grpc_premap_memory_bytes=17179869184 --xla_tpu_enable_sunk_dcn_allreduce_done_with_host_reduction=true' && python src/run_benchmark.py --config=configs/1x_v6e_256.yaml" \
     --num-slices=1 \
-    --docker-image=us-docker.pkg.dev/cloud-tpu-images/jax-stable-stack/tpu:jax0.4.37-rev1 \
+    --docker-image=us-docker.pkg.dev/cloud-tpu-images/jax-stable-stack/tpu:jax0.5.2-rev1 \
     --workload=${WORKLOAD_NAME}
 ```
 

microbenchmarks/trillium/collectives/collectives-1xv6e-256.sh

@@ -5,5 +5,5 @@ python3 ~/xpk/xpk.py workload create \
     --device-type=v6e-256 \
     --command="git clone https://github.com/AI-Hypercomputer/accelerator-microbenchmarks.git && cd accelerator-microbenchmarks && git checkout trillium-collectives && pip install -r requirements.txt && echo '4096 41943040 314572800' > /proc/sys/net/ipv4/tcp_rmem && export LIBTPU_INIT_ARGS='--megascale_grpc_premap_memory_bytes=17179869184 --xla_tpu_enable_sunk_dcn_allreduce_done_with_host_reduction=true' && python src/run_benchmark.py --config=configs/1x_v6e_256.yaml" \
     --num-slices=1 \
-    --docker-image=us-docker.pkg.dev/cloud-tpu-images/jax-stable-stack/tpu:jax0.4.37-rev1 \
+    --docker-image=us-docker.pkg.dev/cloud-tpu-images/jax-stable-stack/tpu:jax0.5.2-rev1 \
     --workload=${WORKLOAD_NAME}

microbenchmarks/trillium/collectives/collectives-2xv6e-256.sh

@@ -5,5 +5,5 @@ python3 ~/xpk/xpk.py workload create \
     --device-type=v6e-256 \
     --command="git clone https://github.com/AI-Hypercomputer/accelerator-microbenchmarks.git && cd accelerator-microbenchmarks && git checkout trillium-collectives && pip install -r requirements.txt && echo '4096 41943040 314572800' > /proc/sys/net/ipv4/tcp_rmem && export LIBTPU_INIT_ARGS='--megascale_grpc_premap_memory_bytes=17179869184 --xla_tpu_enable_sunk_dcn_allreduce_done_with_host_reduction=true' && python src/run_benchmark.py --config=configs/2x_v6e_256.yaml" \
     --num-slices=2 \
-    --docker-image=us-docker.pkg.dev/cloud-tpu-images/jax-stable-stack/tpu:jax0.4.37-rev1 \
+    --docker-image=us-docker.pkg.dev/cloud-tpu-images/jax-stable-stack/tpu:jax0.5.2-rev1 \
     --workload=${WORKLOAD_NAME}

microbenchmarks/trillium/collectives/collectives-4xv6e-256.sh

@@ -5,5 +5,5 @@ python3 ~/dev/xpk/xpk.py workload create \
     --device-type=v6e-256 \
     --command="git clone https://github.com/AI-Hypercomputer/accelerator-microbenchmarks.git && cd accelerator-microbenchmarks && git checkout trillium-collectives && pip install -r requirements.txt && echo '4096 41943040 314572800' > /proc/sys/net/ipv4/tcp_rmem && export LIBTPU_INIT_ARGS='--megascale_grpc_premap_memory_bytes=17179869184 --xla_tpu_enable_sunk_dcn_allreduce_done_with_host_reduction=true' && python src/run_benchmark.py --config=configs/4x_v6e_256.yaml" \
     --num-slices=4 \
-    --docker-image=us-docker.pkg.dev/cloud-tpu-images/jax-stable-stack/tpu:jax0.4.37-rev1 \
+    --docker-image=us-docker.pkg.dev/cloud-tpu-images/jax-stable-stack/tpu:jax0.5.2-rev1 \
     --workload=${WORKLOAD_NAME}