From 18f2d741003764c0da0d30aaa7a629e62a67e2dd Mon Sep 17 00:00:00 2001
From: andy-neuma <andy@neuralmagic.com>
Date: Tue, 27 Feb 2024 16:56:50 -0500
Subject: [PATCH 1/3] cache

---
 .github/actions/nm-hf-cache/action.yml | 11 +++++++++++
 .github/actions/nm-set-env/action.yml  |  1 +
 .github/workflows/build-test.yml       |  4 ++++
 3 files changed, 16 insertions(+)
 create mode 100644 .github/actions/nm-hf-cache/action.yml

diff --git a/.github/actions/nm-hf-cache/action.yml b/.github/actions/nm-hf-cache/action.yml
new file mode 100644
index 0000000000000..379aa775a8fb2
--- /dev/null
+++ b/.github/actions/nm-hf-cache/action.yml
@@ -0,0 +1,11 @@
+name: HF cache
+description: 'mount HF cache'
+runs:
+  using: composite
+  steps:
+  - run: |
+      sudo apt install -y nfs-common
+      sudo mkdir -m 777 -p /EFS
+      sudo mount -t nfs4 -o nfsvers=4.1,rsize=1048576,wsize=1048576,hard,timeo=600,retrans=2,noresvport fs-0204dcb31e253065f.efs.us-east-1.amazonaws.com:/ /EFS
+      sudo chown -R $(whoami):$(whoami) ${HF_HOME}
+    shell: bash
diff --git a/.github/actions/nm-set-env/action.yml b/.github/actions/nm-set-env/action.yml
index c3e1d34d0c95b..18de4e31e716a 100644
--- a/.github/actions/nm-set-env/action.yml
+++ b/.github/actions/nm-set-env/action.yml
@@ -12,6 +12,7 @@ runs:
   steps:
   - run: |
       echo "HF_TOKEN=${HF_TOKEN_SECRET}" >> $GITHUB_ENV
+      echo "HF_HOME=/EFS/hf_home" >> $GITHUB_ENV
       NUM_THREADS=$(./.github/scripts/determine-threading -G ${{ inputs.Gi_per_thread }})
       echo "MAX_JOBS=${NUM_THREADS}" >> $GITHUB_ENV
       echo "VLLM_INSTALL_PUNICA_KERNELS=1" >> $GITHUB_ENV
diff --git a/.github/workflows/build-test.yml b/.github/workflows/build-test.yml
index ceffb577aa2fa..a542f4c538236 100644
--- a/.github/workflows/build-test.yml
+++ b/.github/workflows/build-test.yml
@@ -79,6 +79,10 @@ jobs:
                 python: ${{ inputs.python }}
                 venv: TEST
 
+            - name: hf cache
+              id: hf_cache
+              uses: ./.github/actions/nm-hf-cache/
+
             # TODO: testmo source is currently hardcoded.
             - name: create testmo run
               id: create_testmo_run

From 6200bd6c88260345f49e66875eb32c1e951c4ab7 Mon Sep 17 00:00:00 2001
From: andy-neuma <andy@neuralmagic.com>
Date: Tue, 27 Feb 2024 16:59:04 -0500
Subject: [PATCH 2/3] 2 gpus

---
 .github/scripts/run-tests | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/.github/scripts/run-tests b/.github/scripts/run-tests
index 6fb3ea1bec350..082b6da0bf2fb 100755
--- a/.github/scripts/run-tests
+++ b/.github/scripts/run-tests
@@ -59,9 +59,9 @@ do
     # this is a bit messy and brittle, but certain tests
     # need to be run with specific options
     if [[ "${TEST}" == *"kernels"* ]]; then
-        CUDA_VISIBLE_DEVICES=0 pytest --junitxml=${RESULT_XML} ${TEST} || LOCAL_SUCCESS=$?
+        CUDA_VISIBLE_DEVICES=0,1 pytest --junitxml=${RESULT_XML} ${TEST} || LOCAL_SUCCESS=$?
     elif [[ "${TEST}" == *"samplers"* ]]; then
-        CUDA_VISIBLE_DEVICES=0 pytest --junitxml=${RESULT_XML} ${TEST} || LOCAL_SUCCESS=$?
+        CUDA_VISIBLE_DEVICES=0,1 pytest --junitxml=${RESULT_XML} ${TEST} || LOCAL_SUCCESS=$?
     elif [[ "${TEST}" == *"distributed"* ]]; then
         pytest --forked --junitxml=${RESULT_XML} ${TEST} || LOCAL_SUCCESS=$?
     else

From 8950eb0c5c4ed1af7d8e13979eaf7fad9133da79 Mon Sep 17 00:00:00 2001
From: andy-neuma <andy@neuralmagic.com>
Date: Tue, 27 Feb 2024 17:06:07 -0500
Subject: [PATCH 3/3] obsfucate EFS id

---
 .github/actions/nm-hf-cache/action.yml | 6 +++++-
 .github/workflows/build-test.yml       | 2 ++
 2 files changed, 7 insertions(+), 1 deletion(-)

diff --git a/.github/actions/nm-hf-cache/action.yml b/.github/actions/nm-hf-cache/action.yml
index 379aa775a8fb2..af36c34ea341b 100644
--- a/.github/actions/nm-hf-cache/action.yml
+++ b/.github/actions/nm-hf-cache/action.yml
@@ -1,11 +1,15 @@
 name: HF cache
 description: 'mount HF cache'
+inputs:
+  fs_cache:
+    description: 'filesystem to use for HF cache'
+    required: true
 runs:
   using: composite
   steps:
   - run: |
       sudo apt install -y nfs-common
       sudo mkdir -m 777 -p /EFS
-      sudo mount -t nfs4 -o nfsvers=4.1,rsize=1048576,wsize=1048576,hard,timeo=600,retrans=2,noresvport fs-0204dcb31e253065f.efs.us-east-1.amazonaws.com:/ /EFS
+      sudo mount -t nfs4 -o nfsvers=4.1,rsize=1048576,wsize=1048576,hard,timeo=600,retrans=2,noresvport ${{ inputs.fs_cache }}:/ /EFS
       sudo chown -R $(whoami):$(whoami) ${HF_HOME}
     shell: bash
diff --git a/.github/workflows/build-test.yml b/.github/workflows/build-test.yml
index a542f4c538236..00bfe701ea15a 100644
--- a/.github/workflows/build-test.yml
+++ b/.github/workflows/build-test.yml
@@ -82,6 +82,8 @@ jobs:
             - name: hf cache
               id: hf_cache
               uses: ./.github/actions/nm-hf-cache/
+              with:
+                fs_cache: ${{ secrets.HF_FS_CACHE }}
 
             # TODO: testmo source is currently hardcoded.
             - name: create testmo run