From 6218ac382095beafa0dee121f277a48d31286147 Mon Sep 17 00:00:00 2001
From: derekk-nm <derek@neuralmagic.com>
Date: Wed, 3 Jul 2024 14:47:16 +0000
Subject: [PATCH 1/6] upload RELEASE wheel to pypi.org

passing the workflow category and wheel file name to nm-upload-assets-to-gcp.yml workflow so that they can be used by the publish_whl action.
---
 .github/workflows/nm-build-test.yml           |  2 ++
 .github/workflows/nm-upload-assets-to-gcp.yml | 20 ++++++++++++++++++-
 2 files changed, 21 insertions(+), 1 deletion(-)

diff --git a/.github/workflows/nm-build-test.yml b/.github/workflows/nm-build-test.yml
index 9a5043308a067..2e666ed70af6e 100644
--- a/.github/workflows/nm-build-test.yml
+++ b/.github/workflows/nm-build-test.yml
@@ -169,6 +169,8 @@ jobs:
             label: gcp-k8s-util
             timeout: ${{ inputs.build_timeout }}
             gitref: ${{ github.ref }}
+            wf_category: ${{ inputs.wf_category }}
+            whl: ${{ needs.BUILD.outputs.whl }}
         secrets: inherit
 
     # update docker
diff --git a/.github/workflows/nm-upload-assets-to-gcp.yml b/.github/workflows/nm-upload-assets-to-gcp.yml
index 8d8835271de0a..0dad796b13840 100644
--- a/.github/workflows/nm-upload-assets-to-gcp.yml
+++ b/.github/workflows/nm-upload-assets-to-gcp.yml
@@ -16,7 +16,14 @@ on:
                 description: 'git commit hash or branch name'
                 type: string
                 required: true
-
+            wf_category:
+                description: "workflow category: REMOTE, NIGHTLY, RELEASE"
+                type: string
+                default: "REMOTE"
+            whl:
+                description: "wheel file path"
+                type: string
+                required: true
 jobs:
 
     PUBLISH:
@@ -53,6 +60,16 @@ jobs:
               with:
                   path: assets
 
+            - name: push wheel to pypi.org
+              # this workflow is only run if push-to-pypi is True, and we only
+              # want to push RELEASE wheels to the external pypi.org
+              if: ${{ inputs.wf_category }} == "RELEASE"
+              uses: neuralmagic/nm-actions/actions/publish_whl/action.yml@main
+              with:
+                username: ${{ secrets.PYPI_PUBLIC_USER }}
+                password: ${{ secrets.PYPI_PUBLIC_AUTH }}
+                whl: ${{ inputs.whl }}
+
             # GCP
             - name: 'Authenticate to Google Cloud'
               id: auth
@@ -70,3 +87,4 @@ jobs:
             - name: cp assets
               id: cp-assets
               uses: ./.github/actions/nm-cp-assets/
+

From ef34d580d014d1f9b34b23075cf72923bd7fd6e4 Mon Sep 17 00:00:00 2001
From: derekk-nm <derek@neuralmagic.com>
Date: Tue, 9 Jul 2024 15:59:32 +0000
Subject: [PATCH 2/6] PR revisions

- refactor code from nm-cp-assets that gets wheel and tar.gz files into a new action that can be used for that and our nm-upload-assets-to-gcp.yml workflow
- correct syntax in the if statement
- use nm-actions @v1.0.0
---
 .github/actions/nm-cp-assets/action.yml       | 19 ++++++------
 .../actions/nm_whl_tar_gz_names/action.yml    | 25 +++++++++++++++
 .github/workflows/nm-build-test.yml           |  1 -
 .github/workflows/nm-upload-assets-to-gcp.yml | 31 +++++++++++++------
 4 files changed, 56 insertions(+), 20 deletions(-)
 create mode 100644 .github/actions/nm_whl_tar_gz_names/action.yml

diff --git a/.github/actions/nm-cp-assets/action.yml b/.github/actions/nm-cp-assets/action.yml
index b4957d30d3b25..cf67aaa49e3f6 100644
--- a/.github/actions/nm-cp-assets/action.yml
+++ b/.github/actions/nm-cp-assets/action.yml
@@ -1,18 +1,17 @@
 name: cp assets
 description: "cp whl and tarfile to Google storage 'neuralmagic-public-pypi/dist'"
+inputs:
+  whl:
+    description: "the wheel asset file path"
+    required: true
+  targz:
+    description: "the tar.gz asset file path"
+    required: true
 runs:
   using: composite
   steps:
     - id: cp_assets
       run: |
-        WHL=$(find assets -type f -name "*nm_vllm*.whl")
-        WHL_FILE=$(basename ${WHL})
-        echo "whl: ${WHL}"
-        echo "whl_file: ${WHL_FILE}"
-        TAR=$(find assets -path "*nm-vllm*.tar.gz" -type f -name "nm-vllm*.tar.gz")
-        TAR_FILE=$(basename ${TAR})
-        echo "tar: ${TAR}"
-        echo "tar_file: ${TAR_FILE}"
-        gcloud storage cp ${WHL} gs://neuralmagic-public-pypi/dist/${WHL_FILE}
-        gcloud storage cp ${TAR} gs://neuralmagic-public-pypi/dist/${TAR_FILE}
+        gcloud storage cp ${WHL} gs://neuralmagic-public-pypi/dist/${{ inputs.whl }}
+        gcloud storage cp ${TAR} gs://neuralmagic-public-pypi/dist/${{ inputs.targz }}
       shell: bash
diff --git a/.github/actions/nm_whl_tar_gz_names/action.yml b/.github/actions/nm_whl_tar_gz_names/action.yml
new file mode 100644
index 0000000000000..aa23c2bba00a8
--- /dev/null
+++ b/.github/actions/nm_whl_tar_gz_names/action.yml
@@ -0,0 +1,25 @@
+name: get wheel and tar.gz names
+description: "retrieve the whl and tarfile names from existing assets"
+outputs:
+  whl:
+    description: "the wheel asset file path"
+    value: ${{ steps.whl_targz_names.outputs.whl }}
+  targz:
+    description: "the tar.gz asset file path"
+    value: ${{ steps.whl_targz_names.outputs.targz }}
+runs:
+  using: composite
+  steps:
+    - id: whl_targz_names
+      run: |
+        WHL=$(find assets -type f -name "*nm_vllm*.whl")
+        WHL_FILE=$(basename ${WHL})
+        echo "whl: ${WHL}"
+        echo "whl_file: ${WHL_FILE}"
+        TAR=$(find assets -path "*nm-vllm*.tar.gz" -type f -name "nm-vllm*.tar.gz")
+        TAR_FILE=$(basename ${TAR})
+        echo "tar: ${TAR}"
+        echo "tar_file: ${TAR_FILE}"
+        echo "whl=${WHL_FILE}" >> $GITHUB_OUTPUT
+        echo "targz=${TAR_FILE}" >> $GITHUB_OUTPUT
+      shell: bash
diff --git a/.github/workflows/nm-build-test.yml b/.github/workflows/nm-build-test.yml
index d57acd6a4f8ec..fc64d6dc35d14 100644
--- a/.github/workflows/nm-build-test.yml
+++ b/.github/workflows/nm-build-test.yml
@@ -170,7 +170,6 @@ jobs:
             timeout: ${{ inputs.build_timeout }}
             gitref: ${{ github.ref }}
             wf_category: ${{ inputs.wf_category }}
-            whl: ${{ needs.BUILD.outputs.whl }}
         secrets: inherit
 
     # update docker
diff --git a/.github/workflows/nm-upload-assets-to-gcp.yml b/.github/workflows/nm-upload-assets-to-gcp.yml
index eba427927679d..0e6cb3c14e245 100644
--- a/.github/workflows/nm-upload-assets-to-gcp.yml
+++ b/.github/workflows/nm-upload-assets-to-gcp.yml
@@ -20,10 +20,6 @@ on:
                 description: "workflow category: REMOTE, NIGHTLY, RELEASE"
                 type: string
                 default: "REMOTE"
-            whl:
-                description: "wheel file path"
-                type: string
-                required: true
 
     workflow_dispatch:
         inputs:
@@ -84,15 +80,29 @@ jobs:
               with:
                   path: assets
 
+            - name: get wheel and tar.gz names
+              id: whl_targz_names
+              uses: ./.github/actions/nm_whl_tar_gz_names
+
+            # this workflow is only run if push-to-pypi is True, and we only
+            # want to push RELEASE assets to the external pypi.org
+            # publish the wheel file
             - name: push wheel to pypi.org
-              # this workflow is only run if push-to-pypi is True, and we only
-              # want to push RELEASE wheels to the external pypi.org
-              if: ${{ inputs.wf_category }} == "RELEASE"
-              uses: neuralmagic/nm-actions/actions/publish_whl/action.yml@main
+              if: ${{ inputs.wf_category == "RELEASE" }}
+              uses: neuralmagic/nm-actions/actions/publish_whl/action.yml@v1.0.0
               with:
                 username: ${{ secrets.PYPI_PUBLIC_USER }}
                 password: ${{ secrets.PYPI_PUBLIC_AUTH }}
-                whl: ${{ inputs.whl }}
+                whl: ${{ steps.whl_targz_names.outputs.whl }}
+
+            # publish the tar.gz file
+            - name: push tar.gz to pypi.org
+              if: ${{ inputs.wf_category == "RELEASE" }}
+              uses: neuralmagic/nm-actions/actions/publish_whl/action.yml@v1.0.0
+              with:
+                username: ${{ secrets.PYPI_PUBLIC_USER }}
+                password: ${{ secrets.PYPI_PUBLIC_AUTH }}
+                whl: ${{ steps.whl_targz_names.outputs.targz }}
 
             # GCP
             - name: 'Authenticate to Google Cloud'
@@ -111,6 +121,9 @@ jobs:
             - name: cp assets
               id: cp-assets
               uses: ./.github/actions/nm-cp-assets/
+              with:
+                whl: ${{ steps.whl_targz_names.outputs.whl }}
+                targz: ${{ steps.whl_targz_names.outputs.targz }}
 
             - name: trigger stratus nm-pypi update workflow to update nm-pypi index
               uses: actions/github-script@v6

From 9e5ffa9f3512a7f883b306fe6d9714c024dd7f19 Mon Sep 17 00:00:00 2001
From: derekk-nm <derek@neuralmagic.com>
Date: Tue, 9 Jul 2024 16:08:57 +0000
Subject: [PATCH 3/6] fix if

---
 .github/workflows/nm-upload-assets-to-gcp.yml | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/nm-upload-assets-to-gcp.yml b/.github/workflows/nm-upload-assets-to-gcp.yml
index 0e6cb3c14e245..fb73af2153566 100644
--- a/.github/workflows/nm-upload-assets-to-gcp.yml
+++ b/.github/workflows/nm-upload-assets-to-gcp.yml
@@ -88,7 +88,7 @@ jobs:
             # want to push RELEASE assets to the external pypi.org
             # publish the wheel file
             - name: push wheel to pypi.org
-              if: ${{ inputs.wf_category == "RELEASE" }}
+              if: ${{ inputs.wf_category == 'RELEASE' }}
               uses: neuralmagic/nm-actions/actions/publish_whl/action.yml@v1.0.0
               with:
                 username: ${{ secrets.PYPI_PUBLIC_USER }}
@@ -97,7 +97,7 @@ jobs:
 
             # publish the tar.gz file
             - name: push tar.gz to pypi.org
-              if: ${{ inputs.wf_category == "RELEASE" }}
+              if: ${{ inputs.wf_category == 'RELEASE' }}
               uses: neuralmagic/nm-actions/actions/publish_whl/action.yml@v1.0.0
               with:
                 username: ${{ secrets.PYPI_PUBLIC_USER }}

From 6370dd0b9609572c404135647050a04467f42579 Mon Sep 17 00:00:00 2001
From: Andy Linfoot <78757007+andy-neuma@users.noreply.github.com>
Date: Tue, 9 Jul 2024 16:12:08 -0400
Subject: [PATCH 4/6] expand benchmark to h100's (#371)

SUMMARY:
* updated "build test" to accept an array of benchmarking labels
* updated "remote push" and "nightly" workflows to include benchmarking
on h100's
* adjusted docker job to have same criteria as upload job. did this
since upload could fail, but for auth reasons and this shouldn't stop us
from push docker.

TEST PLAN:
runs on remote push

---------

Co-authored-by: andy-neuma <andy@neuralmagic.com>
---
 .github/workflows/nm-build-test.yml  | 11 +++++++----
 .github/workflows/nm-nightly.yml     |  2 +-
 .github/workflows/nm-remote-push.yml |  2 +-
 3 files changed, 9 insertions(+), 6 deletions(-)

diff --git a/.github/workflows/nm-build-test.yml b/.github/workflows/nm-build-test.yml
index fc64d6dc35d14..2e2b404c81fef 100644
--- a/.github/workflows/nm-build-test.yml
+++ b/.github/workflows/nm-build-test.yml
@@ -58,10 +58,10 @@ on:
         required: true
 
       # benchmark related parameters
-      benchmark_label:
-        description: "requested benchmark label (specifies instance)"
+      benchmark_labels:
+        description: "stringified Json array of benchmark labels"
         type: string
-        default: ""
+        required: true
       benchmark_config_list_file:
         description: "benchmark configs file, e.g. 'nm_benchmark_nightly_configs_list.txt'"
         type: string
@@ -136,9 +136,12 @@ jobs:
     BENCHMARK:
         needs: [BUILD]
         if: success()
+        strategy:
+            matrix:
+                benchmark_label: ${{ fromJson(inputs.benchmark_labels) }}
         uses: ./.github/workflows/nm-benchmark.yml
         with:
-            label: ${{ inputs.benchmark_label }}
+            label: ${{ matrix.benchmark_label }}
             benchmark_config_list_file: ${{ inputs.benchmark_config_list_file }}
             timeout: ${{ inputs.benchmark_timeout }}
             gitref: ${{ github.ref }}
diff --git a/.github/workflows/nm-nightly.yml b/.github/workflows/nm-nightly.yml
index 434f2b9032b1b..494198bf496fc 100644
--- a/.github/workflows/nm-nightly.yml
+++ b/.github/workflows/nm-nightly.yml
@@ -39,7 +39,7 @@ jobs:
                             {"python":"3.11.4","label":"gcp-k8s-l4-solo","test":"neuralmagic/tests/test_skip_env_vars/full.txt"}]'
             test_timeout: 480
 
-            benchmark_label: gcp-k8s-l4-solo
+            benchmark_labels: '["gcp-k8s-l4-solo", "k8s-h100-solo"]'
             benchmark_config_list_file: ./.github/data/nm_benchmark_base_config_list.txt
             benchmark_timeout: 480
             push_benchmark_results_to_gh_pages: "${{ github.event_name == 'schedule' || inputs.push_benchmark_results_to_gh_pages }}"
diff --git a/.github/workflows/nm-remote-push.yml b/.github/workflows/nm-remote-push.yml
index a44274d9e8a11..17108f345f891 100644
--- a/.github/workflows/nm-remote-push.yml
+++ b/.github/workflows/nm-remote-push.yml
@@ -25,7 +25,7 @@ jobs:
                             {"python":"3.11.4","label":"gcp-k8s-l4-solo","test":"neuralmagic/tests/test_skip_env_vars/smoke.txt"}]'
             test_timeout: 480
 
-            benchmark_label: gcp-k8s-l4-solo
+            benchmark_labels: '["gcp-k8s-l4-solo", "k8s-h100-solo"]'
             benchmark_config_list_file: ./.github/data/nm_benchmark_base_config_list.txt
             benchmark_timeout: 480
 

From b4d90208e26025390f3ff509e0d918bef88800fe Mon Sep 17 00:00:00 2001
From: derekk-nm <derek@neuralmagic.com>
Date: Tue, 9 Jul 2024 15:59:32 +0000
Subject: [PATCH 5/6] PR revisions

- refactor code from nm-cp-assets that gets wheel and tar.gz files into a new action that can be used for that and our nm-upload-assets-to-gcp.yml workflow
- correct syntax in the if statement
- use nm-actions @v1.0.0
---
 .github/workflows/nm-upload-assets-to-gcp.yml | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/.github/workflows/nm-upload-assets-to-gcp.yml b/.github/workflows/nm-upload-assets-to-gcp.yml
index fb73af2153566..0e6cb3c14e245 100644
--- a/.github/workflows/nm-upload-assets-to-gcp.yml
+++ b/.github/workflows/nm-upload-assets-to-gcp.yml
@@ -88,7 +88,7 @@ jobs:
             # want to push RELEASE assets to the external pypi.org
             # publish the wheel file
             - name: push wheel to pypi.org
-              if: ${{ inputs.wf_category == 'RELEASE' }}
+              if: ${{ inputs.wf_category == "RELEASE" }}
               uses: neuralmagic/nm-actions/actions/publish_whl/action.yml@v1.0.0
               with:
                 username: ${{ secrets.PYPI_PUBLIC_USER }}
@@ -97,7 +97,7 @@ jobs:
 
             # publish the tar.gz file
             - name: push tar.gz to pypi.org
-              if: ${{ inputs.wf_category == 'RELEASE' }}
+              if: ${{ inputs.wf_category == "RELEASE" }}
               uses: neuralmagic/nm-actions/actions/publish_whl/action.yml@v1.0.0
               with:
                 username: ${{ secrets.PYPI_PUBLIC_USER }}

From f13ee3438ea212df2869f4cbcafd747fdb127b4f Mon Sep 17 00:00:00 2001
From: derekk-nm <derek@neuralmagic.com>
Date: Wed, 10 Jul 2024 11:55:45 +0000
Subject: [PATCH 6/6] PR input

remove an extra "whl" arg.
rename workflow
---
 .github/workflows/nm-build-test.yml                           | 2 +-
 .../{nm-upload-assets-to-gcp.yml => nm-upload-assets.yml}     | 4 ----
 2 files changed, 1 insertion(+), 5 deletions(-)
 rename .github/workflows/{nm-upload-assets-to-gcp.yml => nm-upload-assets.yml} (97%)

diff --git a/.github/workflows/nm-build-test.yml b/.github/workflows/nm-build-test.yml
index 2e2b404c81fef..1b812368c650c 100644
--- a/.github/workflows/nm-build-test.yml
+++ b/.github/workflows/nm-build-test.yml
@@ -167,7 +167,7 @@ jobs:
     UPLOAD:
         needs: [TEST, BENCHMARK, LM-EVAL]
         if: ${{ inputs.push_to_pypi }}
-        uses: ./.github/workflows/nm-upload-assets-to-gcp.yml
+        uses: ./.github/workflows/nm-upload-assets.yml
         with:
             label: gcp-k8s-util
             timeout: ${{ inputs.build_timeout }}
diff --git a/.github/workflows/nm-upload-assets-to-gcp.yml b/.github/workflows/nm-upload-assets.yml
similarity index 97%
rename from .github/workflows/nm-upload-assets-to-gcp.yml
rename to .github/workflows/nm-upload-assets.yml
index 0e6cb3c14e245..2b75316a38575 100644
--- a/.github/workflows/nm-upload-assets-to-gcp.yml
+++ b/.github/workflows/nm-upload-assets.yml
@@ -39,10 +39,6 @@ on:
                 description: "workflow category: REMOTE, NIGHTLY, RELEASE"
                 type: string
                 default: "REMOTE"
-            whl:
-                description: "wheel file path"
-                type: string
-                required: true
 
 jobs: