add nvcc_threads to gha

neuralmagic · Mar 26, 2024 · 3963555 · 3963555
1 parent af5c67a
commit 3963555
Show file tree

Hide file tree

Showing 9 changed files with 50 additions and 4 deletions.
diff --git a/.github/actions/nm-build-vllm/action.yml b/.github/actions/nm-build-vllm/action.yml
@@ -1,9 +1,6 @@
 name: build nm-vllm
 description: 'build nm-vllm'
 inputs:
-  Gi_per_thread:
-    description: 'requested GiB to reserve per thread'
-    required: true
   python:
     description: 'python version, e.g. 3.10.12'
     required: true

diff --git a/.github/actions/nm-set-env/action.yml b/.github/actions/nm-set-env/action.yml
@@ -7,6 +7,10 @@ inputs:
   Gi_per_thread:
     description: 'requested GiB to reserve per thread'
     required: true
+  nvcc_threads:
+    description: "number of threads nvcc build threads"
+    type: string
+    required: true
 runs:
   using: composite
   steps:
@@ -16,6 +20,7 @@ runs:
       echo "HF_HOME=/EFS/hf_home" >> $GITHUB_ENV
       NUM_THREADS=$(./.github/scripts/determine-threading -G ${{ inputs.Gi_per_thread }})
       echo "MAX_JOBS=${NUM_THREADS}" >> $GITHUB_ENV
+      echo "NVCC_THREADS=${{ inputs.nvcc_threads }}" >> $GITHUB_ENV
       echo "VLLM_INSTALL_PUNICA_KERNELS=1" >> $GITHUB_ENV
       echo "NCCL_IGNORE_DISABLED_P2P=1" >> $GITHUB_ENV
       echo "PYENV_ROOT=/usr/local/apps/pyenv" >> $GITHUB_ENV

diff --git a/.github/workflows/build-test.yml b/.github/workflows/build-test.yml
@@ -19,6 +19,10 @@ on:
         description: 'requested GiB to reserve per thread'
         type: string
         required: true
+      nvcc_threads:
+        description: "number of threads nvcc build threads"
+        type: string
+        required: true
       python:
         description: "python version, e.g. 3.10.12"
         type: string
@@ -47,6 +51,10 @@ on:
         description: 'requested GiB to reserve per thread'
         type: string
         required: true
+      nvcc_threads:
+        description: "number of threads nvcc build threads"
+        type: string
+        required: true
       python:
         description: "python version, e.g. 3.10.12"
         type: string
@@ -79,6 +87,7 @@ jobs:
               with:
                 hf_token: ${{ secrets.NM_HF_TOKEN }}
                 Gi_per_thread: ${{ inputs.Gi_per_thread }}
+                nvcc_threads: ${{ inputs.nvcc_threads }}
 
             - name: set python
               id: set_python

diff --git a/.github/workflows/build-whl.yml b/.github/workflows/build-whl.yml
@@ -19,6 +19,10 @@ on:
         description: 'requested GiB to reserve per thread'
         type: string
         required: true
+      nvcc_threads:
+        description: "number of threads nvcc build threads"
+        type: string
+        required: true
       python:
         description: "python version, e.g. 3.10.12"
         type: string
@@ -43,6 +47,10 @@ on:
         description: 'requested GiB to reserve per thread'
         type: string
         required: true
+      nvcc_threads:
+        description: "number of threads nvcc build threads"
+        type: string
+        required: true
       python:
         description: "python version, e.g. 3.10.12"
         type: string
@@ -76,6 +84,7 @@ jobs:
               with:
                 hf_token: ${{ secrets.NM_HF_TOKEN }}
                 Gi_per_thread: ${{ inputs.Gi_per_thread }}
+                nvcc_threads: ${{ inputs.nvcc_threads }}
 
             - name: set python
               id: set_python

diff --git a/.github/workflows/gen-whl.yml b/.github/workflows/gen-whl.yml
@@ -20,5 +20,6 @@ jobs:
         timeout: 30
         gitref: ${{ inputs.gitref }}
         Gi_per_thread: 4
+        nvcc_threads: 8
         python: ${{ matrix.python }}
     secrets: inherit
diff --git a/.github/workflows/nightly.yml b/.github/workflows/nightly.yml
@@ -24,6 +24,7 @@ jobs:
             timeout: 240
             gitref: ${{ github.ref }}
             Gi_per_thread: 4
+            nvcc_threads: 8
             python: 3.10.12
             test_skip_list:
         secrets: inherit
@@ -35,6 +36,7 @@ jobs:
             timeout: 300
             gitref: ${{ github.ref }}
             Gi_per_thread: 12
+            nvcc_threads: 1
             python: 3.11.4
             test_skip_list:
         secrets: inherit
@@ -48,6 +50,7 @@ jobs:
     #         timeout: 480
     #         gitref: '${{ github.ref }}'
     #         Gi_per_thread: 4
+    #         nvcc_threads: 8
     #         python: "3.10.12"
     #         # Always push if it is a scheduled job
     #         push_benchmark_results_to_gh_pages: "${{ github.event_name == 'schedule' || inputs.push_benchmark_results_to_gh_pages }}"
@@ -62,6 +65,7 @@ jobs:
             timeout: 720
             gitref: '${{ github.ref }}'
             Gi_per_thread: 12
+            nvcc_threads: 1
             python: "3.10.12"
             # Always push if it is a scheduled job
             push_benchmark_results_to_gh_pages: "${{ github.event_name == 'schedule' || inputs.push_benchmark_results_to_gh_pages }}"
@@ -75,5 +79,6 @@ jobs:
             timeout: 60
             gitref: '${{ github.ref }}'
             Gi_per_thread: 12
+            nvcc_threads: 1
             python: "3.10.12"
         secrets: inherit
diff --git a/.github/workflows/nm-benchmark.yml b/.github/workflows/nm-benchmark.yml
@@ -23,6 +23,10 @@ on:
         description: 'requested GiB to reserve per thread'
         type: string
         required: true
+      nvcc_threads:
+        description: "number of threads nvcc build threads"
+        type: string
+        required: true
       python:
         description: "python version, e.g. 3.10.12"
         type: string
@@ -55,6 +59,10 @@ on:
         description: 'requested GiB to reserve per thread'
         type: string
         required: true
+      nvcc_threads:
+        description: "number of threads nvcc build threads"
+        type: string
+        required: true
       python:
         description: "python version, e.g. 3.10.12"
         type: string
@@ -89,6 +97,7 @@ jobs:
         with:
           hf_token: ${{ secrets.NM_HF_TOKEN }}
           Gi_per_thread: ${{ inputs.Gi_per_thread }}
+          nvcc_threads: ${{ inputs.nvcc_threads }}
 
       - name: set python
         id: set_python

diff --git a/.github/workflows/nm-lm-eval-accuracy.yml b/.github/workflows/nm-lm-eval-accuracy.yml
@@ -19,6 +19,10 @@ on:
         description: 'requested GiB to reserve per thread'
         type: string
         required: true
+      nvcc_threads:
+        description: "number of threads nvcc build threads"
+        type: string
+        required: true
       python:
         description: "python version, e.g. 3.10.12"
         type: string
@@ -43,6 +47,10 @@ on:
         description: 'requested GiB to reserve per thread'
         type: string
         required: true
+      nvcc_threads:
+        description: "number of threads nvcc build threads"
+        type: string
+        required: true
       python:
         description: "python version, e.g. 3.10.12"
         type: string
@@ -68,6 +76,7 @@ jobs:
         with:
           hf_token: ${{ secrets.NM_HF_TOKEN }}
           Gi_per_thread: ${{ inputs.Gi_per_thread }}
+          nvcc_threads: ${{ inputs.nvcc_threads }}
 
       - name: set python
         id: set_python

diff --git a/.github/workflows/remote-push.yml b/.github/workflows/remote-push.yml
@@ -24,11 +24,11 @@ jobs:
     #        timeout: 240
     #        gitref: '${{ github.ref }}'
     #        Gi_per_thread: 4
+    #        nvcc_threads: 8
     #        python: ${{ matrix.python }}
     #        test_skip_list: neuralmagic/tests/skip-for-remote-push.txt
     #    secrets: inherit
 
-    # multi-gpu
     AWS-AVX2-32G-A10G-24G:
         strategy:
             matrix:
@@ -39,6 +39,7 @@ jobs:
             timeout: 240
             gitref: '${{ github.ref }}'
             Gi_per_thread: 12
+            nvcc_threads: 1
             python: ${{ matrix.python }}
             test_skip_list: neuralmagic/tests/skip-for-remote-push.txt
         secrets: inherit
@@ -52,6 +53,7 @@ jobs:
     #        timeout: 60
     #        gitref: '${{ github.ref }}'
     #        Gi_per_thread: 12
+    #        nvcc_threads: 1
     #        python: "3.10.12"
     #        push_benchmark_results_to_gh_pages: "false"
     #    secrets: inherit