fix: vllm e2e dependencies and env_vars

defenseunicorns · Jul 26, 2024 · 6092128 · 6092128
1 parent 8acaff6
commit 6092128
Show file tree

Hide file tree

Showing 2 changed files with 4 additions and 0 deletions.
diff --git a/.github/workflows/e2e-vllm.yaml b/.github/workflows/e2e-vllm.yaml
@@ -69,6 +69,7 @@ jobs:
           run: |
             python -m pip install "."
             python -m pip install -U "huggingface_hub[cli,hf_transfer]"
+            python -m pip install confz
 
         - name: Setup UDS Environment
           uses: defenseunicorns/uds-common/.github/actions/setup@05f42bb3117b66ebef8c72ae050b34bce19385f5

diff --git a/packages/vllm/zarf.yaml b/packages/vllm/zarf.yaml
@@ -25,6 +25,7 @@ components:
           - "vllm-values.yaml"
     images:
       - ghcr.io/defenseunicorns/leapfrogai/vllm:###ZARF_PKG_TMPL_IMAGE_VERSION###
+      - cgr.dev/chainguard/bash:latest
     dataInjections:
       - source: .model/
         target:
@@ -41,3 +42,5 @@ components:
             env:
               - LAI_REPO_ID=TheBloke/Synthia-7B-v2.0-GPTQ
               - LAI_REVISION=gptq-4bit-32g-actorder_True
+              - LAI_QUANTIZATION=gptq
+              - LAI_HF_HUB_ENABLE_HF_TRANSFER=1