From 6ab04343af63ace6f9c7161e16e5f6987a77cf62 Mon Sep 17 00:00:00 2001 From: Justin Law Date: Mon, 16 Sep 2024 14:26:20 -0400 Subject: [PATCH] Zarf-native create time model download --- packages/vllm/zarf-config.yaml | 10 ++++++++++ packages/vllm/zarf.yaml | 13 +++++++------ 2 files changed, 17 insertions(+), 6 deletions(-) create mode 100644 packages/vllm/zarf-config.yaml diff --git a/packages/vllm/zarf-config.yaml b/packages/vllm/zarf-config.yaml new file mode 100644 index 000000000..3a3941848 --- /dev/null +++ b/packages/vllm/zarf-config.yaml @@ -0,0 +1,10 @@ +package: + create: + set: + # x-release-please-start-version + image_version: "0.12.2" + # x-release-please-end + + model_repo_id: "TheBloke/Synthia-7B-v2.0-GPTQ" + model_revision: "gptq-4bit-32g-actorder_True" + model_path: "/data/.model/" diff --git a/packages/vllm/zarf.yaml b/packages/vllm/zarf.yaml index b23cf84f0..997af4705 100644 --- a/packages/vllm/zarf.yaml +++ b/packages/vllm/zarf.yaml @@ -11,12 +11,13 @@ constants: value: "###ZARF_PKG_TMPL_IMAGE_VERSION###" - name: MODEL_REPO_ID description: "The HuggingFace repository ID" - value: "TheBloke/Synthia-7B-v2.0-GPTQ" + value: "###ZARF_PKG_TMPL_MODEL_REPO_ID###" - name: MODEL_REVISION - value: "gptq-4bit-32g-actorder_True" + description: "The HuggingFace git branch or commit hash" + value: "###ZARF_PKG_TMPL_MODEL_REVISION###" - name: MODEL_PATH - description: "Location of the Zarf Injected model files" - value: "/data/.model/" + description: "Defines the location of the Zarf Injected model files in the vLLM container" + value: "###ZARF_PKG_TMPL_MODEL_PATH###" variables: # vLLM runtime configuration (usually influenced by .env in local development) @@ -132,5 +133,5 @@ components: # NOTE: This assumes python is installed and in $PATH and 'huggingface_hub[cli,hf_transfer]' has been installed - cmd: "python src/model_download.py" env: - - LFAI_REPO_ID="###ZARF_CONST_MODEL_REPO_ID###" - - LFAI_REVISION="###ZARF_CONST_MODEL_REVISION###" + - LFAI_REPO_ID=###ZARF_PKG_TMPL_MODEL_REPO_ID### + - LFAI_REVISION=###ZARF_PKG_TMPL_MODEL_REVISION###