chore: re-tag vllm worker as the default worker image

kubeagi · Mar 25, 2024 · 05d09bb · 05d09bb
1 parent 00dd4b9
commit 05d09bb
Show file tree

Hide file tree

Showing 3 changed files with 13 additions and 16 deletions.
diff --git a/deploy/llms/Dockerfile.fastchat-worker b/deploy/llms/Dockerfile.fastchat-worker
@@ -39,5 +39,15 @@ RUN python3.9 -m pip install einops scipy transformers_stream_generator==0.0.4 d
 # Install requirements for Qutantization with auto-gptq
 RUN python3.9 -m pip install auto-gptq optimum -i ${PYTHON_INDEX_URL}
 
+# Install requirements for vllm worker
+# Ray v2.9.3 and vllm v0.3.3
+RUN python3.9 -m pip install vllm -i ${PYTHON_INDEX_URL}
+
+# Allow to use environment variable to set ray & python version to pass the version check
+# for now, ray: 2.9.0, python: 3.9.x
+# this utils.py file is from ray 2.9.0 ray-ml image
+# search 'KubeAGI' in utils.py for what's changed
+COPY deploy/llms/utils.py /usr/local/lib/python3.9/dist-packages/ray/_private/utils.py
+
 COPY deploy/llms/start-worker.sh /
 ENTRYPOINT ["/start-worker.sh"]
diff --git a/deploy/llms/Dockerfile.fastchat-worker-vllm b/deploy/llms/Dockerfile.fastchat-worker-vllm
diff --git a/pkg/worker/runner.go b/pkg/worker/runner.go
@@ -33,8 +33,9 @@ import (
 
 const (
 	// tag is the same version as fastchat
-	defaultFastChatImage     = "kubeagi/arcadia-fastchat-worker:v0.2.36"
-	defaultFastchatVLLMImage = "kubeagi/arcadia-fastchat-worker:vllm-v0.2.36"
+	defaultFastChatImage = "kubeagi/arcadia-fastchat-worker:v0.2.36"
+	// For ease of maintainance and stability, VLLM module is now included in standard image as a default feature.
+	defaultFastchatVLLMImage = "kubeagi/arcadia-fastchat-worker:v0.2.36"
 	// defaultKubeAGIImage for RunnerKubeAGI
 	defaultKubeAGIImage = "kubeagi/core-library-cli:v0.0.1"
 )