.

build_lora_image.sh

@@ -0,0 +1 @@
+docker build   --build-arg http_proxy=http://127.0.0.1:7890   --build-arg https_proxy=http://127.0.0.1:7890   --network=host   --build-arg CUDA_VERSION=12.6.1 --target runtime-autobuild --build-arg MAX_JOBS=96  --load -t sglang_lora .

build_lora_wheels.sh

@@ -0,0 +1,2 @@
+export DOCKER_BUILDKIT=1 
+docker build --build-arg http_proxy=http://127.0.0.1:7890   --build-arg https_proxy=http://127.0.0.1:7890   --network=host   --build-arg CUDA_VERSION=12.6.1 --target wheelhouse --output type=local,dest=./_wheelhouse  --build-arg MAX_JOBS=96 .

run_docker_lora_image.sh

@@ -0,0 +1,11 @@
+docker run -d \
+  --name sglang_lora \
+  --gpus all \
+  --shm-size=8g \
+  -e NVIDIA_VISIBLE_DEVICES=all \
+  -e SUPABASE_URL=http://183.36.35.42 \
+  -p 30000:30000 \
+  -p 30001:30001 \
+  -v /home/jzy/Qwen3-32B:/root/.cradle/external/llm \
+  -v /home/jzy/adapters:/root/.cradle/external/lora/q3 \
+  sglang_lora:latest

supervisord.conf

@@ -5,7 +5,7 @@ logfile_maxbytes=0
 loglevel=info
 
 [program:sglang]
-command=python3 -m sglang.launch_server --host 0.0.0.0 --port 30000 --model-path /root/.cradle/external/llm/ --lora-paths q3=/root/.cradle/external/lora/q3 --disable-radix-cache --tp 4 --api-key token-abc123 --enable-metrics
+command=python3 -m sglang.launch_server --host 0.0.0.0 --port 30000 --model-path /root/.cradle/external/llm/ --lora-paths q3=/root/.cradle/external/lora/q3 --lora-target-modules q_proj,k_proj,v_proj,o_proj,gate_proj,up_proj,down_proj --max-lora-rank 16 --served-model-name qwen3-32b --disable-radix-cache --tp 4 --api-key token-abc123 --enable-metrics --log-requests --log-requests-level 2
 autostart=true
 autorestart=true
 stdout_logfile=/dev/stdout