Unverified Commit 4c4130e3 authored by ishandhanani's avatar ishandhanani Committed by GitHub
Browse files

chore: typo and new commands (#3617)

parent 5744548b
...@@ -61,7 +61,7 @@ spec: ...@@ -61,7 +61,7 @@ spec:
image: my-registry/sglang-runtime:my-tag image: my-registry/sglang-runtime:my-tag
workingDir: /workspace/components/backends/sglang workingDir: /workspace/components/backends/sglang
command: command:
- python3E - python3
- -m - -m
- dynamo.sglang - dynamo.sglang
args: args:
......
...@@ -16,10 +16,6 @@ trap cleanup EXIT INT TERM ...@@ -16,10 +16,6 @@ trap cleanup EXIT INT TERM
python3 -m dynamo.frontend --http-port=8000 & python3 -m dynamo.frontend --http-port=8000 &
DYNAMO_PID=$! DYNAMO_PID=$!
# Set the expert distribution recording directory
mkdir -p /tmp/sglang_expert_distribution_record
export SGLANG_EXPERT_DISTRIBUTION_RECORDER_DIR=/tmp/sglang_expert_distribution_record
# run prefill worker # run prefill worker
python3 -m dynamo.sglang \ python3 -m dynamo.sglang \
--model-path silence09/DeepSeek-R1-Small-2layers \ --model-path silence09/DeepSeek-R1-Small-2layers \
...@@ -31,7 +27,7 @@ python3 -m dynamo.sglang \ ...@@ -31,7 +27,7 @@ python3 -m dynamo.sglang \
--trust-remote-code \ --trust-remote-code \
--disaggregation-mode prefill \ --disaggregation-mode prefill \
--disaggregation-transfer-backend nixl \ --disaggregation-transfer-backend nixl \
--expert-distribution-recorder-mode stat \ --load-balance-method round_robin \
--port 30000 & --port 30000 &
PREFILL_PID=$! PREFILL_PID=$!
...@@ -46,5 +42,5 @@ CUDA_VISIBLE_DEVICES=2,3 python3 -m dynamo.sglang \ ...@@ -46,5 +42,5 @@ CUDA_VISIBLE_DEVICES=2,3 python3 -m dynamo.sglang \
--trust-remote-code \ --trust-remote-code \
--disaggregation-mode decode \ --disaggregation-mode decode \
--disaggregation-transfer-backend nixl \ --disaggregation-transfer-backend nixl \
--expert-distribution-recorder-mode stat \ --prefill-round-robin-balance \
--port 31000 --port 31000
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment