Revert "Change dsr1 fp8 image to lmsysorg/sglang 0.5.5.post3 and fp4 image to…" (#295)

This reverts commit 0fe9dcc.

Author: cquil11

.github/configs/amd-master.yaml

@@ -1,5 +1,5 @@
 dsr1-fp4-mi355x-sglang:
-  image: lmsysorg/sglang:v0.5.5.post2-rocm700-mi35x
+  image: rocm/7.0:rocm7.0_ubuntu_22.04_sgl-dev-v0.5.2-rocm7.0-mi35x-20250915
   model: amd/DeepSeek-R1-0528-MXFP4-Preview
   model-prefix: dsr1
   runner: mi355x
@@ -63,7 +63,7 @@ dsr1-fp8-mi325x-sglang:
     - { tp: 8, conc-start: 4, conc-end: 64 }
 
 dsr1-fp8-mi355x-sglang:
-  image: lmsysorg/sglang:v0.5.5.post3-rocm700-mi35x
+  image: rocm/7.0:rocm7.0_ubuntu_22.04_sgl-dev-v0.5.2-rocm7.0-mi35x-20250915
   model: deepseek-ai/DeepSeek-R1-0528
   model-prefix: dsr1
   runner: mi355x

benchmarks/dsr1_fp4_mi355x_docker.sh

@@ -11,7 +11,6 @@
 # RESULT_FILENAME
 # NUM_PROMPTS
 export SGLANG_USE_AITER=1
-export ROCM_QUICK_REDUCE_QUANTIZATION=INT4
 
 PREFILL_SIZE=196608
 if [[ "$ISL" == "8192" && "$OSL" == "1024" ]]; then

benchmarks/dsr1_fp4_mi355x_slurm.sh

@@ -10,7 +10,6 @@
 # RANDOM_RANGE_RATIO
 # RESULT_FILENAME
 export SGLANG_USE_AITER=1
-export ROCM_QUICK_REDUCE_QUANTIZATION=INT4
 SERVER_LOG=$(mktemp /tmp/server-XXXXXX.log)
 
 PREFILL_SIZE=196608

benchmarks/dsr1_fp8_mi355x_docker.sh

@@ -14,14 +14,10 @@
 # https://rocm.docs.amd.com/en/docs-7.0-docker/benchmark-docker/inference-sglang-deepseek-r1-fp8.html
 
 export SGLANG_USE_AITER=1
-export RCCL_MSCCL_ENABLE=0
-export ROCM_QUICK_REDUCE_QUANTIZATION=INT4
-
 
 SERVER_LOG=$(mktemp /tmp/server-XXXXXX.log)
 
 python3 -m sglang.launch_server \
-    --attention-backend aiter \
     --model-path $MODEL \
     --host=0.0.0.0 \
     --port $PORT \
@@ -31,7 +27,6 @@ python3 -m sglang.launch_server \
     --mem-fraction-static 0.8 --disable-radix-cache \
     --num-continuous-decode-steps 4 \
     --max-prefill-tokens 196608 \
-    --enable-torch-compile \
     --cuda-graph-max-bs 128 > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!

benchmarks/dsr1_fp8_mi355x_slurm.sh

@@ -12,14 +12,11 @@
 
 export HF_MODULES_CACHE="/tmp/hf_modules_cache/"
 export SGLANG_USE_AITER=1
-export RCCL_MSCCL_ENABLE=0
-export ROCM_QUICK_REDUCE_QUANTIZATION=INT4
 
 SERVER_LOG=$(mktemp /tmp/server-XXXXXX.log)
 
 set -x
 python3 -m sglang.launch_server \
-    --attention-backend aiter \
     --model-path $MODEL \
     --host=0.0.0.0 \
     --port $PORT \
@@ -30,8 +27,7 @@ python3 -m sglang.launch_server \
     --disable-radix-cache \
     --num-continuous-decode-steps 4 \
     --max-prefill-tokens 196608 \
-    --cuda-graph-max-bs 128 \
-    --enable-torch-compile > $SERVER_LOG 2>&1 &
+    --cuda-graph-max-bs 128 > $SERVER_LOG 2>&1 &
 
 SERVER_PID=$!