revert attention-backend, set image to vllm/vllm-openai:v0.26.0

Author: xinli-sw

benchmarks/single_node/fixed_seq_len/dsv4_fp4_b200_vllm.sh

@@ -100,7 +100,6 @@ vllm serve "$MODEL" --host 0.0.0.0 --port "$PORT" \
     "${PREFILL_SCHEDULE_ARGS[@]}" \
     "${DEP_COMPILE_ARGS[@]}" \
     --attention_config.use_fp4_indexer_cache=True \
-    --attention-backend FLASHMLA_SPARSE_DSV4 \
     --tokenizer-mode deepseek_v4 \
     --tool-call-parser deepseek_v4 \
     --enable-auto-tool-choice \

configs/nvidia-master.yaml

@@ -929,7 +929,7 @@ dsv4-fp4-b200-sglang:
       - { tp: 8, ep: 8, dp-attn: true, conc-start: 256, conc-end: 1024 }
 
 dsv4-fp4-b200-vllm:
-  image: vllm/vllm-openai:nightly-700d39b55813fce6fe5339b334beb55d2a39f598
+  image: vllm/vllm-openai:v0.26.0
   model: deepseek-ai/DeepSeek-V4-Pro
   model-prefix: dsv4
   runner: cluster:b200-dgxc