remove unused env var

Author: Gasoonjia

.ci/scripts/test_model_e2e.sh

@@ -355,8 +355,6 @@ EOF
     ;;
   qwen3_5_moe)
     RUNNER_ARGS="$RUNNER_ARGS --tokenizer_path ${MODEL_DIR}/$TOKENIZER_FILE --prompt 'What is the capital of France?' --max_new_tokens 128 --temperature 0 --cuda_graph"
-    # CUDA graph capture requires cudaMallocAsync backend for stream-ordered allocations
-    export PYTORCH_CUDA_ALLOC_CONF=backend:cudaMallocAsync
     ;;
   voxtral_realtime)
     RUNNER_ARGS="--model_path ${MODEL_DIR}/model.pte --tokenizer_path ${MODEL_DIR}/$TOKENIZER_FILE --preprocessor_path ${MODEL_DIR}/$PREPROCESSOR --audio_path ${MODEL_DIR}/$AUDIO_FILE --temperature 0"

examples/models/qwen3_5_moe/main.cpp

@@ -133,8 +133,8 @@ int main(int argc, char** argv) {
 
   // Use prefill method for T>=2, decode method for T=1
   // (prefill was exported with min seq_len=2)
-  std::string run_method = prefill_method;
-  if (dual_method && num_prompt_tokens == 1) {
+  std::string run_method = "prefill";
+  if (num_prompt_tokens == 1) {
     run_method = "decode";
   }