diff --git a/computility-run.yaml b/computility-run.yaml
index 7d356832..b009c3fb 100644
--- a/computility-run.yaml
+++ b/computility-run.yaml
@@ -49,6 +49,4 @@ env:
       value: '1'
     - name: PYTORCH_CUDA_ALLOC_CONF
       value: max_split_size_mb:512
-    - name: LD_PRELOAD
-      value: /workspace/qwen3_6_scripts/cccl_preload/libcccl_allocator.so
 
diff --git a/qwen3_6_scripts/patch_xformers_sdpa_seq.py b/qwen3_6_scripts/patch_xformers_sdpa_seq.py
index af7037c2..8de1565a 100644
--- a/qwen3_6_scripts/patch_xformers_sdpa_seq.py
+++ b/qwen3_6_scripts/patch_xformers_sdpa_seq.py
@@ -200,10 +200,6 @@ FALLBACK_METHOD = '''
         max_seqlen = max(seq_lens_list)
 
         try:
-            # Skip flash_attn during profiling — OOMs on large dummy batch
-            import os
-            if os.environ.get("BI100_IN_STARTUP_PROFILE") == "1":
-                raise RuntimeError("skip flash_attn during profiling")
             out = _ixf.flash_attn_varlen_func(
                 q_flat.to(torch.float16),
                 k_flat.to(torch.float16),