From aebc660a102679b2b2e4838eb72ff8d2ffbac590 Mon Sep 17 00:00:00 2001
From: project6-dev <dev@project6.ai>
Date: Thu, 13 Aug 2026 15:08:40 +0000
Subject: [PATCH] revert: restore to 8c8c0286 (last confirmed build success)

Revert LD_PRELOAD addition and patch_xformers profiling skip.
Need to identify which change caused build failure before re-adding.
---
 computility-run.yaml                       | 2 --
 qwen3_6_scripts/patch_xformers_sdpa_seq.py | 4 ----
 2 files changed, 6 deletions(-)

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),