This is an automated email from the ASF dual-hosted git repository.
github-bot pushed a change to branch nightly
in repository https://gitbox.apache.org/repos/asf/tvm.git
from e56c5e1ef2 [Bugfix][Transform] Preserve symbolic variables in FuseOps
(#16637)
add e5420436a0 [Dlight] Skip GeMV when normalization fails (#16665)
add 3b25588926 [TOPI] improve inclusive_scan for thrust (#16652)
add 5718ff35ef [Relax][Runtime] Support Unpack API for NDArrayCache
(#16648)
add 73b01eec4a Fixing workload comment (#16662)
add ad1da4ee57 [Runtime][Builtin] Using float32 accumulation in attention
kernel (#16667)
No new revisions were added by this update.
Summary of changes:
python/tvm/auto_scheduler/workload_registry.py | 2 +-
python/tvm/dlight/gpu/gemv.py | 2 ++
python/tvm/topi/cuda/scan.py | 42 +++++++++++++++++-----
src/runtime/relax_vm/ndarray_cache_support.cc | 15 +++++++-
tests/python/dlight/test_gpu_gemv.py | 33 +++++++++++++++++
...runtime_builtin_paged_attention_kv_cache_tir.py | 12 +++----
6 files changed, 89 insertions(+), 17 deletions(-)
