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 9404fb5a59 [CI] Unpin pytest-profiling (#17620)
add 9898039456 [docs] Download 3rd party embeds to local files
add 3eb5ad6711 [KVCache] TIR attention kernel support for MLA (#17618)
No new revisions were added by this update.
Summary of changes:
docs/Makefile | 4 +
docs/download_3rdparty_embeds.py | 314 ++++
docs/script_convert.py | 100 --
python/tvm/relax/frontend/nn/llm/kv_cache.py | 1895 ++++++++++++++------
python/tvm/relax/frontend/nn/llm/tree_attn.py | 24 +-
src/runtime/relax_vm/kv_state.cc | 9 +
src/runtime/relax_vm/kv_state.h | 24 +-
src/runtime/relax_vm/paged_kv_cache.cc | 226 ++-
tests/python/ci/test_script_converter.py | 173 --
..._builtin_paged_attention_kv_cache_flashinfer.py | 229 +--
...ime_builtin_paged_attention_kv_cache_mla_tir.py | 456 +++++
...runtime_builtin_paged_attention_kv_cache_tir.py | 30 +-
tests/scripts/task_python_docs.sh | 2 +-
13 files changed, 2343 insertions(+), 1143 deletions(-)
create mode 100644 docs/download_3rdparty_embeds.py
delete mode 100644 docs/script_convert.py
delete mode 100644 tests/python/ci/test_script_converter.py
create mode 100644
tests/python/relax/test_runtime_builtin_paged_attention_kv_cache_mla_tir.py
