vllm kvprune wo:v1.1.1

d4645504 · chenzk · d29c39ca · d29c39ca · d29c39ca · d29c39ca
Commit d4645504 authored Apr 30, 2026 by chenzk
20 changed files
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/compaction.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/compaction.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/compaction_details/__init__.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/compaction_details/__init__.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/compaction_details/_masked_compaction.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/compaction_details/_masked_compaction.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/__init__.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/__init__.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/_common.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/_common.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/_matmul_ogs.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/_matmul_ogs.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/_p_matmul_ogs.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/_p_matmul_ogs.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/_reduce_grouped.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/_reduce_grouped.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/opt_flags.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/opt_flags.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/opt_flags_details/__init__.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/opt_flags_details/__init__.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/opt_flags_details/opt_flags_amd.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/opt_flags_details/opt_flags_amd.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/opt_flags_details/opt_flags_nvidia.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/matmul_ogs_details/opt_flags_details/opt_flags_nvidia.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics_details/__init__.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics_details/__init__.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics_details/flexpoint.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics_details/flexpoint.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics_details/mxfp.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics_details/mxfp.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics_details/mxfp_details/__init__.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics_details/mxfp_details/__init__.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics_details/mxfp_details/_downcast_to_mxfp.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics_details/mxfp_details/_downcast_to_mxfp.py
--- a/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics_details/mxfp_details/_upcast_from_mxfp.py
+++ b/vllm/compactor-vllm/src/compactor_vllm/triton_kernels/numerics_details/mxfp_details/_upcast_from_mxfp.py