Toggle theme

gemma.cpp

Workflow: Detection type:

MCP

Show Harden Runner enabled runs only

Showing workflow runs for branch:test_991744020

Clear filter

When `MxFp4Codec::QuantizeBlock` encounters an all-zero input block, the early-out path zeroes `kBlockSize` (32) bytes instead of the encoded block size `sizeof(MxFp4Stream)` (17 bytes: 1 scale byte plus 16 nibble bytes). Writing 15 bytes past the end of the block corrupts the adjacent block when blocks are encoded out of order (such as across parallel batch boundaries in `Compress` or via `packed_ofs` sub-ranges) and writes past `PackedEnd` on the final block of a tensor. test_991744020 build #8212build.yml10 days ago 5Destinations 7Actions

Insights

When `MxFp4Codec::QuantizeBlock` encounters an all-zero input block, the early-out path zeroes `kBlockSize` (32) bytes instead of the encoded block size `sizeof(MxFp4Stream)` (17 bytes: 1 scale byte plus 16 nibble bytes). Writing 15 bytes past the end of the block corrupts the adjacent block when blocks are encoded out of order (such as across parallel batch boundaries in `Compress` or via `packed_ofs` sub-ranges) and writes past `PackedEnd` on the final block of a tensor. test_991744020 build #8211build.yml10 days ago 5Destinations 7Actions

Insights