v0.9.6

0.9.6 (2026-03-27)

Bug Fixes

• graph: skip all quantized storage in EnsureSlotsGPU/EnsureCaptureInputsGPU (0b38668)

v0.9.5

0.9.5 (2026-03-27)

Bug Fixes

• compute: skip MmapStorage entirely in UploadWeights (8796fd0)

v0.9.4

0.9.4 (2026-03-27)

Bug Fixes

• compute: copy mmap bytes to heap before cudaMemcpy upload (c2d68e7)

v0.9.3

0.9.3 (2026-03-27)

Bug Fixes

• graph: skip quantized storage in PreUploadFrozenWeights (4b8388c)

v0.9.2

0.9.2 (2026-03-27)

Bug Fixes

• compute: skip F32 MmapStorage in quantized upload path (51ed3e7)

v0.9.1

0.9.1 (2026-03-27)

Bug Fixes

• tensor: delegate K-quant MmapStorage dequant to reference implementations (3ef8261)

v0.9.0

0.9.0 (2026-03-27)

Features

• compute: add MmapStorage GPU dispatch for quantized GEMV/GEMM (62f3db1)

v0.8.0

0.8.0 (2026-03-27)

Features

• tensor: add Q4_1/Q5_0/Q5_1 support for MmapStorage (8adb879)

v0.7.0

0.7.0 (2026-03-27)

Features

• tensor: add MmapStorage type and platform mmap helpers (f8b48bb)

v0.6.3

0.6.3 (2026-03-27)

Bug Fixes

• compute: change Repeat to repeat-each semantics for GQA correctness (d3e6b96)