Glq releases

No immediate action

v0.5.1 Breaking risk · 7h

Default inline-dequant E8 KV

Open

No immediate action

v0.5.0 New feature · 3d

Inline-dequant E8 KV cache

Open

No immediate action

v0.3.5 New feature · 16d

Auto‑PIECEWISE downgrade

Open

No immediate action

v0.3.4 New feature · 16d

CUDA-graph capture size tuning

Open

No immediate action

v0.3.3 Bug fix · 17d

E8 KV regression fix + GLQ ops

Open

No immediate action

v0.3.2 Breaking risk · 17d

Faster decoding without flags

Open

No immediate action

v0.3.1 Bug fix · 17d

GLQShardedParameter duplicate allocation fix

Open

No immediate action

v0.2.13 Performance · 1mo

Throughput +19 %

Open

No immediate action

v0.2.12 New feature · 1mo

HF integration + fused MoE + torch pin

Open

No immediate action

v0.2.11 New feature · 1mo

CUDA graph buckets + N-stage matmul

Open

No immediate action

v0.2.10 New feature · 1mo

Decode speed boost

Open

No immediate action

v0.2.9 New feature · 1mo

Block‑diagonal FHT + N‑stage RVQ

Open

No immediate action

v0.2.8 Mixed · 2mo

Kernel + Inference + Quantization

Open

No immediate action

v0.2.7 New feature · 2mo

CUDA Graph + INT8 KV cache

Open

No immediate action

v0.2.6 New feature · 2mo

Inline PTX performance boost

Open

No immediate action

v0.2.5 Mixed · 2mo

CUDA C kernel speedups + deadlock fix

Open

No immediate action

v0.2.2 New feature · 2mo

Sensitivity profiling + bit‑allocation

Open

No immediate action

v0.1.6 New feature · 2mo

Tiled Triton kernel speedup

Open

No immediate action

v0.1.5 Bug fix · 2mo

Quantization fix

Open

No immediate action

v0.1.4 New feature · 2mo

CPU offloading for 7B+ quantization

Open

No immediate action

v0.1.3 Feature · 2mo

Triton kernel + benchmark script

Open

No immediate action

v0.1.2 Feature · 2mo

Triton fused dequant+matmul

Open

No immediate action

v0.1.1 Maintenance · 2mo

Routine maintenance and dependency updates.

Open

No immediate action

v0.1.0 Maintenance · 2mo

Routine maintenance and dependency updates.

Open