Search results

Research: 32K-vocab English-optimized small model for quant.cpp

enhancementNew feature or requestNew feature or request

Status: Open.

#92 In quantumaikr/quant.cpp;

· unamedkr opened on Apr 12, 2026

Port Phi-3 fused QKV/FFN from quant.h to libturboquant (enables Metal GPU)

enhancementNew feature or requestNew feature or request

Status: Open.

#91 In quantumaikr/quant.cpp;

· unamedkr opened on Apr 12, 2026

RLV Locator: add lightweight semantic search as third RRF signal

enhancementNew feature or requestNew feature or request

Status: Open.

#89 In quantumaikr/quant.cpp;

· unamedkr opened on Apr 12, 2026

Single Source of Truth: auto-generate split sources from quant.h

enhancementNew feature or requestNew feature or request

Status: Open.

#85 In quantumaikr/quant.cpp;

· unamedkr opened on Apr 12, 2026

KV cache pre-build for RLV: question latency 35s → 5s

enhancementNew feature or requestNew feature or request

Status: Open.

#83 In quantumaikr/quant.cpp;

· unamedkr opened on Apr 12, 2026

Qwen3.5-4B DeltaNet layers: FP32 dequant bottleneck causes 0.7 tok/s

enhancementNew feature or requestNew feature or request

Status: Open.

#70 In quantumaikr/quant.cpp;

· unamedkr opened on Apr 12, 2026

Qwen3.5-4B: quant_generate works but quant_ask produces empty/garbage output

bugSomething isn't workingSomething isn't working

Status: Open.

#69 In quantumaikr/quant.cpp;

· unamedkr opened on Apr 12, 2026

Metal backend is currently slower than CPU-only on all tested models

Status: Open.

#16 In quantumaikr/quant.cpp;

· unamedkr opened on Apr 8, 2026

TurboQuant follow-ups: per-channel outliers, Llama 3.1 8B reproduction, 5-bit codebook

Status: Open.

#15 In quantumaikr/quant.cpp;

· unamedkr opened on Apr 7, 2026