Issues

• CUDA vec FA kernel: turbo4 V-cache branch consumes 4 of 8 elements per iteration — half of every head's V output is zero on batch-1 decode (patch attached)

  TheTom/turboquant_plus#99 · Filter by author morpheos-llc opened on Sep 3, 2026
  1

• uv pip install turboquant also required

  TheTom/turboquant_plus#97 · Filter by author jwtiyar opened on Jul 26, 2026

• Can it be used together with the MTP draft model?

  TheTom/turboquant_plus#92 · Filter by author serchcc opened on May 10, 2026
  7

• [ROCm] Scale operation fails with "invalid device function" during Gemma 4 loading

  TheTom/turboquant_plus#86 · Filter by author http403 opened on Apr 23, 2026

• Feature request: prebuilt binary for Windows CPU only

  TheTom/turboquant_plus#85 · Filter by author jadams777 opened on Apr 23, 2026

• Diagnostic: [RTX 3090]

  TheTom/turboquant_plus#84 · Filter by author chertykov opened on Apr 20, 2026

• Question about QJL abliation study

  TheTom/turboquant_plus#82 · Filter by author ryusaeba opened on Apr 16, 2026

• time and performance overhead of quantization and dequantization

  TheTom/turboquant_plus#80 · Filter by author Szh1107 opened on Apr 13, 2026

• Reproducibility of QJL resurrection issue

  TheTom/turboquant_plus#79 · Filter by author zhangsipeng opened on Apr 13, 2026

• it run

  TheTom/turboquant_plus#74 · Filter by author ayttop opened on Apr 9, 2026
  2

• End-to-end inference test fails with --cache-type turbo4 on CPU

  TheTom/turboquant_plus#72 · Filter by author MELANCHOLY828 opened on Apr 9, 2026

• Error when building from source

  TheTom/turboquant_plus#70 · Filter by author gabiiblt opened on Apr 3, 2026

• polar coordinate transformation operations

  TheTom/turboquant_plus#69 · Filter by author Keshawn082 opened on Apr 3, 2026

• llama-quantize crashes with ios_base::failbit when using --tensor-type-file or --tensor-type (regression from #20503)

  TheTom/turboquant_plus#68 · Filter by author dogukanatakul opened on Apr 3, 2026

• [Bug] Numerical instability (NaN/Infinite Loop) with Qwen 2.5/3.5 models using turbo4 on RTX 4070Ti

  TheTom/turboquant_plus#60 · Filter by author david01333 opened on Apr 1, 2026

• Greetings from TurboQuant-MLX & Some Insights to Share! (System Prompt vs Boundary Layers)

  TheTom/turboquant_plus#59 · Filter by author helgklaizar opened on Apr 1, 2026

• Unsupported gpu architecture 'compute_120a'

  TheTom/turboquant_plus#58 · Filter by author Ezzz-dev opened on Mar 31, 2026
  2

• turbo - cant compile in Windows

  TheTom/turboquant_plus#57 · Filter by author Ezzz-dev opened on Mar 31, 2026
  1

• turbo3/turbo4 cache produces garbled output on NVIDIA Blackwell GPU (RTX 5070 Laptop, compute capability 12.0)

  TheTom/turboquant_plus#56 · Filter by author arifkurnaz opened on Mar 31, 2026
  1

• Norm correction validated on vLLM + Nemotron hybrid

  TheTom/turboquant_plus#55 · Filter by author MidasMining opened on Mar 31, 2026

• Diagnostic: Mac Studio M2 Ultra

  TheTom/turboquant_plus#50 · Filter by author dylanzr opened on Mar 31, 2026

• feat: GPU as async dequant coprocessor for CPU-primary inference

  TheTom/turboquant_plus#49 · Filter by author homeofe opened on Mar 30, 2026

• No Difference in tokens/sec - Ministral3 8B Q5_K_M

  TheTom/turboquant_plus#48 · Filter by author MrMuhannadObeidat opened on Mar 29, 2026
  3

• CUDA doesn't work

  TheTom/turboquant_plus#47 · Filter by author DragonRollDev opened on Mar 29, 2026

• Don't work on vulkan device

  TheTom/turboquant_plus#46 · Filter by author vincalkr opened on Mar 28, 2026
  3