Search results

CUDA SETUP ERROR: Missing dependency: libnvJitLink.so.13 - Google Colab

Status: Open.

#1905 In bitsandbytes-foundation/bitsandbytes;

· tanvircr7 opened on Mar 25, 2026

Params4bit.__getattr__ breaks torch.compile - use @property instead

Status: Open.

#1904 In bitsandbytes-foundation/bitsandbytes;

· kbabiuchx opened on Mar 23, 2026

Question: intentional FP16-only path for int8_vectorwise_quant / LLM.int8 activation quant? (BF16 support + removing casts)

Status: Open.

Feature

#1868 In bitsandbytes-foundation/bitsandbytes;

· sanghyunna opened on Feb 16, 2026

Default LLM.int8() mixed-precision decomposition causes 17-147% energy overhead across consumer and datacenter GPUs

Status: Open.

#1867 In bitsandbytes-foundation/bitsandbytes;

· hongping-zh opened on Feb 15, 2026

gemv_4bit silently produces wrong results when weight is quantized in (in_features, out_features) layout

Status: Open.

#1862 In bitsandbytes-foundation/bitsandbytes;

· TimDettmers opened on Feb 14, 2026

[Feature Gap] CUDA compared to other backends like XPU/CPU

Intel

Status: Open.

#1852 In bitsandbytes-foundation/bitsandbytes;

· jiqing-feng opened on Jan 30, 2026

[Performance/Energy] 4-bit NF4 shows significant energy efficiency penalty on Blackwell (RTX 5090) for small models

Status: Open.

#1851 In bitsandbytes-foundation/bitsandbytes;

· hongping-zh opened on Jan 29, 2026

Failed to quant MoE models with fused expert weights in transformers v5

Hugging Face IntegrationAn issue or PR that is related to the interaction between bitsandbytes and HF libraries.An issue or PR that is related to the interaction between bitsandbytes and HF libraries.

Status: Open.

#1849 In bitsandbytes-foundation/bitsandbytes;

· ITcarrot opened on Jan 25, 2026

# 70B 4-bit LLM decode bottlenecked by HIP kernel (kgemm_4bit_inference_naive) efficiency — 49% vs 91% memory bandwidth on ROCm/gfx1151

ROCm

Status: Open.

#1842 In bitsandbytes-foundation/bitsandbytes;

· BellaDoggie opened on Jan 19, 2026

Support quantizing tensors when numel() > INT_MAX

CUDAIssues and PRs related to the CUDA backend, excluding installation/support help.Issues and PRs related to the CUDA backend, excluding installation/support help.

Status: Open.

Feature

#1785 In bitsandbytes-foundation/bitsandbytes;

· matthewdouglas opened on Oct 22, 2025

·  

v0.50.0

Reduce CUDA build matrix

Build

CUDAIssues and PRs related to the CUDA backend, excluding installation/support help.Issues and PRs related to the CUDA backend, excluding installation/support help.

Status: Open.

Task

#1778 In bitsandbytes-foundation/bitsandbytes;

· matthewdouglas opened on Oct 3, 2025

·  

v0.50.0

Can't get llm_int8_skip_modules to work: 'Parameter' object has no attribute 'SCB'

BugSomething isn't workingSomething isn't working

Status: Open.

#1634 In bitsandbytes-foundation/bitsandbytes;

· redbrain opened on May 12, 2025