Search results

How is alternating optimization for MSE-optimized grids implemented in MR-GPTQ / FP-Quant?

Status: Open.

#25 In IST-DASLab/FP-Quant;

· efsotr opened on Apr 21, 2026

gsm8k evaluation score is extremely low when evaluating Llama-3.1-8B-Instruct

Status: Open.

#24 In IST-DASLab/FP-Quant;

· ShiyiWang01 opened on Mar 10, 2026

Question: Any reason for not using random H matrix?

Status: Open.

#23 In IST-DASLab/FP-Quant;

· chkim1103 opened on Feb 27, 2026

About default order implementation and performance

Status: Open.

#22 In IST-DASLab/FP-Quant;

· JeffLee1874 opened on Feb 3, 2026

Global_scale is passed but not applied in MXFP quantization, yet used in GEMM, could this cause numerical inequivalence?

Status: Open.

#21 In IST-DASLab/FP-Quant;

· manyizhang opened on Jan 20, 2026

AttributeError: 'InputCollector' object has no attribute 'attention_type' during GPTQ quantization with nvfp format

Status: Open.

#20 In IST-DASLab/FP-Quant;

· 547435524 opened on Jan 6, 2026

Regarding the Hadamard transform accuracy drop for nvfp4 with RTN in the paper

Status: Open.

#19 In IST-DASLab/FP-Quant;

· TheTinyTeddy opened on Dec 4, 2025

When using transformers.set_seed() instead of src.utils.common_utils.fix_seed(), W4A4 openllm chat-template–related task accuracies drop to 0, but bf16 accuracies are fine.

Status: Open.

#17 In IST-DASLab/FP-Quant;

· WeiweiZhang1 opened on Nov 25, 2025

BUG: Duplicate bias addition in PseudoQuant4x16NoMasterFn causes incorrect inference output

Status: Open.

#16 In IST-DASLab/FP-Quant;

· manyizhang opened on Nov 22, 2025

Realquant checkpoint vllm compatibility bug

Status: Open.

#15 In IST-DASLab/FP-Quant;

· songsm921 opened on Nov 20, 2025

Evaluation Extremely Slow After Quantization

Status: Open.

#8 In IST-DASLab/FP-Quant;

· slchenchn opened on Sep 22, 2025

Environmens

Status: Open.

#4 In IST-DASLab/FP-Quant;

· Aaronhuang-778 opened on Aug 18, 2025