Search results

Test Accuracy dropped a lot after Dynamic Quantization for resnet18.

bugSomething isn't workingSomething isn't working

Status: Open.

#4292 In pytorch/ao;

· pentatracy5 opened on Apr 17, 2026

Rewrite deleted tutorials without AQT

docathon:hard

topic: documentationUse this tag if this PR adds or improves documentationUse this tag if this PR adds or improves documentation

Status: Open.

#4289 In pytorch/ao;

· andrewor14 opened on Apr 16, 2026

Update QAT docs (Apr 26)

docathon:medium

topic: documentationUse this tag if this PR adds or improves documentationUse this tag if this PR adds or improves documentation

Status: Open.

#4288 In pytorch/ao;

· andrewor14 opened on Apr 16, 2026

Fine-tuning docs still refer to torchtune

docathon:medium

topic: documentationUse this tag if this PR adds or improves documentationUse this tag if this PR adds or improves documentation

Status: Open.

#4287 In pytorch/ao;

· andrewor14 opened on Apr 16, 2026

[CPU][x86] Int8DynamicActivationIntxWeightConfig with DA8W4 configuration does not have packed-kernel ops for x86 CPUs?

cpu

Status: Open.

#4243 In pytorch/ao;

· aadwived opened on Apr 6, 2026

Moving BlockSparseWeightConfig to prototype

Status: Open.

#4230 In pytorch/ao;

· jerryzh168 opened on Apr 3, 2026

rocm should run an allowlist of tests in CI?

topic: rocm

Status: Open.

#4219 In pytorch/ao;

· jerryzh168 opened on Apr 2, 2026

Support 32x32 scaling for weights in MXFP8 weight quantization kernel

moe

mx

Status: Open.

#4185 In pytorch/ao;

· danielvegamyhre opened on Mar 26, 2026

·  

MXFP8 Training

fuse dynamic per group padding into cutedsl 2d MXFP8 quantization kernel

moe

mx

Status: Open.

#4184 In pytorch/ao;

· danielvegamyhre opened on Mar 26, 2026

·  

MXFP8 Training

[MXFP8 MoE] mx_block_rearrange_2d_M_groups_cuda fails for models with > 32 experts

module: trainingquantize_ api training flowquantize_ api training flow

moe

mx

Status: Open.

#4163 In pytorch/ao;

· yzhautouskay opened on Mar 24, 2026

·  

MXFP8 Training

mxfp8 RCEIL scaling should use optimized PTX instructions on CUDA

Status: Open.

#4140 In pytorch/ao;

· vkuzo opened on Mar 23, 2026

Buffer mutation tracking lost after convert_pt2e for index_put_

module: pt2e_quantpt2 export quantization (prepare_pt2e, convert_pt2e, quantizer)pt2 export quantization (prepare_pt2e, convert_pt2e, quantizer)

Status: Open.

#4138 In pytorch/ao;

· tom-arm opened on Mar 23, 2026