Bump torchao from 0.17.0 to 0.18.0

[dependabot]

Bumps torchao from 0.17.0 to 0.18.0.

Release notes

Sourced from torchao's releases.

v0.18.0

Highlights

NVFP4 training

NVFP4 for dense linears is now supported for training (prototype), implementing the recipe from Pretraining Large Language Models with NVFP4. This is a prototype feature and the API may change without notice. Hardening, MoE support, and torchtitan integration are planned.

Example usage:

import torch
from torchao.quantization import quantize_
from torchao.prototype.moe_training.nvfp4_training.nvfp4_training import (
    NVFP4TrainingConfig,
)
Blackwell (SM100+); bfloat16 model with Linear dims divisible by 128
model = model.cuda().bfloat16()
Replace nn.Linear with NVFP4Linear
quantize_(model, NVFP4TrainingConfig())
Train as usual
out = model(x)
out.sum().backward()

• [nvfp4_training] Support NVFP4 linear training on Blackwell by @​rdspring1 in pytorch/ao#4369
• [nvfp4_training] Add Triton kernel for global amax of columnwise RHT (SM90+) by @​rdspring1 in pytorch/ao#4247

BC-Breaking / Deprecations

Minimum PyTorch version is now 2.11

Support for PyTorch older than 2.11 has been dropped from the code, and CI has moved to newer PyTorch versions.

• drop PyTorch 2.7 and older support from the code by @​vkuzo in pytorch/ao#4395
• drop PyTorch<2.8 support from the code by @​vkuzo in pytorch/ao#4398
• drop PyTorch<2.9 support by @​vkuzo in pytorch/ao#4399
• drop PyTorch <2.10 support by @​vkuzo in pytorch/ao#4400
• drop PyTorch <2.11 support by @​vkuzo in pytorch/ao#4401
• switch CI to PyTorch 2.12 by @​vkuzo in pytorch/ao#4402
• Pin CI to PyTorch 2.13 for release; drop unused torchaudio by @​vkuzo in pytorch/ao#4638

Removed: v1 tensor-subclass / layout system

The deprecated v1 AffineQuantizedTensor stack and layout system have been fully removed, completing the migration to v2 tensor subclasses.

• Delete AffineQuantizedTensor, AQTTensorImpl, and Layout by @​andrewor14 in pytorch/ao#4245
• Delete deprecated TensorCoreTiledLayout and related code by @​jerryzh168 in pytorch/ao#4153
• Delete deprecated PlainLayout, PlainAQTTensorImpl and related v1 code paths by @​jerryzh168 in pytorch/ao#4151
• Remove LinearActivationQuantizedTensor and all related code by @​jerryzh168 in pytorch/ao#4258
• Remove LinearActivationWeightObservedTensor and insert_observers_ by @​jerryzh168 in pytorch/ao#4261
• Remove WeightTensorWithLinearActivationScaleMetadata and related code by @​jerryzh168 in pytorch/ao#4262
• Remove WeightTensorWithLinearActivationQuantizationMetadata by @​jerryzh168 in pytorch/ao#4263

... (truncated)

Commits

• 5f2baf9 update test-infra branch to 0.18
• 79a65df update README for optional dependencies (#4666)
• e42806f update CODEOWNERS (#4662)
• 721217b fix h100 sparse CI (#4661)
• 55c8e6e float8: emit tl.where inline for the Triton fp8 store workaround (#4657)
• e054801 Route ROCm regression CI to MI350 ecosystem runners (#4621)
• ac7bcee delete the microbenchmarking system (#4642)
• dbef699 mxfp8: make RCEIL NaN handling CSE-friendly (#4654)
• e81aa6e mxfp8: use versioned TMA lookup with CUDA 13 (#4653)
• 3ffa31c float8: work around Triton fp8 store miscompile in compiled training (#4652)
• Additional commits viewable in compare view