-
Notifications
You must be signed in to change notification settings - Fork 590
All issues
Issue creation is restricted in this repository
Issues
is:issue state:open
is:issue state:open
Search results
Build Linux Wheels (x86, cuda13.0): pip install torch nightly fails with ProtocolError IncompleteRead (network interruption)
ci-failureAutomated report of CI failure on mainAutomated report of CI failure on mainStatus: Open.#4757 In pytorch/ao;MPS Experimental Tests: self-hosted macOS runner lost communication during "Setup miniconda" (no tests ran)
ci-failureAutomated report of CI failure on mainAutomated report of CI failure on mainStatus: Open.#4756 In pytorch/ao;- Status: Open.#4713 In pytorch/ao;
test_fp32_fp8_multi_module_parity (FSDP2 float8): NaN/0.0 loss parity mismatch on 4xH100 (torch 2.13.0)
ci-failureAutomated report of CI failure on mainAutomated report of CI failure on mainStatus: Open.#4708 In pytorch/ao;ConvRot — Group-wise Regular Hadamard Rotation for W4A4/W8A8 Diffusion Transformer Quantization
enhancementNew feature or requestNew feature or requestStatus: Open.#4695 In pytorch/ao;ZeroPointDomain (class) vs zero_point_domain (variable) typo
bugSomething isn't workingSomething isn't workingStatus: Open.#4693 In pytorch/ao;convrot for int8 and fp4
enhancementNew feature or requestNew feature or requestStatus: Open.#4685 In pytorch/ao;ROCm Nightly (gfx950): fp8 hipSPARSELt sparse tests fail "hipSPARSELt not supported" plus gfx950 mxfp8/GPTQ numeric mismatches (6 failed)
ci-failureAutomated report of CI failure on mainAutomated report of CI failure on mainStatus: Open.#4663 In pytorch/ao;test_fp8_cutlass_sparse: torchao::to_sparse_semi_structured_cutlass_sm9x_f8 CUDA op not registered at runtime on torch nightly (1xH100)
ci-failureAutomated report of CI failure on mainAutomated report of CI failure on mainStatus: Open.#4655 In pytorch/ao;Blockwise FP8 training GEMMs read past the end of both scale tensors
bugSomething isn't workingSomething isn't workingStatus: Open.#4634 In pytorch/ao;int_scaled_matmulreads out of bounds and returns wrong results when K is even but not a multiple of `BLOCK_KbugSomething isn't workingSomething isn't workingStatus: Open.#4633 In pytorch/ao;RFC: 2:4 sparsity for Linear and Conv2d
enhancementNew feature or requestNew feature or requestStatus: Open.#4626 In pytorch/ao;