1 file changed
+1
-1
lines changed- .github/workflows/1xH100_tests.yml+3-3
- .github/workflows/1xL4_tests.yml+1-1
- benchmarks/_models/eval_hf_models.py+4-3
- benchmarks/float8/training/llama3.sh
- benchmarks/float8/training/llama4.sh+41
- benchmarks/mx_formats/cast_bench.py+54-22
- benchmarks/prototype/moe_training/benchmark_kernels.py+15-14
- benchmarks/prototype/moe_training/benchmark_moe_layer.py+179
- benchmarks/prototype/moe_training/benchmark_per_group_scaling_kernels.py+190
- benchmarks/prototype/moe_training/benchmark_rowwise_3d_quant_kernels.py+151
- benchmarks/prototype/moe_training/benchmark_scaled_grouped_mm.py
- test/core/test_config.py+15-11
- test/dtypes/test_affine_quantized_float.py+53-24
- test/dtypes/test_affine_quantized_tensor_parallel.py+3
- test/dtypes/test_fbgemm_fp8.py-153
- test/dtypes/test_nf4.py+4
- test/float8/test_base.py+3-3
- test/integration/test_integration.py+1
- test/integration/test_loading_deprecated_checkpoint.py+70
- test/prototype/moe_training/test_kernels.py+43-4
- test/prototype/moe_training/test_scaled_grouped_mm.py+1
- test/prototype/moe_training/test_training.py+118-7
- test/prototype/test_dynamic_activation_lut.py+1-1
- test/prototype/test_quantized_training.py+4
- test/prototype/test_smoothquant.py+216-137
- test/quantization/quantize_/workflows/float8/test_float8_tensor.py+563
- test/quantization/test_qat.py+40
- test/test_low_bit_optim.py+4
- torchao/core/config.py+31-34
- torchao/dtypes/floatx/float8_layout.py+4
- torchao/experimental/CMakeLists.txt+32-14
- torchao/experimental/kernels/cpu/aarch64/CMakeLists.txt-17
- torchao/experimental/ops/tests/CMakeLists.txt+21
- torchao/experimental/tests/test_quant_passes.py+6-2
- torchao/float8/config.py+1
- torchao/prototype/moe_training/conversion_utils.py+14-3
- torchao/prototype/moe_training/kernels/__init__.py+3
- torchao/prototype/moe_training/kernels/float8_rowwise.py+255
- torchao/prototype/moe_training/kernels/jagged_float8_scales.py+47-35
- torchao/prototype/moe_training/scaled_grouped_mm.py+36-38
- torchao/prototype/moe_training/tensor.py+27-8
- torchao/prototype/moe_training/utils.py+38-2
- torchao/prototype/mx_formats/kernels.py+10
- torchao/prototype/mx_formats/utils.py+1-1
- torchao/prototype/quantization/codebook_groupwise/codebook_quantized_tensor.py+220
- torchao/quantization/__init__.py+2
- torchao/quantization/pt2e/utils.py+2
- torchao/quantization/qat/api.py+15-7
- torchao/quantization/qat/fake_quantize_config.py+17-1
- torchao/quantization/qat/utils.py+32
- torchao/quantization/quant_api.py+84-35
- torchao/quantization/quantize_/common/__init__.py+11
- torchao/quantization/quantize_/common/kernel_preference.py+37
- torchao/quantization/quantize_/common/quantize_tensor_kwargs.py+56
- torchao/quantization/quantize_/workflows/__init__.py+6
- torchao/quantization/quantize_/workflows/float8/__init__.py
- torchao/quantization/quantize_/workflows/float8/float8_tensor.py+613
- torchao/testing/training/roofline_utils.py+16-8
0 commit comments