Promote latest kernel artifacts to main
Browse files- README.md +113 -6
- build/torch212-cxx11-cu130-x86_64-linux/__init__.py +74 -0
- build/torch212-cxx11-cu130-x86_64-linux/{_fp4_gemm_cuda_55c4885.abi3.so → _fp4_gemm_cuda_230e467.abi3.so} +2 -2
- build/torch212-cxx11-cu130-x86_64-linux/_ops.py +3 -3
- build/torch212-cxx11-cu130-x86_64-linux/metadata.json +5 -5
- build/torch212-cxx11-cu132-x86_64-linux/__init__.py +74 -0
- build/torch212-cxx11-cu132-x86_64-linux/{_fp4_gemm_cuda_55c4885.abi3.so → _fp4_gemm_cuda_230e467.abi3.so} +2 -2
- build/torch212-cxx11-cu132-x86_64-linux/_ops.py +3 -3
- build/torch212-cxx11-cu132-x86_64-linux/metadata.json +5 -5
- build/torch213-cxx11-cu130-x86_64-linux/__init__.py +74 -0
- build/torch213-cxx11-cu130-x86_64-linux/{_fp4_gemm_cuda_55c4885.abi3.so → _fp4_gemm_cuda_230e467.abi3.so} +2 -2
- build/torch213-cxx11-cu130-x86_64-linux/_ops.py +3 -3
- build/torch213-cxx11-cu130-x86_64-linux/metadata.json +5 -5
- build/torch213-cxx11-cu132-x86_64-linux/__init__.py +74 -0
- build/torch213-cxx11-cu132-x86_64-linux/{_fp4_gemm_cuda_55c4885.abi3.so → _fp4_gemm_cuda_230e467.abi3.so} +2 -2
- build/torch213-cxx11-cu132-x86_64-linux/_ops.py +3 -3
- build/torch213-cxx11-cu132-x86_64-linux/metadata.json +5 -5
README.md
CHANGED
|
@@ -1,9 +1,116 @@
|
|
| 1 |
-
#
|
| 2 |
|
| 3 |
-
|
| 4 |
-
that resolve repositories through the default Hugging Face model repo API.
|
| 5 |
|
| 6 |
-
|
|
|
|
|
|
|
|
|
|
| 7 |
|
| 8 |
-
|
| 9 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# fp4-gemm
|
| 2 |
|
| 3 |
+
FlashRT native Blackwell NVFP4 A4W4 GEMM kernels.
|
|
|
|
| 4 |
|
| 5 |
+
This package consumes packed FP4 E2M1 tensors plus CUTLASS Sm1xx SFA/SFB scale
|
| 6 |
+
buffers and produces BF16 output. It is designed to pair with
|
| 7 |
+
`flashrt/fp4-fused-ops` and other static low-bit transformer/diffuser runtime
|
| 8 |
+
paths.
|
| 9 |
|
| 10 |
+
## Available Functions
|
| 11 |
+
|
| 12 |
+
- `sfa_size_bytes(rows, dim)`
|
| 13 |
+
- `quantize_fp4_sfa_fp16(x, packed=None, sfa=None, is_sfb=False)`
|
| 14 |
+
- `quantize_fp4_sfa_bf16(x, packed=None, sfa=None, is_sfb=False)`
|
| 15 |
+
- `dequantize_fp4_sfa_fp16(packed, sfa, out=None, is_sfb=False)`
|
| 16 |
+
- `nvfp4_gemm_bf16(a_packed, b_packed, sfa, sfb, alpha=1.0, out=None, variant=-1)`
|
| 17 |
+
- `nvfp4_gemm_bias_bf16(a_packed, b_packed, sfa, sfb, bias, out=None)`
|
| 18 |
+
- `nvfp4_gemm_bias_residual_bf16(a_packed, b_packed, sfa, sfb, bias, residual, out=None)`
|
| 19 |
+
- `nvfp4_gemm_residual_bf16(a_packed, b_packed, sfa, sfb, residual, alpha=1.0, out=None)`
|
| 20 |
+
- `nvfp4_gemm_bias_gelu_bf16(a_packed, b_packed, sfa, sfb, bias, alpha=1.0, out=None)`
|
| 21 |
+
- `nvfp4_gemm_bias_gelu_nvfp4(a_packed, b_packed, sfa, sfb, bias, alpha=1.0, out_packed=None, out_sfa=None)`
|
| 22 |
+
- `nvfp4_gemm_streamk_bf16(a_packed, b_packed, sfa, sfb, alpha=1.0, out=None)`
|
| 23 |
+
- `nvfp4_gemm_streamk_bias_bf16(a_packed, b_packed, sfa, sfb, bias, alpha=1.0, out=None)`
|
| 24 |
+
- `fp4_w4a16_linear_bf16(...)` is retained as a compatibility alias
|
| 25 |
+
|
| 26 |
+
## Tensor Contract
|
| 27 |
+
|
| 28 |
+
- `a_packed`: `torch.uint8`, shape `(M, K / 2)`.
|
| 29 |
+
- `b_packed`: `torch.uint8`, shape `(N, K / 2)`.
|
| 30 |
+
- `sfa`: `torch.uint8`, CUTLASS SFA layout for `(M, K)`.
|
| 31 |
+
- `sfb`: `torch.uint8`, CUTLASS SFB layout for `(N, K)`.
|
| 32 |
+
- output: `torch.bfloat16`, shape `(M, N)`.
|
| 33 |
+
- `K` must be divisible by 16.
|
| 34 |
+
- Targets: Blackwell `sm_110a` (Jetson AGX Thor, CUDA 13+) and `sm_120a`
|
| 35 |
+
(RTX Blackwell, CUDA 12.8+).
|
| 36 |
+
|
| 37 |
+
`variant` selects the CUTLASS schedule:
|
| 38 |
+
|
| 39 |
+
- `-1`: architecture-aware auto-dispatch (public default).
|
| 40 |
+
- `0`: default `<128,128,256>` cooperative schedule.
|
| 41 |
+
- `1`: widen `<128,256,128>` schedule, intended for very large `N`.
|
| 42 |
+
- `2`: pingpong schedule for A/B testing shape-specific wins.
|
| 43 |
+
|
| 44 |
+
The canonical linear API and FP4/SFA quantize/dequantize helpers are available
|
| 45 |
+
on both SM110 and SM120. SM110 additionally provides the GROOT N1.7 production
|
| 46 |
+
epilogues `nvfp4_gemm_bias_bf16`, `nvfp4_gemm_bias_residual_bf16`, and
|
| 47 |
+
`nvfp4_gemm_bias_gelu_nvfp4`. The latter emits packed FP4 plus CUTLASS SFA so
|
| 48 |
+
the following projection can consume it without a BF16 materialization and a
|
| 49 |
+
standalone quantization launch. Stream-K and the older BF16 GELU epilogue keep
|
| 50 |
+
their existing SM120 dispatch and reject unsupported architectures explicitly.
|
| 51 |
+
|
| 52 |
+
The SM110 release gate includes the production `(M,N,K)` shapes
|
| 53 |
+
`(41,4608,1536)`, `(41,6144,1536)`, and `(41,1536,6144)`, plus the legacy
|
| 54 |
+
`M=51` compatibility row. The kernels are the native sources used by FlashRT's
|
| 55 |
+
GROOT N1.7 Thor NVFP4 pipeline.
|
| 56 |
+
|
| 57 |
+
## Minimal Usage
|
| 58 |
+
|
| 59 |
+
```python
|
| 60 |
+
from kernels import get_kernel
|
| 61 |
+
import torch
|
| 62 |
+
|
| 63 |
+
ops = get_kernel("flashrt/fp4-gemm", version=1, trust_remote_code=True)
|
| 64 |
+
|
| 65 |
+
x = torch.randn((32, 256), device="cuda", dtype=torch.float16)
|
| 66 |
+
w = torch.randn((512, 256), device="cuda", dtype=torch.float16)
|
| 67 |
+
|
| 68 |
+
a_packed, sfa = ops.quantize_fp4_sfa_fp16(x, is_sfb=False)
|
| 69 |
+
b_packed, sfb = ops.quantize_fp4_sfa_fp16(w, is_sfb=True)
|
| 70 |
+
|
| 71 |
+
y = ops.nvfp4_gemm_bf16(a_packed, b_packed, sfa, sfb, alpha=1.0)
|
| 72 |
+
```
|
| 73 |
+
|
| 74 |
+
For BF16 model activations, use the direct producer so the hot path does not
|
| 75 |
+
materialize an intermediate FP16 tensor:
|
| 76 |
+
|
| 77 |
+
```python
|
| 78 |
+
x_bf16 = torch.randn((1, 5120), device="cuda", dtype=torch.bfloat16)
|
| 79 |
+
a_packed, sfa = ops.quantize_fp4_sfa_bf16(x_bf16)
|
| 80 |
+
```
|
| 81 |
+
|
| 82 |
+
The BF16 entry writes the same E2M1 bytes and CUTLASS SFA/SFB layout as
|
| 83 |
+
`quantize_fp4_sfa_fp16(x_bf16.to(torch.float16))` for finite FP16-range
|
| 84 |
+
inputs. It is an additive API; the existing FP16 producer remains unchanged.
|
| 85 |
+
|
| 86 |
+
The quantize/dequantize helpers are included for examples and validation. A
|
| 87 |
+
production runtime should keep weights prepacked and should avoid quantizing in
|
| 88 |
+
the hot path unless that producer kernel is part of the intended low-bit block.
|
| 89 |
+
|
| 90 |
+
Use the bias/GELU and residual variants to avoid returning to BF16
|
| 91 |
+
elementwise code between low-bit GEMMs. Stream-K variants are selected only
|
| 92 |
+
for the validated large down-projection shapes; unsupported shapes reject
|
| 93 |
+
rather than silently selecting a losing schedule.
|
| 94 |
+
|
| 95 |
+
## Validation
|
| 96 |
+
|
| 97 |
+
```bash
|
| 98 |
+
python fp4-gemm/tests/test_fp4_gemm.py --backend source --mode full
|
| 99 |
+
python fp4-gemm/tests/test_fp4_gemm.py --backend installed --mode full \
|
| 100 |
+
--artifact fp4-gemm/build/torch211-cxx11-cu128-x86_64-linux
|
| 101 |
+
python fp4-gemm/benchmarks/benchmark.py --backend installed --mode headline \
|
| 102 |
+
--artifact fp4-gemm/build/torch211-cxx11-cu128-x86_64-linux
|
| 103 |
+
|
| 104 |
+
# Thor model-shape gate
|
| 105 |
+
python fp4-gemm/tests/test_fp4_gemm.py --backend installed \
|
| 106 |
+
--mode thor-models \
|
| 107 |
+
--artifact fp4-gemm/build/torch211-cxx11-cu130-aarch64-linux
|
| 108 |
+
```
|
| 109 |
+
|
| 110 |
+
The correctness reference dequantizes the same FP4/SFA and FP4/SFB inputs used
|
| 111 |
+
by the kernel, then computes the PyTorch GEMM reference from those dequantized
|
| 112 |
+
low-bit values.
|
| 113 |
+
|
| 114 |
+
The producer gate also checks the BF16 direct entry byte-for-byte against the
|
| 115 |
+
established FP16 compatibility chain at decode widths 5120, 6144 and 17408,
|
| 116 |
+
plus multi-row activation and SFB layouts.
|
build/torch212-cxx11-cu130-x86_64-linux/__init__.py
CHANGED
|
@@ -94,6 +94,21 @@ def _quant_fake(x: torch.Tensor, packed: torch.Tensor, sfa: torch.Tensor, is_sfb
|
|
| 94 |
return None
|
| 95 |
|
| 96 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
@torch.library.register_fake(add_op_namespace_prefix("quantize_fp4_sfa_bf16"))
|
| 98 |
def _quant_bf16_fake(x: torch.Tensor, packed: torch.Tensor, sfa: torch.Tensor, is_sfb: bool = False) -> None:
|
| 99 |
return None
|
|
@@ -143,6 +158,62 @@ def quantize_fp4_sfa_fp16(
|
|
| 143 |
return packed, sfa
|
| 144 |
|
| 145 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 146 |
def quantize_fp4_sfa_bf16(
|
| 147 |
x: torch.Tensor,
|
| 148 |
packed: torch.Tensor | None = None,
|
|
@@ -457,6 +528,7 @@ def nvfp4_gemm_streamk_bias_bf16(
|
|
| 457 |
|
| 458 |
__all__ = [
|
| 459 |
"dequantize_fp4_sfa_fp16",
|
|
|
|
| 460 |
"fp4_w4a16_linear_bf16",
|
| 461 |
"fp4_w4a4_gemv_warpsplit_bf16",
|
| 462 |
"nvfp4_gemm_bf16",
|
|
@@ -469,9 +541,11 @@ __all__ = [
|
|
| 469 |
"nvfp4_gemm_bias_gelu_nvfp4",
|
| 470 |
"nvfp4_gemm_bias_residual_bf16",
|
| 471 |
"nvfp4_gemm_residual_bf16",
|
|
|
|
| 472 |
"nvfp4_gemm_streamk_bf16",
|
| 473 |
"nvfp4_gemm_streamk_bias_bf16",
|
| 474 |
"quantize_fp4_sfa_fp16",
|
|
|
|
| 475 |
"quantize_fp4_sfa_bf16",
|
| 476 |
"sfa_size_bytes",
|
| 477 |
]
|
|
|
|
| 94 |
return None
|
| 95 |
|
| 96 |
|
| 97 |
+
@torch.library.register_fake(add_op_namespace_prefix("quantize_e0m3_sfa_fp16"))
|
| 98 |
+
def _quant_e0m3_fake(x, packed, sfa, is_sfb: bool = False) -> None:
|
| 99 |
+
return None
|
| 100 |
+
|
| 101 |
+
|
| 102 |
+
@torch.library.register_fake(add_op_namespace_prefix("e0m3_weight_gemm_fp16"))
|
| 103 |
+
def _e0m3_gemm_fake(a, b, sfa, sfb, out, alpha: float = 1.0, a_format: int = 1) -> None:
|
| 104 |
+
return None
|
| 105 |
+
|
| 106 |
+
|
| 107 |
+
@torch.library.register_fake(add_op_namespace_prefix("nvfp4_gemm_relu2_nvfp4"))
|
| 108 |
+
def _relu2_gemm_fake(a, b, sfa, sfb, out_packed, out_sfa) -> None:
|
| 109 |
+
return None
|
| 110 |
+
|
| 111 |
+
|
| 112 |
@torch.library.register_fake(add_op_namespace_prefix("quantize_fp4_sfa_bf16"))
|
| 113 |
def _quant_bf16_fake(x: torch.Tensor, packed: torch.Tensor, sfa: torch.Tensor, is_sfb: bool = False) -> None:
|
| 114 |
return None
|
|
|
|
| 158 |
return packed, sfa
|
| 159 |
|
| 160 |
|
| 161 |
+
def quantize_e0m3_sfa_fp16(
|
| 162 |
+
x: torch.Tensor,
|
| 163 |
+
packed: torch.Tensor | None = None,
|
| 164 |
+
sfa: torch.Tensor | None = None,
|
| 165 |
+
is_sfb: bool = False,
|
| 166 |
+
):
|
| 167 |
+
"""Quantize FP16 to packed E0M3 with CUTLASS SFA/SFB layout."""
|
| 168 |
+
if packed is None or sfa is None:
|
| 169 |
+
packed, sfa = _alloc_fp4(x.shape[0], x.shape[1], x.device)
|
| 170 |
+
ops.quantize_e0m3_sfa_fp16(x, packed, sfa, bool(is_sfb))
|
| 171 |
+
return packed, sfa
|
| 172 |
+
|
| 173 |
+
|
| 174 |
+
def e0m3_weight_gemm_fp16(
|
| 175 |
+
a_packed: torch.Tensor,
|
| 176 |
+
b_packed: torch.Tensor,
|
| 177 |
+
sfa: torch.Tensor,
|
| 178 |
+
sfb: torch.Tensor,
|
| 179 |
+
*,
|
| 180 |
+
alpha: float = 1.0,
|
| 181 |
+
a_format: int = 1,
|
| 182 |
+
out: torch.Tensor | None = None,
|
| 183 |
+
):
|
| 184 |
+
"""SM110 GEMM with E0M3 weights and E2M1 or E0M3 activations."""
|
| 185 |
+
if out is None:
|
| 186 |
+
out = torch.empty(
|
| 187 |
+
(a_packed.shape[0], b_packed.shape[0]),
|
| 188 |
+
device=a_packed.device,
|
| 189 |
+
dtype=torch.float16,
|
| 190 |
+
)
|
| 191 |
+
ops.e0m3_weight_gemm_fp16(
|
| 192 |
+
a_packed, b_packed, sfa, sfb, out, float(alpha), int(a_format)
|
| 193 |
+
)
|
| 194 |
+
return out
|
| 195 |
+
|
| 196 |
+
|
| 197 |
+
def nvfp4_gemm_relu2_nvfp4(
|
| 198 |
+
a_packed: torch.Tensor,
|
| 199 |
+
b_packed: torch.Tensor,
|
| 200 |
+
sfa: torch.Tensor,
|
| 201 |
+
sfb: torch.Tensor,
|
| 202 |
+
*,
|
| 203 |
+
out_packed: torch.Tensor | None = None,
|
| 204 |
+
out_sfa: torch.Tensor | None = None,
|
| 205 |
+
):
|
| 206 |
+
"""SM110 NVFP4 GEMM with fused ReLU-squared and NVFP4 output."""
|
| 207 |
+
if out_packed is None or out_sfa is None:
|
| 208 |
+
out_packed, out_sfa = _alloc_fp4(
|
| 209 |
+
a_packed.shape[0], b_packed.shape[0], a_packed.device
|
| 210 |
+
)
|
| 211 |
+
ops.nvfp4_gemm_relu2_nvfp4(
|
| 212 |
+
a_packed, b_packed, sfa, sfb, out_packed, out_sfa
|
| 213 |
+
)
|
| 214 |
+
return out_packed, out_sfa
|
| 215 |
+
|
| 216 |
+
|
| 217 |
def quantize_fp4_sfa_bf16(
|
| 218 |
x: torch.Tensor,
|
| 219 |
packed: torch.Tensor | None = None,
|
|
|
|
| 528 |
|
| 529 |
__all__ = [
|
| 530 |
"dequantize_fp4_sfa_fp16",
|
| 531 |
+
"e0m3_weight_gemm_fp16",
|
| 532 |
"fp4_w4a16_linear_bf16",
|
| 533 |
"fp4_w4a4_gemv_warpsplit_bf16",
|
| 534 |
"nvfp4_gemm_bf16",
|
|
|
|
| 541 |
"nvfp4_gemm_bias_gelu_nvfp4",
|
| 542 |
"nvfp4_gemm_bias_residual_bf16",
|
| 543 |
"nvfp4_gemm_residual_bf16",
|
| 544 |
+
"nvfp4_gemm_relu2_nvfp4",
|
| 545 |
"nvfp4_gemm_streamk_bf16",
|
| 546 |
"nvfp4_gemm_streamk_bias_bf16",
|
| 547 |
"quantize_fp4_sfa_fp16",
|
| 548 |
+
"quantize_e0m3_sfa_fp16",
|
| 549 |
"quantize_fp4_sfa_bf16",
|
| 550 |
"sfa_size_bytes",
|
| 551 |
]
|
build/torch212-cxx11-cu130-x86_64-linux/{_fp4_gemm_cuda_55c4885.abi3.so → _fp4_gemm_cuda_230e467.abi3.so}
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5e319eb3cf472a0ead98df64005aca885558f238a7b682e0767a27189d170e46
|
| 3 |
+
size 7181592
|
build/torch212-cxx11-cu130-x86_64-linux/_ops.py
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
import torch
|
| 2 |
-
from . import
|
| 3 |
-
ops = torch.ops.
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
-
return f"
|
|
|
|
| 1 |
import torch
|
| 2 |
+
from . import _fp4_gemm_cuda_230e467
|
| 3 |
+
ops = torch.ops._fp4_gemm_cuda_230e467
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_fp4_gemm_cuda_230e467::{op_name}"
|
build/torch212-cxx11-cu130-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "fp4-gemm",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
@@ -14,9 +14,9 @@
|
|
| 14 |
"digest": {
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
-
"__init__.py": "
|
| 18 |
-
"
|
| 19 |
-
"_ops.py": "
|
| 20 |
}
|
| 21 |
},
|
| 22 |
"provenance": {
|
|
@@ -26,7 +26,7 @@
|
|
| 26 |
"dirty": false
|
| 27 |
},
|
| 28 |
"kernel": {
|
| 29 |
-
"sha": "
|
| 30 |
"dirty": false
|
| 31 |
}
|
| 32 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"name": "fp4-gemm",
|
| 3 |
+
"id": "_fp4_gemm_cuda_230e467",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 14 |
"digest": {
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
+
"__init__.py": "Qfdvg/9EedWLGh34xAalDfC0IsAHCOL44L4ShaP/w5Y=",
|
| 18 |
+
"_fp4_gemm_cuda_230e467.abi3.so": "XjGes89HKg6tmN9kAFrKiFVY8jintoLgdnonGJ0XDkY=",
|
| 19 |
+
"_ops.py": "CxLLKrT37bp1ie5kAVTryKkFcmkRJqJ2fYXrSYuMHcA="
|
| 20 |
}
|
| 21 |
},
|
| 22 |
"provenance": {
|
|
|
|
| 26 |
"dirty": false
|
| 27 |
},
|
| 28 |
"kernel": {
|
| 29 |
+
"sha": "230e467129aa8e5fc687611fbb2aa40a12e09ad1",
|
| 30 |
"dirty": false
|
| 31 |
}
|
| 32 |
}
|
build/torch212-cxx11-cu132-x86_64-linux/__init__.py
CHANGED
|
@@ -94,6 +94,21 @@ def _quant_fake(x: torch.Tensor, packed: torch.Tensor, sfa: torch.Tensor, is_sfb
|
|
| 94 |
return None
|
| 95 |
|
| 96 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
@torch.library.register_fake(add_op_namespace_prefix("quantize_fp4_sfa_bf16"))
|
| 98 |
def _quant_bf16_fake(x: torch.Tensor, packed: torch.Tensor, sfa: torch.Tensor, is_sfb: bool = False) -> None:
|
| 99 |
return None
|
|
@@ -143,6 +158,62 @@ def quantize_fp4_sfa_fp16(
|
|
| 143 |
return packed, sfa
|
| 144 |
|
| 145 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 146 |
def quantize_fp4_sfa_bf16(
|
| 147 |
x: torch.Tensor,
|
| 148 |
packed: torch.Tensor | None = None,
|
|
@@ -457,6 +528,7 @@ def nvfp4_gemm_streamk_bias_bf16(
|
|
| 457 |
|
| 458 |
__all__ = [
|
| 459 |
"dequantize_fp4_sfa_fp16",
|
|
|
|
| 460 |
"fp4_w4a16_linear_bf16",
|
| 461 |
"fp4_w4a4_gemv_warpsplit_bf16",
|
| 462 |
"nvfp4_gemm_bf16",
|
|
@@ -469,9 +541,11 @@ __all__ = [
|
|
| 469 |
"nvfp4_gemm_bias_gelu_nvfp4",
|
| 470 |
"nvfp4_gemm_bias_residual_bf16",
|
| 471 |
"nvfp4_gemm_residual_bf16",
|
|
|
|
| 472 |
"nvfp4_gemm_streamk_bf16",
|
| 473 |
"nvfp4_gemm_streamk_bias_bf16",
|
| 474 |
"quantize_fp4_sfa_fp16",
|
|
|
|
| 475 |
"quantize_fp4_sfa_bf16",
|
| 476 |
"sfa_size_bytes",
|
| 477 |
]
|
|
|
|
| 94 |
return None
|
| 95 |
|
| 96 |
|
| 97 |
+
@torch.library.register_fake(add_op_namespace_prefix("quantize_e0m3_sfa_fp16"))
|
| 98 |
+
def _quant_e0m3_fake(x, packed, sfa, is_sfb: bool = False) -> None:
|
| 99 |
+
return None
|
| 100 |
+
|
| 101 |
+
|
| 102 |
+
@torch.library.register_fake(add_op_namespace_prefix("e0m3_weight_gemm_fp16"))
|
| 103 |
+
def _e0m3_gemm_fake(a, b, sfa, sfb, out, alpha: float = 1.0, a_format: int = 1) -> None:
|
| 104 |
+
return None
|
| 105 |
+
|
| 106 |
+
|
| 107 |
+
@torch.library.register_fake(add_op_namespace_prefix("nvfp4_gemm_relu2_nvfp4"))
|
| 108 |
+
def _relu2_gemm_fake(a, b, sfa, sfb, out_packed, out_sfa) -> None:
|
| 109 |
+
return None
|
| 110 |
+
|
| 111 |
+
|
| 112 |
@torch.library.register_fake(add_op_namespace_prefix("quantize_fp4_sfa_bf16"))
|
| 113 |
def _quant_bf16_fake(x: torch.Tensor, packed: torch.Tensor, sfa: torch.Tensor, is_sfb: bool = False) -> None:
|
| 114 |
return None
|
|
|
|
| 158 |
return packed, sfa
|
| 159 |
|
| 160 |
|
| 161 |
+
def quantize_e0m3_sfa_fp16(
|
| 162 |
+
x: torch.Tensor,
|
| 163 |
+
packed: torch.Tensor | None = None,
|
| 164 |
+
sfa: torch.Tensor | None = None,
|
| 165 |
+
is_sfb: bool = False,
|
| 166 |
+
):
|
| 167 |
+
"""Quantize FP16 to packed E0M3 with CUTLASS SFA/SFB layout."""
|
| 168 |
+
if packed is None or sfa is None:
|
| 169 |
+
packed, sfa = _alloc_fp4(x.shape[0], x.shape[1], x.device)
|
| 170 |
+
ops.quantize_e0m3_sfa_fp16(x, packed, sfa, bool(is_sfb))
|
| 171 |
+
return packed, sfa
|
| 172 |
+
|
| 173 |
+
|
| 174 |
+
def e0m3_weight_gemm_fp16(
|
| 175 |
+
a_packed: torch.Tensor,
|
| 176 |
+
b_packed: torch.Tensor,
|
| 177 |
+
sfa: torch.Tensor,
|
| 178 |
+
sfb: torch.Tensor,
|
| 179 |
+
*,
|
| 180 |
+
alpha: float = 1.0,
|
| 181 |
+
a_format: int = 1,
|
| 182 |
+
out: torch.Tensor | None = None,
|
| 183 |
+
):
|
| 184 |
+
"""SM110 GEMM with E0M3 weights and E2M1 or E0M3 activations."""
|
| 185 |
+
if out is None:
|
| 186 |
+
out = torch.empty(
|
| 187 |
+
(a_packed.shape[0], b_packed.shape[0]),
|
| 188 |
+
device=a_packed.device,
|
| 189 |
+
dtype=torch.float16,
|
| 190 |
+
)
|
| 191 |
+
ops.e0m3_weight_gemm_fp16(
|
| 192 |
+
a_packed, b_packed, sfa, sfb, out, float(alpha), int(a_format)
|
| 193 |
+
)
|
| 194 |
+
return out
|
| 195 |
+
|
| 196 |
+
|
| 197 |
+
def nvfp4_gemm_relu2_nvfp4(
|
| 198 |
+
a_packed: torch.Tensor,
|
| 199 |
+
b_packed: torch.Tensor,
|
| 200 |
+
sfa: torch.Tensor,
|
| 201 |
+
sfb: torch.Tensor,
|
| 202 |
+
*,
|
| 203 |
+
out_packed: torch.Tensor | None = None,
|
| 204 |
+
out_sfa: torch.Tensor | None = None,
|
| 205 |
+
):
|
| 206 |
+
"""SM110 NVFP4 GEMM with fused ReLU-squared and NVFP4 output."""
|
| 207 |
+
if out_packed is None or out_sfa is None:
|
| 208 |
+
out_packed, out_sfa = _alloc_fp4(
|
| 209 |
+
a_packed.shape[0], b_packed.shape[0], a_packed.device
|
| 210 |
+
)
|
| 211 |
+
ops.nvfp4_gemm_relu2_nvfp4(
|
| 212 |
+
a_packed, b_packed, sfa, sfb, out_packed, out_sfa
|
| 213 |
+
)
|
| 214 |
+
return out_packed, out_sfa
|
| 215 |
+
|
| 216 |
+
|
| 217 |
def quantize_fp4_sfa_bf16(
|
| 218 |
x: torch.Tensor,
|
| 219 |
packed: torch.Tensor | None = None,
|
|
|
|
| 528 |
|
| 529 |
__all__ = [
|
| 530 |
"dequantize_fp4_sfa_fp16",
|
| 531 |
+
"e0m3_weight_gemm_fp16",
|
| 532 |
"fp4_w4a16_linear_bf16",
|
| 533 |
"fp4_w4a4_gemv_warpsplit_bf16",
|
| 534 |
"nvfp4_gemm_bf16",
|
|
|
|
| 541 |
"nvfp4_gemm_bias_gelu_nvfp4",
|
| 542 |
"nvfp4_gemm_bias_residual_bf16",
|
| 543 |
"nvfp4_gemm_residual_bf16",
|
| 544 |
+
"nvfp4_gemm_relu2_nvfp4",
|
| 545 |
"nvfp4_gemm_streamk_bf16",
|
| 546 |
"nvfp4_gemm_streamk_bias_bf16",
|
| 547 |
"quantize_fp4_sfa_fp16",
|
| 548 |
+
"quantize_e0m3_sfa_fp16",
|
| 549 |
"quantize_fp4_sfa_bf16",
|
| 550 |
"sfa_size_bytes",
|
| 551 |
]
|
build/torch212-cxx11-cu132-x86_64-linux/{_fp4_gemm_cuda_55c4885.abi3.so → _fp4_gemm_cuda_230e467.abi3.so}
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:42c8b744abb8d1df65520ba3135b88c58a002b31bb4fcc1b66b823c961ee815d
|
| 3 |
+
size 7169248
|
build/torch212-cxx11-cu132-x86_64-linux/_ops.py
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
import torch
|
| 2 |
-
from . import
|
| 3 |
-
ops = torch.ops.
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
-
return f"
|
|
|
|
| 1 |
import torch
|
| 2 |
+
from . import _fp4_gemm_cuda_230e467
|
| 3 |
+
ops = torch.ops._fp4_gemm_cuda_230e467
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_fp4_gemm_cuda_230e467::{op_name}"
|
build/torch212-cxx11-cu132-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "fp4-gemm",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
@@ -14,9 +14,9 @@
|
|
| 14 |
"digest": {
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
-
"__init__.py": "
|
| 18 |
-
"
|
| 19 |
-
"_ops.py": "
|
| 20 |
}
|
| 21 |
},
|
| 22 |
"provenance": {
|
|
@@ -26,7 +26,7 @@
|
|
| 26 |
"dirty": false
|
| 27 |
},
|
| 28 |
"kernel": {
|
| 29 |
-
"sha": "
|
| 30 |
"dirty": false
|
| 31 |
}
|
| 32 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"name": "fp4-gemm",
|
| 3 |
+
"id": "_fp4_gemm_cuda_230e467",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 14 |
"digest": {
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
+
"__init__.py": "Qfdvg/9EedWLGh34xAalDfC0IsAHCOL44L4ShaP/w5Y=",
|
| 18 |
+
"_fp4_gemm_cuda_230e467.abi3.so": "Qsi3RKu40d9lUgujE1uIxYoAKzG7T8wbZrgjyWHugV0=",
|
| 19 |
+
"_ops.py": "CxLLKrT37bp1ie5kAVTryKkFcmkRJqJ2fYXrSYuMHcA="
|
| 20 |
}
|
| 21 |
},
|
| 22 |
"provenance": {
|
|
|
|
| 26 |
"dirty": false
|
| 27 |
},
|
| 28 |
"kernel": {
|
| 29 |
+
"sha": "230e467129aa8e5fc687611fbb2aa40a12e09ad1",
|
| 30 |
"dirty": false
|
| 31 |
}
|
| 32 |
}
|
build/torch213-cxx11-cu130-x86_64-linux/__init__.py
CHANGED
|
@@ -94,6 +94,21 @@ def _quant_fake(x: torch.Tensor, packed: torch.Tensor, sfa: torch.Tensor, is_sfb
|
|
| 94 |
return None
|
| 95 |
|
| 96 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
@torch.library.register_fake(add_op_namespace_prefix("quantize_fp4_sfa_bf16"))
|
| 98 |
def _quant_bf16_fake(x: torch.Tensor, packed: torch.Tensor, sfa: torch.Tensor, is_sfb: bool = False) -> None:
|
| 99 |
return None
|
|
@@ -143,6 +158,62 @@ def quantize_fp4_sfa_fp16(
|
|
| 143 |
return packed, sfa
|
| 144 |
|
| 145 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 146 |
def quantize_fp4_sfa_bf16(
|
| 147 |
x: torch.Tensor,
|
| 148 |
packed: torch.Tensor | None = None,
|
|
@@ -457,6 +528,7 @@ def nvfp4_gemm_streamk_bias_bf16(
|
|
| 457 |
|
| 458 |
__all__ = [
|
| 459 |
"dequantize_fp4_sfa_fp16",
|
|
|
|
| 460 |
"fp4_w4a16_linear_bf16",
|
| 461 |
"fp4_w4a4_gemv_warpsplit_bf16",
|
| 462 |
"nvfp4_gemm_bf16",
|
|
@@ -469,9 +541,11 @@ __all__ = [
|
|
| 469 |
"nvfp4_gemm_bias_gelu_nvfp4",
|
| 470 |
"nvfp4_gemm_bias_residual_bf16",
|
| 471 |
"nvfp4_gemm_residual_bf16",
|
|
|
|
| 472 |
"nvfp4_gemm_streamk_bf16",
|
| 473 |
"nvfp4_gemm_streamk_bias_bf16",
|
| 474 |
"quantize_fp4_sfa_fp16",
|
|
|
|
| 475 |
"quantize_fp4_sfa_bf16",
|
| 476 |
"sfa_size_bytes",
|
| 477 |
]
|
|
|
|
| 94 |
return None
|
| 95 |
|
| 96 |
|
| 97 |
+
@torch.library.register_fake(add_op_namespace_prefix("quantize_e0m3_sfa_fp16"))
|
| 98 |
+
def _quant_e0m3_fake(x, packed, sfa, is_sfb: bool = False) -> None:
|
| 99 |
+
return None
|
| 100 |
+
|
| 101 |
+
|
| 102 |
+
@torch.library.register_fake(add_op_namespace_prefix("e0m3_weight_gemm_fp16"))
|
| 103 |
+
def _e0m3_gemm_fake(a, b, sfa, sfb, out, alpha: float = 1.0, a_format: int = 1) -> None:
|
| 104 |
+
return None
|
| 105 |
+
|
| 106 |
+
|
| 107 |
+
@torch.library.register_fake(add_op_namespace_prefix("nvfp4_gemm_relu2_nvfp4"))
|
| 108 |
+
def _relu2_gemm_fake(a, b, sfa, sfb, out_packed, out_sfa) -> None:
|
| 109 |
+
return None
|
| 110 |
+
|
| 111 |
+
|
| 112 |
@torch.library.register_fake(add_op_namespace_prefix("quantize_fp4_sfa_bf16"))
|
| 113 |
def _quant_bf16_fake(x: torch.Tensor, packed: torch.Tensor, sfa: torch.Tensor, is_sfb: bool = False) -> None:
|
| 114 |
return None
|
|
|
|
| 158 |
return packed, sfa
|
| 159 |
|
| 160 |
|
| 161 |
+
def quantize_e0m3_sfa_fp16(
|
| 162 |
+
x: torch.Tensor,
|
| 163 |
+
packed: torch.Tensor | None = None,
|
| 164 |
+
sfa: torch.Tensor | None = None,
|
| 165 |
+
is_sfb: bool = False,
|
| 166 |
+
):
|
| 167 |
+
"""Quantize FP16 to packed E0M3 with CUTLASS SFA/SFB layout."""
|
| 168 |
+
if packed is None or sfa is None:
|
| 169 |
+
packed, sfa = _alloc_fp4(x.shape[0], x.shape[1], x.device)
|
| 170 |
+
ops.quantize_e0m3_sfa_fp16(x, packed, sfa, bool(is_sfb))
|
| 171 |
+
return packed, sfa
|
| 172 |
+
|
| 173 |
+
|
| 174 |
+
def e0m3_weight_gemm_fp16(
|
| 175 |
+
a_packed: torch.Tensor,
|
| 176 |
+
b_packed: torch.Tensor,
|
| 177 |
+
sfa: torch.Tensor,
|
| 178 |
+
sfb: torch.Tensor,
|
| 179 |
+
*,
|
| 180 |
+
alpha: float = 1.0,
|
| 181 |
+
a_format: int = 1,
|
| 182 |
+
out: torch.Tensor | None = None,
|
| 183 |
+
):
|
| 184 |
+
"""SM110 GEMM with E0M3 weights and E2M1 or E0M3 activations."""
|
| 185 |
+
if out is None:
|
| 186 |
+
out = torch.empty(
|
| 187 |
+
(a_packed.shape[0], b_packed.shape[0]),
|
| 188 |
+
device=a_packed.device,
|
| 189 |
+
dtype=torch.float16,
|
| 190 |
+
)
|
| 191 |
+
ops.e0m3_weight_gemm_fp16(
|
| 192 |
+
a_packed, b_packed, sfa, sfb, out, float(alpha), int(a_format)
|
| 193 |
+
)
|
| 194 |
+
return out
|
| 195 |
+
|
| 196 |
+
|
| 197 |
+
def nvfp4_gemm_relu2_nvfp4(
|
| 198 |
+
a_packed: torch.Tensor,
|
| 199 |
+
b_packed: torch.Tensor,
|
| 200 |
+
sfa: torch.Tensor,
|
| 201 |
+
sfb: torch.Tensor,
|
| 202 |
+
*,
|
| 203 |
+
out_packed: torch.Tensor | None = None,
|
| 204 |
+
out_sfa: torch.Tensor | None = None,
|
| 205 |
+
):
|
| 206 |
+
"""SM110 NVFP4 GEMM with fused ReLU-squared and NVFP4 output."""
|
| 207 |
+
if out_packed is None or out_sfa is None:
|
| 208 |
+
out_packed, out_sfa = _alloc_fp4(
|
| 209 |
+
a_packed.shape[0], b_packed.shape[0], a_packed.device
|
| 210 |
+
)
|
| 211 |
+
ops.nvfp4_gemm_relu2_nvfp4(
|
| 212 |
+
a_packed, b_packed, sfa, sfb, out_packed, out_sfa
|
| 213 |
+
)
|
| 214 |
+
return out_packed, out_sfa
|
| 215 |
+
|
| 216 |
+
|
| 217 |
def quantize_fp4_sfa_bf16(
|
| 218 |
x: torch.Tensor,
|
| 219 |
packed: torch.Tensor | None = None,
|
|
|
|
| 528 |
|
| 529 |
__all__ = [
|
| 530 |
"dequantize_fp4_sfa_fp16",
|
| 531 |
+
"e0m3_weight_gemm_fp16",
|
| 532 |
"fp4_w4a16_linear_bf16",
|
| 533 |
"fp4_w4a4_gemv_warpsplit_bf16",
|
| 534 |
"nvfp4_gemm_bf16",
|
|
|
|
| 541 |
"nvfp4_gemm_bias_gelu_nvfp4",
|
| 542 |
"nvfp4_gemm_bias_residual_bf16",
|
| 543 |
"nvfp4_gemm_residual_bf16",
|
| 544 |
+
"nvfp4_gemm_relu2_nvfp4",
|
| 545 |
"nvfp4_gemm_streamk_bf16",
|
| 546 |
"nvfp4_gemm_streamk_bias_bf16",
|
| 547 |
"quantize_fp4_sfa_fp16",
|
| 548 |
+
"quantize_e0m3_sfa_fp16",
|
| 549 |
"quantize_fp4_sfa_bf16",
|
| 550 |
"sfa_size_bytes",
|
| 551 |
]
|
build/torch213-cxx11-cu130-x86_64-linux/{_fp4_gemm_cuda_55c4885.abi3.so → _fp4_gemm_cuda_230e467.abi3.so}
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c5c6596baac50bb9b9ebb3ac5acc544c62cb33959c4ea64bdd8e8fd039bdb294
|
| 3 |
+
size 7181432
|
build/torch213-cxx11-cu130-x86_64-linux/_ops.py
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
import torch
|
| 2 |
-
from . import
|
| 3 |
-
ops = torch.ops.
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
-
return f"
|
|
|
|
| 1 |
import torch
|
| 2 |
+
from . import _fp4_gemm_cuda_230e467
|
| 3 |
+
ops = torch.ops._fp4_gemm_cuda_230e467
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_fp4_gemm_cuda_230e467::{op_name}"
|
build/torch213-cxx11-cu130-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "fp4-gemm",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
@@ -14,9 +14,9 @@
|
|
| 14 |
"digest": {
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
-
"__init__.py": "
|
| 18 |
-
"
|
| 19 |
-
"_ops.py": "
|
| 20 |
}
|
| 21 |
},
|
| 22 |
"provenance": {
|
|
@@ -26,7 +26,7 @@
|
|
| 26 |
"dirty": false
|
| 27 |
},
|
| 28 |
"kernel": {
|
| 29 |
-
"sha": "
|
| 30 |
"dirty": false
|
| 31 |
}
|
| 32 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"name": "fp4-gemm",
|
| 3 |
+
"id": "_fp4_gemm_cuda_230e467",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 14 |
"digest": {
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
+
"__init__.py": "Qfdvg/9EedWLGh34xAalDfC0IsAHCOL44L4ShaP/w5Y=",
|
| 18 |
+
"_fp4_gemm_cuda_230e467.abi3.so": "xcZZa6rFC7m567OsWsxUTGLLM5WcTqZL3Y6P0Dm9spQ=",
|
| 19 |
+
"_ops.py": "CxLLKrT37bp1ie5kAVTryKkFcmkRJqJ2fYXrSYuMHcA="
|
| 20 |
}
|
| 21 |
},
|
| 22 |
"provenance": {
|
|
|
|
| 26 |
"dirty": false
|
| 27 |
},
|
| 28 |
"kernel": {
|
| 29 |
+
"sha": "230e467129aa8e5fc687611fbb2aa40a12e09ad1",
|
| 30 |
"dirty": false
|
| 31 |
}
|
| 32 |
}
|
build/torch213-cxx11-cu132-x86_64-linux/__init__.py
CHANGED
|
@@ -94,6 +94,21 @@ def _quant_fake(x: torch.Tensor, packed: torch.Tensor, sfa: torch.Tensor, is_sfb
|
|
| 94 |
return None
|
| 95 |
|
| 96 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 97 |
@torch.library.register_fake(add_op_namespace_prefix("quantize_fp4_sfa_bf16"))
|
| 98 |
def _quant_bf16_fake(x: torch.Tensor, packed: torch.Tensor, sfa: torch.Tensor, is_sfb: bool = False) -> None:
|
| 99 |
return None
|
|
@@ -143,6 +158,62 @@ def quantize_fp4_sfa_fp16(
|
|
| 143 |
return packed, sfa
|
| 144 |
|
| 145 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 146 |
def quantize_fp4_sfa_bf16(
|
| 147 |
x: torch.Tensor,
|
| 148 |
packed: torch.Tensor | None = None,
|
|
@@ -457,6 +528,7 @@ def nvfp4_gemm_streamk_bias_bf16(
|
|
| 457 |
|
| 458 |
__all__ = [
|
| 459 |
"dequantize_fp4_sfa_fp16",
|
|
|
|
| 460 |
"fp4_w4a16_linear_bf16",
|
| 461 |
"fp4_w4a4_gemv_warpsplit_bf16",
|
| 462 |
"nvfp4_gemm_bf16",
|
|
@@ -469,9 +541,11 @@ __all__ = [
|
|
| 469 |
"nvfp4_gemm_bias_gelu_nvfp4",
|
| 470 |
"nvfp4_gemm_bias_residual_bf16",
|
| 471 |
"nvfp4_gemm_residual_bf16",
|
|
|
|
| 472 |
"nvfp4_gemm_streamk_bf16",
|
| 473 |
"nvfp4_gemm_streamk_bias_bf16",
|
| 474 |
"quantize_fp4_sfa_fp16",
|
|
|
|
| 475 |
"quantize_fp4_sfa_bf16",
|
| 476 |
"sfa_size_bytes",
|
| 477 |
]
|
|
|
|
| 94 |
return None
|
| 95 |
|
| 96 |
|
| 97 |
+
@torch.library.register_fake(add_op_namespace_prefix("quantize_e0m3_sfa_fp16"))
|
| 98 |
+
def _quant_e0m3_fake(x, packed, sfa, is_sfb: bool = False) -> None:
|
| 99 |
+
return None
|
| 100 |
+
|
| 101 |
+
|
| 102 |
+
@torch.library.register_fake(add_op_namespace_prefix("e0m3_weight_gemm_fp16"))
|
| 103 |
+
def _e0m3_gemm_fake(a, b, sfa, sfb, out, alpha: float = 1.0, a_format: int = 1) -> None:
|
| 104 |
+
return None
|
| 105 |
+
|
| 106 |
+
|
| 107 |
+
@torch.library.register_fake(add_op_namespace_prefix("nvfp4_gemm_relu2_nvfp4"))
|
| 108 |
+
def _relu2_gemm_fake(a, b, sfa, sfb, out_packed, out_sfa) -> None:
|
| 109 |
+
return None
|
| 110 |
+
|
| 111 |
+
|
| 112 |
@torch.library.register_fake(add_op_namespace_prefix("quantize_fp4_sfa_bf16"))
|
| 113 |
def _quant_bf16_fake(x: torch.Tensor, packed: torch.Tensor, sfa: torch.Tensor, is_sfb: bool = False) -> None:
|
| 114 |
return None
|
|
|
|
| 158 |
return packed, sfa
|
| 159 |
|
| 160 |
|
| 161 |
+
def quantize_e0m3_sfa_fp16(
|
| 162 |
+
x: torch.Tensor,
|
| 163 |
+
packed: torch.Tensor | None = None,
|
| 164 |
+
sfa: torch.Tensor | None = None,
|
| 165 |
+
is_sfb: bool = False,
|
| 166 |
+
):
|
| 167 |
+
"""Quantize FP16 to packed E0M3 with CUTLASS SFA/SFB layout."""
|
| 168 |
+
if packed is None or sfa is None:
|
| 169 |
+
packed, sfa = _alloc_fp4(x.shape[0], x.shape[1], x.device)
|
| 170 |
+
ops.quantize_e0m3_sfa_fp16(x, packed, sfa, bool(is_sfb))
|
| 171 |
+
return packed, sfa
|
| 172 |
+
|
| 173 |
+
|
| 174 |
+
def e0m3_weight_gemm_fp16(
|
| 175 |
+
a_packed: torch.Tensor,
|
| 176 |
+
b_packed: torch.Tensor,
|
| 177 |
+
sfa: torch.Tensor,
|
| 178 |
+
sfb: torch.Tensor,
|
| 179 |
+
*,
|
| 180 |
+
alpha: float = 1.0,
|
| 181 |
+
a_format: int = 1,
|
| 182 |
+
out: torch.Tensor | None = None,
|
| 183 |
+
):
|
| 184 |
+
"""SM110 GEMM with E0M3 weights and E2M1 or E0M3 activations."""
|
| 185 |
+
if out is None:
|
| 186 |
+
out = torch.empty(
|
| 187 |
+
(a_packed.shape[0], b_packed.shape[0]),
|
| 188 |
+
device=a_packed.device,
|
| 189 |
+
dtype=torch.float16,
|
| 190 |
+
)
|
| 191 |
+
ops.e0m3_weight_gemm_fp16(
|
| 192 |
+
a_packed, b_packed, sfa, sfb, out, float(alpha), int(a_format)
|
| 193 |
+
)
|
| 194 |
+
return out
|
| 195 |
+
|
| 196 |
+
|
| 197 |
+
def nvfp4_gemm_relu2_nvfp4(
|
| 198 |
+
a_packed: torch.Tensor,
|
| 199 |
+
b_packed: torch.Tensor,
|
| 200 |
+
sfa: torch.Tensor,
|
| 201 |
+
sfb: torch.Tensor,
|
| 202 |
+
*,
|
| 203 |
+
out_packed: torch.Tensor | None = None,
|
| 204 |
+
out_sfa: torch.Tensor | None = None,
|
| 205 |
+
):
|
| 206 |
+
"""SM110 NVFP4 GEMM with fused ReLU-squared and NVFP4 output."""
|
| 207 |
+
if out_packed is None or out_sfa is None:
|
| 208 |
+
out_packed, out_sfa = _alloc_fp4(
|
| 209 |
+
a_packed.shape[0], b_packed.shape[0], a_packed.device
|
| 210 |
+
)
|
| 211 |
+
ops.nvfp4_gemm_relu2_nvfp4(
|
| 212 |
+
a_packed, b_packed, sfa, sfb, out_packed, out_sfa
|
| 213 |
+
)
|
| 214 |
+
return out_packed, out_sfa
|
| 215 |
+
|
| 216 |
+
|
| 217 |
def quantize_fp4_sfa_bf16(
|
| 218 |
x: torch.Tensor,
|
| 219 |
packed: torch.Tensor | None = None,
|
|
|
|
| 528 |
|
| 529 |
__all__ = [
|
| 530 |
"dequantize_fp4_sfa_fp16",
|
| 531 |
+
"e0m3_weight_gemm_fp16",
|
| 532 |
"fp4_w4a16_linear_bf16",
|
| 533 |
"fp4_w4a4_gemv_warpsplit_bf16",
|
| 534 |
"nvfp4_gemm_bf16",
|
|
|
|
| 541 |
"nvfp4_gemm_bias_gelu_nvfp4",
|
| 542 |
"nvfp4_gemm_bias_residual_bf16",
|
| 543 |
"nvfp4_gemm_residual_bf16",
|
| 544 |
+
"nvfp4_gemm_relu2_nvfp4",
|
| 545 |
"nvfp4_gemm_streamk_bf16",
|
| 546 |
"nvfp4_gemm_streamk_bias_bf16",
|
| 547 |
"quantize_fp4_sfa_fp16",
|
| 548 |
+
"quantize_e0m3_sfa_fp16",
|
| 549 |
"quantize_fp4_sfa_bf16",
|
| 550 |
"sfa_size_bytes",
|
| 551 |
]
|
build/torch213-cxx11-cu132-x86_64-linux/{_fp4_gemm_cuda_55c4885.abi3.so → _fp4_gemm_cuda_230e467.abi3.so}
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:e941f703da2103bf557f43020d96a34f7f3f2d18d5bfdf5e37731945e6d6e792
|
| 3 |
+
size 7169088
|
build/torch213-cxx11-cu132-x86_64-linux/_ops.py
CHANGED
|
@@ -1,9 +1,9 @@
|
|
| 1 |
import torch
|
| 2 |
-
from . import
|
| 3 |
-
ops = torch.ops.
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
-
return f"
|
|
|
|
| 1 |
import torch
|
| 2 |
+
from . import _fp4_gemm_cuda_230e467
|
| 3 |
+
ops = torch.ops._fp4_gemm_cuda_230e467
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_fp4_gemm_cuda_230e467::{op_name}"
|
build/torch213-cxx11-cu132-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "fp4-gemm",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
@@ -14,9 +14,9 @@
|
|
| 14 |
"digest": {
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
-
"__init__.py": "
|
| 18 |
-
"
|
| 19 |
-
"_ops.py": "
|
| 20 |
}
|
| 21 |
},
|
| 22 |
"provenance": {
|
|
@@ -26,7 +26,7 @@
|
|
| 26 |
"dirty": false
|
| 27 |
},
|
| 28 |
"kernel": {
|
| 29 |
-
"sha": "
|
| 30 |
"dirty": false
|
| 31 |
}
|
| 32 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"name": "fp4-gemm",
|
| 3 |
+
"id": "_fp4_gemm_cuda_230e467",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 14 |
"digest": {
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
+
"__init__.py": "Qfdvg/9EedWLGh34xAalDfC0IsAHCOL44L4ShaP/w5Y=",
|
| 18 |
+
"_fp4_gemm_cuda_230e467.abi3.so": "6UH3A9ohA79Vf0MCDZajT38/LRjVv99eN3MZRebW55I=",
|
| 19 |
+
"_ops.py": "CxLLKrT37bp1ie5kAVTryKkFcmkRJqJ2fYXrSYuMHcA="
|
| 20 |
}
|
| 21 |
},
|
| 22 |
"provenance": {
|
|
|
|
| 26 |
"dirty": false
|
| 27 |
},
|
| 28 |
"kernel": {
|
| 29 |
+
"sha": "230e467129aa8e5fc687611fbb2aa40a12e09ad1",
|
| 30 |
"dirty": false
|
| 31 |
}
|
| 32 |
}
|