Uploaded using `kernel-builder`.
Browse files- benchmarks/benchmark.py +143 -3
- build/torch211-cxx11-cu128-x86_64-linux/_ops.py +3 -3
- build/torch211-cxx11-cu128-x86_64-linux/{_weight_only_ffn_cuda_b07568a.abi3.so → _weight_only_ffn_cuda_38d37b6.abi3.so} +2 -2
- build/torch211-cxx11-cu128-x86_64-linux/metadata.json +4 -4
- build/torch211-cxx11-cu130-x86_64-linux/_ops.py +3 -3
- build/torch211-cxx11-cu130-x86_64-linux/{_weight_only_ffn_cuda_b07568a.abi3.so → _weight_only_ffn_cuda_38d37b6.abi3.so} +2 -2
- build/torch211-cxx11-cu130-x86_64-linux/metadata.json +5 -4
- build/torch212-cxx11-cu130-x86_64-linux/_ops.py +3 -3
- build/torch212-cxx11-cu130-x86_64-linux/{_weight_only_ffn_cuda_b07568a.abi3.so → _weight_only_ffn_cuda_38d37b6.abi3.so} +2 -2
- build/torch212-cxx11-cu130-x86_64-linux/metadata.json +5 -4
- build/torch212-cxx11-cu132-x86_64-linux/_ops.py +3 -3
- build/torch212-cxx11-cu132-x86_64-linux/{_weight_only_ffn_cuda_b07568a.abi3.so → _weight_only_ffn_cuda_38d37b6.abi3.so} +2 -2
- build/torch212-cxx11-cu132-x86_64-linux/metadata.json +5 -4
- build/torch213-cxx11-cu130-x86_64-linux/_ops.py +3 -3
- build/torch213-cxx11-cu130-x86_64-linux/_weight_only_ffn_cuda_38d37b6.abi3.so +3 -0
- build/torch213-cxx11-cu130-x86_64-linux/_weight_only_ffn_cuda_b07568a.abi3.so +0 -3
- build/torch213-cxx11-cu130-x86_64-linux/metadata.json +5 -4
- build/torch213-cxx11-cu132-x86_64-linux/_ops.py +3 -3
- build/torch213-cxx11-cu132-x86_64-linux/_weight_only_ffn_cuda_38d37b6.abi3.so +3 -0
- build/torch213-cxx11-cu132-x86_64-linux/_weight_only_ffn_cuda_b07568a.abi3.so +0 -3
- build/torch213-cxx11-cu132-x86_64-linux/metadata.json +5 -4
benchmarks/benchmark.py
CHANGED
|
@@ -28,6 +28,17 @@ SHAPES = {
|
|
| 28 |
"vision_m4": (4, 1536, 6144, 1536),
|
| 29 |
}
|
| 30 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 31 |
|
| 32 |
def bench(fn, warmup: int, iterations: int, repeats: int = 3) -> float:
|
| 33 |
for _ in range(warmup):
|
|
@@ -77,6 +88,26 @@ class SourceModule:
|
|
| 77 |
self.ops.dequantize_w8_weight_bf16(packed, scale, out)
|
| 78 |
return out
|
| 79 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 80 |
def _gated(self, bits, x, gu_w, gu_s, dn_w, dn_s, *, gelu,
|
| 81 |
gate_up_bias, down_bias, variant, workspace, out):
|
| 82 |
gu, hidden = workspace
|
|
@@ -132,9 +163,15 @@ def load_source_module():
|
|
| 132 |
root = Path(__file__).resolve().parents[1]
|
| 133 |
registration = root.parent.parent / "kernels" / "kernel-builder" / "src" / "pyproject" / "templates" / "torch"
|
| 134 |
major, minor = torch.cuda.get_device_capability(0)
|
| 135 |
-
if major
|
| 136 |
-
|
| 137 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 138 |
namespace = "weight_only_ffn_benchmark_source"
|
| 139 |
load(
|
| 140 |
name=namespace,
|
|
@@ -260,7 +297,15 @@ def run_case(module, name: str, shape, bits: int, activation: str,
|
|
| 260 |
f"than 5% slower than diagnostic variant {best_diagnostic_variant} "
|
| 261 |
f"at {best_diagnostic_us:.3f} us"
|
| 262 |
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 263 |
return {
|
|
|
|
| 264 |
"shape": name,
|
| 265 |
"M": m,
|
| 266 |
"K": k,
|
|
@@ -285,6 +330,94 @@ def run_case(module, name: str, shape, bits: int, activation: str,
|
|
| 285 |
}
|
| 286 |
|
| 287 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 288 |
def main() -> int:
|
| 289 |
parser = argparse.ArgumentParser()
|
| 290 |
parser.add_argument("--backend", choices=["source", "installed"], default="source")
|
|
@@ -302,6 +435,13 @@ def main() -> int:
|
|
| 302 |
for activation in ("swiglu", "geglu", "gelu"):
|
| 303 |
rows.append(run_case(module, name, SHAPES[name], bits, activation,
|
| 304 |
args.warmup, args.iterations))
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 305 |
payload = {
|
| 306 |
"device": torch.cuda.get_device_name(),
|
| 307 |
"capability": list(torch.cuda.get_device_capability()),
|
|
|
|
| 28 |
"vision_m4": (4, 1536, 6144, 1536),
|
| 29 |
}
|
| 30 |
|
| 31 |
+
LINEAR_SHAPES = {
|
| 32 |
+
"llm_square_m1": (1, 4096, 4096),
|
| 33 |
+
"llm_wide_m1": (1, 4096, 11008),
|
| 34 |
+
"vla_wide_m1": (1, 1024, 4096),
|
| 35 |
+
"vision_wide_m1": (1, 1536, 6144),
|
| 36 |
+
"llm_square_m2": (2, 4096, 4096),
|
| 37 |
+
"llm_wide_m2": (2, 4096, 11008),
|
| 38 |
+
"vla_wide_m2": (2, 1024, 4096),
|
| 39 |
+
"vision_wide_m2": (2, 1536, 6144),
|
| 40 |
+
}
|
| 41 |
+
|
| 42 |
|
| 43 |
def bench(fn, warmup: int, iterations: int, repeats: int = 3) -> float:
|
| 44 |
for _ in range(warmup):
|
|
|
|
| 88 |
self.ops.dequantize_w8_weight_bf16(packed, scale, out)
|
| 89 |
return out
|
| 90 |
|
| 91 |
+
def w4a16_linear_bf16(self, x, weight, scale, *, variant=0, out=None):
|
| 92 |
+
if out is None:
|
| 93 |
+
out = torch.empty(
|
| 94 |
+
(x.shape[0], weight.shape[0]), device=x.device,
|
| 95 |
+
dtype=torch.bfloat16,
|
| 96 |
+
)
|
| 97 |
+
self.ops.w4a16_linear_bf16(
|
| 98 |
+
x, weight, scale, 1.0, variant, out,
|
| 99 |
+
)
|
| 100 |
+
return out
|
| 101 |
+
|
| 102 |
+
def w8a16_linear_bf16(self, x, weight, scale, *, variant=0, out=None):
|
| 103 |
+
if out is None:
|
| 104 |
+
out = torch.empty(
|
| 105 |
+
(x.shape[0], weight.shape[0]), device=x.device,
|
| 106 |
+
dtype=torch.bfloat16,
|
| 107 |
+
)
|
| 108 |
+
self.ops.w8a16_linear_bf16(x, weight, scale, variant, out)
|
| 109 |
+
return out
|
| 110 |
+
|
| 111 |
def _gated(self, bits, x, gu_w, gu_s, dn_w, dn_s, *, gelu,
|
| 112 |
gate_up_bias, down_bias, variant, workspace, out):
|
| 113 |
gu, hidden = workspace
|
|
|
|
| 163 |
root = Path(__file__).resolve().parents[1]
|
| 164 |
registration = root.parent.parent / "kernels" / "kernel-builder" / "src" / "pyproject" / "templates" / "torch"
|
| 165 |
major, minor = torch.cuda.get_device_capability(0)
|
| 166 |
+
if major == 11 and minor == 0:
|
| 167 |
+
arch = "11.0a"
|
| 168 |
+
elif major == 12 and minor == 1:
|
| 169 |
+
arch = "12.1"
|
| 170 |
+
elif major >= 12:
|
| 171 |
+
arch = "12.0a"
|
| 172 |
+
else:
|
| 173 |
+
raise RuntimeError("source benchmark requires Blackwell SM110/SM120/SM121")
|
| 174 |
+
os.environ.setdefault("TORCH_CUDA_ARCH_LIST", arch)
|
| 175 |
namespace = "weight_only_ffn_benchmark_source"
|
| 176 |
load(
|
| 177 |
name=namespace,
|
|
|
|
| 297 |
f"than 5% slower than diagnostic variant {best_diagnostic_variant} "
|
| 298 |
f"at {best_diagnostic_us:.3f} us"
|
| 299 |
)
|
| 300 |
+
if auto_us is not None and auto_us * 1.02 >= min(eager_us, compiled_us):
|
| 301 |
+
raise AssertionError(
|
| 302 |
+
f"{name} W{bits}A16 {activation}: accepted auto path must beat "
|
| 303 |
+
f"the strongest eager/compile baseline by at least 2%; "
|
| 304 |
+
f"auto={auto_us:.3f} us, eager={eager_us:.3f} us, "
|
| 305 |
+
f"compile={compiled_us:.3f} us"
|
| 306 |
+
)
|
| 307 |
return {
|
| 308 |
+
"region": "ffn",
|
| 309 |
"shape": name,
|
| 310 |
"M": m,
|
| 311 |
"K": k,
|
|
|
|
| 330 |
}
|
| 331 |
|
| 332 |
|
| 333 |
+
def run_linear_case(module, name: str, shape, bits: int, warmup: int,
|
| 334 |
+
iterations: int):
|
| 335 |
+
m, k, n = shape
|
| 336 |
+
generator = torch.Generator(device="cuda").manual_seed(
|
| 337 |
+
92000 + m + k + n + bits
|
| 338 |
+
)
|
| 339 |
+
x = (torch.randn((m, k), generator=generator, device="cuda") * 0.1).bfloat16()
|
| 340 |
+
weight = (
|
| 341 |
+
torch.randn((n, k), generator=generator, device="cuda") * 0.02
|
| 342 |
+
).bfloat16()
|
| 343 |
+
packed, scale, dequant = quantize(module, bits, weight)
|
| 344 |
+
out = torch.empty((m, n), device="cuda", dtype=torch.bfloat16)
|
| 345 |
+
linear = getattr(module, f"w{bits}a16_linear_bf16")
|
| 346 |
+
|
| 347 |
+
def kernel(variant: int):
|
| 348 |
+
return linear(x, packed, scale, variant=variant, out=out)
|
| 349 |
+
|
| 350 |
+
def reference():
|
| 351 |
+
return F.linear(x, dequant)
|
| 352 |
+
|
| 353 |
+
eager_us = bench(reference, warmup, iterations)
|
| 354 |
+
compiled = torch.compile(
|
| 355 |
+
reference, fullgraph=True, mode="max-autotune-no-cudagraphs"
|
| 356 |
+
)
|
| 357 |
+
compiled()
|
| 358 |
+
torch.cuda.synchronize()
|
| 359 |
+
compiled_us = bench(compiled, warmup, iterations)
|
| 360 |
+
variants = {
|
| 361 |
+
str(variant): bench(
|
| 362 |
+
lambda variant=variant: kernel(variant), warmup, iterations
|
| 363 |
+
)
|
| 364 |
+
for variant in (1, 2, 3)
|
| 365 |
+
}
|
| 366 |
+
auto_error = None
|
| 367 |
+
try:
|
| 368 |
+
auto_us = bench(lambda: kernel(0), warmup, iterations)
|
| 369 |
+
kernel(0)
|
| 370 |
+
except RuntimeError as exc:
|
| 371 |
+
if "no qualified fast path" not in str(exc):
|
| 372 |
+
raise
|
| 373 |
+
auto_us = None
|
| 374 |
+
auto_error = str(exc)
|
| 375 |
+
kernel(int(min(variants, key=variants.get)))
|
| 376 |
+
ref = reference()
|
| 377 |
+
torch.cuda.synchronize()
|
| 378 |
+
diff = (out.float() - ref.float()).abs().flatten()
|
| 379 |
+
cosine = F.cosine_similarity(
|
| 380 |
+
out.float().flatten(), ref.float().flatten(), dim=0
|
| 381 |
+
)
|
| 382 |
+
best_variant = min(variants, key=variants.get)
|
| 383 |
+
best_us = variants[best_variant]
|
| 384 |
+
if auto_us is not None and auto_us > best_us * 1.05:
|
| 385 |
+
raise AssertionError(
|
| 386 |
+
f"{name} W{bits}A16 linear: auto {auto_us:.3f} us is more than "
|
| 387 |
+
f"5% slower than diagnostic variant {best_variant} at {best_us:.3f} us"
|
| 388 |
+
)
|
| 389 |
+
if auto_us is not None and auto_us * 1.02 >= min(eager_us, compiled_us):
|
| 390 |
+
raise AssertionError(
|
| 391 |
+
f"{name} W{bits}A16 linear: accepted auto path must beat the "
|
| 392 |
+
f"strongest eager/compile baseline by at least 2%; "
|
| 393 |
+
f"auto={auto_us:.3f} us, eager={eager_us:.3f} us, "
|
| 394 |
+
f"compile={compiled_us:.3f} us"
|
| 395 |
+
)
|
| 396 |
+
return {
|
| 397 |
+
"region": "linear",
|
| 398 |
+
"shape": name,
|
| 399 |
+
"M": m,
|
| 400 |
+
"K": k,
|
| 401 |
+
"N": n,
|
| 402 |
+
"precision": f"W{bits}A16",
|
| 403 |
+
"op": "linear",
|
| 404 |
+
"eager_us": eager_us,
|
| 405 |
+
"compile_us": compiled_us,
|
| 406 |
+
"variant_us": variants,
|
| 407 |
+
"auto_status": "accepted" if auto_us is not None else "rejected",
|
| 408 |
+
"auto_us": auto_us,
|
| 409 |
+
"auto_error": auto_error,
|
| 410 |
+
"auto_speedup_vs_eager": eager_us / auto_us if auto_us else None,
|
| 411 |
+
"auto_speedup_vs_compile": compiled_us / auto_us if auto_us else None,
|
| 412 |
+
"best_diagnostic_variant": int(best_variant),
|
| 413 |
+
"best_diagnostic_us": best_us,
|
| 414 |
+
"max_abs": float(diff.max()),
|
| 415 |
+
"mean_abs": float(diff.mean()),
|
| 416 |
+
"p99_abs": float(torch.quantile(diff, 0.99)),
|
| 417 |
+
"cosine": float(cosine),
|
| 418 |
+
}
|
| 419 |
+
|
| 420 |
+
|
| 421 |
def main() -> int:
|
| 422 |
parser = argparse.ArgumentParser()
|
| 423 |
parser.add_argument("--backend", choices=["source", "installed"], default="source")
|
|
|
|
| 435 |
for activation in ("swiglu", "geglu", "gelu"):
|
| 436 |
rows.append(run_case(module, name, SHAPES[name], bits, activation,
|
| 437 |
args.warmup, args.iterations))
|
| 438 |
+
linear_names = ["llm_square_m1"] if args.mode == "smoke" else list(LINEAR_SHAPES)
|
| 439 |
+
for name in linear_names:
|
| 440 |
+
for bits in (4, 8):
|
| 441 |
+
rows.append(run_linear_case(
|
| 442 |
+
module, name, LINEAR_SHAPES[name], bits,
|
| 443 |
+
args.warmup, args.iterations,
|
| 444 |
+
))
|
| 445 |
payload = {
|
| 446 |
"device": torch.cuda.get_device_name(),
|
| 447 |
"capability": list(torch.cuda.get_device_capability()),
|
build/torch211-cxx11-cu128-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 _weight_only_ffn_cuda_38d37b6
|
| 3 |
+
ops = torch.ops._weight_only_ffn_cuda_38d37b6
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_weight_only_ffn_cuda_38d37b6::{op_name}"
|
build/torch211-cxx11-cu128-x86_64-linux/{_weight_only_ffn_cuda_b07568a.abi3.so → _weight_only_ffn_cuda_38d37b6.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 sha256:7ee391e51802ca24eda6c3c73cb1b711e61864dbe9272b9f6b79cce491deaffe
|
| 3 |
+
size 767928
|
build/torch211-cxx11-cu128-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "weight-only-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
@@ -14,8 +14,8 @@
|
|
| 14 |
"algorithm": "sha256",
|
| 15 |
"files": {
|
| 16 |
"__init__.py": "0q6TMj65una1LdyHLC291Dwy2VzrBbAD8Wm8MEdmqwA=",
|
| 17 |
-
"_ops.py": "
|
| 18 |
-
"
|
| 19 |
"weight_only_ffn/__init__.py": "DFYPlrhXwYjEqCl/8n0SmWGZV8NFml5DPhMjKfv98GY="
|
| 20 |
}
|
| 21 |
},
|
|
@@ -26,7 +26,7 @@
|
|
| 26 |
"dirty": false
|
| 27 |
},
|
| 28 |
"kernel": {
|
| 29 |
-
"sha": "
|
| 30 |
"dirty": false
|
| 31 |
}
|
| 32 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"name": "weight-only-ffn",
|
| 3 |
+
"id": "_weight_only_ffn_cuda_38d37b6",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 14 |
"algorithm": "sha256",
|
| 15 |
"files": {
|
| 16 |
"__init__.py": "0q6TMj65una1LdyHLC291Dwy2VzrBbAD8Wm8MEdmqwA=",
|
| 17 |
+
"_ops.py": "iqY+HoE4nGHqvVlEZTwEAcyu1/w+QrjwuuW9sdTI2uk=",
|
| 18 |
+
"_weight_only_ffn_cuda_38d37b6.abi3.so": "fuOR5RgCyiTtpsPHPLG3EeYYZNvpJyufa3nM5JHer/4=",
|
| 19 |
"weight_only_ffn/__init__.py": "DFYPlrhXwYjEqCl/8n0SmWGZV8NFml5DPhMjKfv98GY="
|
| 20 |
}
|
| 21 |
},
|
|
|
|
| 26 |
"dirty": false
|
| 27 |
},
|
| 28 |
"kernel": {
|
| 29 |
+
"sha": "38d37b67acfe940c99d7b3db721fa67163b82cec",
|
| 30 |
"dirty": false
|
| 31 |
}
|
| 32 |
}
|
build/torch211-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 _weight_only_ffn_cuda_38d37b6
|
| 3 |
+
ops = torch.ops._weight_only_ffn_cuda_38d37b6
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_weight_only_ffn_cuda_38d37b6::{op_name}"
|
build/torch211-cxx11-cu130-x86_64-linux/{_weight_only_ffn_cuda_b07568a.abi3.so → _weight_only_ffn_cuda_38d37b6.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 sha256:6d98010a7e6bbe8b925689ee60f1026519835d2d841d6f1a58fb7922714d274d
|
| 3 |
+
size 1896120
|
build/torch211-cxx11-cu130-x86_64-linux/metadata.json
CHANGED
|
@@ -1,12 +1,13 @@
|
|
| 1 |
{
|
| 2 |
"name": "weight-only-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
| 7 |
"backend": {
|
| 8 |
"type": "cuda",
|
| 9 |
"archs": [
|
|
|
|
| 10 |
"12.0a",
|
| 11 |
"12.1"
|
| 12 |
]
|
|
@@ -15,8 +16,8 @@
|
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
"__init__.py": "0q6TMj65una1LdyHLC291Dwy2VzrBbAD8Wm8MEdmqwA=",
|
| 18 |
-
"_ops.py": "
|
| 19 |
-
"
|
| 20 |
"weight_only_ffn/__init__.py": "DFYPlrhXwYjEqCl/8n0SmWGZV8NFml5DPhMjKfv98GY="
|
| 21 |
}
|
| 22 |
},
|
|
@@ -27,7 +28,7 @@
|
|
| 27 |
"dirty": false
|
| 28 |
},
|
| 29 |
"kernel": {
|
| 30 |
-
"sha": "
|
| 31 |
"dirty": false
|
| 32 |
}
|
| 33 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"name": "weight-only-ffn",
|
| 3 |
+
"id": "_weight_only_ffn_cuda_38d37b6",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
| 7 |
"backend": {
|
| 8 |
"type": "cuda",
|
| 9 |
"archs": [
|
| 10 |
+
"11.0a",
|
| 11 |
"12.0a",
|
| 12 |
"12.1"
|
| 13 |
]
|
|
|
|
| 16 |
"algorithm": "sha256",
|
| 17 |
"files": {
|
| 18 |
"__init__.py": "0q6TMj65una1LdyHLC291Dwy2VzrBbAD8Wm8MEdmqwA=",
|
| 19 |
+
"_ops.py": "iqY+HoE4nGHqvVlEZTwEAcyu1/w+QrjwuuW9sdTI2uk=",
|
| 20 |
+
"_weight_only_ffn_cuda_38d37b6.abi3.so": "bZgBCn5rvouSVonuYPECZRmDXS2EHW8aWPt5InFNJ00=",
|
| 21 |
"weight_only_ffn/__init__.py": "DFYPlrhXwYjEqCl/8n0SmWGZV8NFml5DPhMjKfv98GY="
|
| 22 |
}
|
| 23 |
},
|
|
|
|
| 28 |
"dirty": false
|
| 29 |
},
|
| 30 |
"kernel": {
|
| 31 |
+
"sha": "38d37b67acfe940c99d7b3db721fa67163b82cec",
|
| 32 |
"dirty": false
|
| 33 |
}
|
| 34 |
}
|
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 _weight_only_ffn_cuda_38d37b6
|
| 3 |
+
ops = torch.ops._weight_only_ffn_cuda_38d37b6
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_weight_only_ffn_cuda_38d37b6::{op_name}"
|
build/torch212-cxx11-cu130-x86_64-linux/{_weight_only_ffn_cuda_b07568a.abi3.so → _weight_only_ffn_cuda_38d37b6.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:ebcf5ff2f3270a0327712bb8c917a6aa036140f7a2c065849dd897a58898261b
|
| 3 |
+
size 1906560
|
build/torch212-cxx11-cu130-x86_64-linux/metadata.json
CHANGED
|
@@ -1,12 +1,13 @@
|
|
| 1 |
{
|
| 2 |
"name": "weight-only-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
| 7 |
"backend": {
|
| 8 |
"type": "cuda",
|
| 9 |
"archs": [
|
|
|
|
| 10 |
"12.0a",
|
| 11 |
"12.1"
|
| 12 |
]
|
|
@@ -15,8 +16,8 @@
|
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
"__init__.py": "0q6TMj65una1LdyHLC291Dwy2VzrBbAD8Wm8MEdmqwA=",
|
| 18 |
-
"_ops.py": "
|
| 19 |
-
"
|
| 20 |
"weight_only_ffn/__init__.py": "DFYPlrhXwYjEqCl/8n0SmWGZV8NFml5DPhMjKfv98GY="
|
| 21 |
}
|
| 22 |
},
|
|
@@ -27,7 +28,7 @@
|
|
| 27 |
"dirty": false
|
| 28 |
},
|
| 29 |
"kernel": {
|
| 30 |
-
"sha": "
|
| 31 |
"dirty": false
|
| 32 |
}
|
| 33 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"name": "weight-only-ffn",
|
| 3 |
+
"id": "_weight_only_ffn_cuda_38d37b6",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
| 7 |
"backend": {
|
| 8 |
"type": "cuda",
|
| 9 |
"archs": [
|
| 10 |
+
"11.0a",
|
| 11 |
"12.0a",
|
| 12 |
"12.1"
|
| 13 |
]
|
|
|
|
| 16 |
"algorithm": "sha256",
|
| 17 |
"files": {
|
| 18 |
"__init__.py": "0q6TMj65una1LdyHLC291Dwy2VzrBbAD8Wm8MEdmqwA=",
|
| 19 |
+
"_ops.py": "iqY+HoE4nGHqvVlEZTwEAcyu1/w+QrjwuuW9sdTI2uk=",
|
| 20 |
+
"_weight_only_ffn_cuda_38d37b6.abi3.so": "689f8vMnCgMncSu4yRemqgNhQPeiwGWEndiXpYiYJhs=",
|
| 21 |
"weight_only_ffn/__init__.py": "DFYPlrhXwYjEqCl/8n0SmWGZV8NFml5DPhMjKfv98GY="
|
| 22 |
}
|
| 23 |
},
|
|
|
|
| 28 |
"dirty": false
|
| 29 |
},
|
| 30 |
"kernel": {
|
| 31 |
+
"sha": "38d37b67acfe940c99d7b3db721fa67163b82cec",
|
| 32 |
"dirty": false
|
| 33 |
}
|
| 34 |
}
|
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 _weight_only_ffn_cuda_38d37b6
|
| 3 |
+
ops = torch.ops._weight_only_ffn_cuda_38d37b6
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_weight_only_ffn_cuda_38d37b6::{op_name}"
|
build/torch212-cxx11-cu132-x86_64-linux/{_weight_only_ffn_cuda_b07568a.abi3.so → _weight_only_ffn_cuda_38d37b6.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 sha256:8ef5420caa2f23097e3e8f9777250c9bbb98618fe76493d492e5cb5a5f09336c
|
| 3 |
+
size 1906560
|
build/torch212-cxx11-cu132-x86_64-linux/metadata.json
CHANGED
|
@@ -1,12 +1,13 @@
|
|
| 1 |
{
|
| 2 |
"name": "weight-only-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
| 7 |
"backend": {
|
| 8 |
"type": "cuda",
|
| 9 |
"archs": [
|
|
|
|
| 10 |
"12.0a",
|
| 11 |
"12.1"
|
| 12 |
]
|
|
@@ -15,8 +16,8 @@
|
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
"__init__.py": "0q6TMj65una1LdyHLC291Dwy2VzrBbAD8Wm8MEdmqwA=",
|
| 18 |
-
"_ops.py": "
|
| 19 |
-
"
|
| 20 |
"weight_only_ffn/__init__.py": "DFYPlrhXwYjEqCl/8n0SmWGZV8NFml5DPhMjKfv98GY="
|
| 21 |
}
|
| 22 |
},
|
|
@@ -27,7 +28,7 @@
|
|
| 27 |
"dirty": false
|
| 28 |
},
|
| 29 |
"kernel": {
|
| 30 |
-
"sha": "
|
| 31 |
"dirty": false
|
| 32 |
}
|
| 33 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"name": "weight-only-ffn",
|
| 3 |
+
"id": "_weight_only_ffn_cuda_38d37b6",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
| 7 |
"backend": {
|
| 8 |
"type": "cuda",
|
| 9 |
"archs": [
|
| 10 |
+
"11.0a",
|
| 11 |
"12.0a",
|
| 12 |
"12.1"
|
| 13 |
]
|
|
|
|
| 16 |
"algorithm": "sha256",
|
| 17 |
"files": {
|
| 18 |
"__init__.py": "0q6TMj65una1LdyHLC291Dwy2VzrBbAD8Wm8MEdmqwA=",
|
| 19 |
+
"_ops.py": "iqY+HoE4nGHqvVlEZTwEAcyu1/w+QrjwuuW9sdTI2uk=",
|
| 20 |
+
"_weight_only_ffn_cuda_38d37b6.abi3.so": "jvVCDKovIwl+Po+XdyUMm7uYYY/nZJPUkuXLWl8JM2w=",
|
| 21 |
"weight_only_ffn/__init__.py": "DFYPlrhXwYjEqCl/8n0SmWGZV8NFml5DPhMjKfv98GY="
|
| 22 |
}
|
| 23 |
},
|
|
|
|
| 28 |
"dirty": false
|
| 29 |
},
|
| 30 |
"kernel": {
|
| 31 |
+
"sha": "38d37b67acfe940c99d7b3db721fa67163b82cec",
|
| 32 |
"dirty": false
|
| 33 |
}
|
| 34 |
}
|
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 _weight_only_ffn_cuda_38d37b6
|
| 3 |
+
ops = torch.ops._weight_only_ffn_cuda_38d37b6
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_weight_only_ffn_cuda_38d37b6::{op_name}"
|
build/torch213-cxx11-cu130-x86_64-linux/_weight_only_ffn_cuda_38d37b6.abi3.so
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2c2083db3ce3c5ad71f964444503f9ca653986e9e878d20ea08572f4c2c7eec9
|
| 3 |
+
size 1906400
|
build/torch213-cxx11-cu130-x86_64-linux/_weight_only_ffn_cuda_b07568a.abi3.so
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:38e91d7f9250b503c947bf3d0f911f9e584547ab020ec50138f43afaf34cc90d
|
| 3 |
-
size 1391928
|
|
|
|
|
|
|
|
|
|
|
|
build/torch213-cxx11-cu130-x86_64-linux/metadata.json
CHANGED
|
@@ -1,12 +1,13 @@
|
|
| 1 |
{
|
| 2 |
"name": "weight-only-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
| 7 |
"backend": {
|
| 8 |
"type": "cuda",
|
| 9 |
"archs": [
|
|
|
|
| 10 |
"12.0a",
|
| 11 |
"12.1"
|
| 12 |
]
|
|
@@ -15,8 +16,8 @@
|
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
"__init__.py": "0q6TMj65una1LdyHLC291Dwy2VzrBbAD8Wm8MEdmqwA=",
|
| 18 |
-
"_ops.py": "
|
| 19 |
-
"
|
| 20 |
"weight_only_ffn/__init__.py": "DFYPlrhXwYjEqCl/8n0SmWGZV8NFml5DPhMjKfv98GY="
|
| 21 |
}
|
| 22 |
},
|
|
@@ -27,7 +28,7 @@
|
|
| 27 |
"dirty": false
|
| 28 |
},
|
| 29 |
"kernel": {
|
| 30 |
-
"sha": "
|
| 31 |
"dirty": false
|
| 32 |
}
|
| 33 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"name": "weight-only-ffn",
|
| 3 |
+
"id": "_weight_only_ffn_cuda_38d37b6",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
| 7 |
"backend": {
|
| 8 |
"type": "cuda",
|
| 9 |
"archs": [
|
| 10 |
+
"11.0a",
|
| 11 |
"12.0a",
|
| 12 |
"12.1"
|
| 13 |
]
|
|
|
|
| 16 |
"algorithm": "sha256",
|
| 17 |
"files": {
|
| 18 |
"__init__.py": "0q6TMj65una1LdyHLC291Dwy2VzrBbAD8Wm8MEdmqwA=",
|
| 19 |
+
"_ops.py": "iqY+HoE4nGHqvVlEZTwEAcyu1/w+QrjwuuW9sdTI2uk=",
|
| 20 |
+
"_weight_only_ffn_cuda_38d37b6.abi3.so": "LCCD2zzjxa1x+WRERQP5ymU5hunoeNIOoIVy9MLH7sk=",
|
| 21 |
"weight_only_ffn/__init__.py": "DFYPlrhXwYjEqCl/8n0SmWGZV8NFml5DPhMjKfv98GY="
|
| 22 |
}
|
| 23 |
},
|
|
|
|
| 28 |
"dirty": false
|
| 29 |
},
|
| 30 |
"kernel": {
|
| 31 |
+
"sha": "38d37b67acfe940c99d7b3db721fa67163b82cec",
|
| 32 |
"dirty": false
|
| 33 |
}
|
| 34 |
}
|
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 _weight_only_ffn_cuda_38d37b6
|
| 3 |
+
ops = torch.ops._weight_only_ffn_cuda_38d37b6
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_weight_only_ffn_cuda_38d37b6::{op_name}"
|
build/torch213-cxx11-cu132-x86_64-linux/_weight_only_ffn_cuda_38d37b6.abi3.so
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:93ff16c4abd9057f9166fb2dd0b765053ea8d4bd69ddcb514b4fc3e6de56bb10
|
| 3 |
+
size 1906400
|
build/torch213-cxx11-cu132-x86_64-linux/_weight_only_ffn_cuda_b07568a.abi3.so
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:09cd6d00eef5a01d237971204f9696ac011d1313cff15a00c99f0c158ee1f53c
|
| 3 |
-
size 1391928
|
|
|
|
|
|
|
|
|
|
|
|
build/torch213-cxx11-cu132-x86_64-linux/metadata.json
CHANGED
|
@@ -1,12 +1,13 @@
|
|
| 1 |
{
|
| 2 |
"name": "weight-only-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
| 7 |
"backend": {
|
| 8 |
"type": "cuda",
|
| 9 |
"archs": [
|
|
|
|
| 10 |
"12.0a",
|
| 11 |
"12.1"
|
| 12 |
]
|
|
@@ -15,8 +16,8 @@
|
|
| 15 |
"algorithm": "sha256",
|
| 16 |
"files": {
|
| 17 |
"__init__.py": "0q6TMj65una1LdyHLC291Dwy2VzrBbAD8Wm8MEdmqwA=",
|
| 18 |
-
"_ops.py": "
|
| 19 |
-
"
|
| 20 |
"weight_only_ffn/__init__.py": "DFYPlrhXwYjEqCl/8n0SmWGZV8NFml5DPhMjKfv98GY="
|
| 21 |
}
|
| 22 |
},
|
|
@@ -27,7 +28,7 @@
|
|
| 27 |
"dirty": false
|
| 28 |
},
|
| 29 |
"kernel": {
|
| 30 |
-
"sha": "
|
| 31 |
"dirty": false
|
| 32 |
}
|
| 33 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"name": "weight-only-ffn",
|
| 3 |
+
"id": "_weight_only_ffn_cuda_38d37b6",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
| 7 |
"backend": {
|
| 8 |
"type": "cuda",
|
| 9 |
"archs": [
|
| 10 |
+
"11.0a",
|
| 11 |
"12.0a",
|
| 12 |
"12.1"
|
| 13 |
]
|
|
|
|
| 16 |
"algorithm": "sha256",
|
| 17 |
"files": {
|
| 18 |
"__init__.py": "0q6TMj65una1LdyHLC291Dwy2VzrBbAD8Wm8MEdmqwA=",
|
| 19 |
+
"_ops.py": "iqY+HoE4nGHqvVlEZTwEAcyu1/w+QrjwuuW9sdTI2uk=",
|
| 20 |
+
"_weight_only_ffn_cuda_38d37b6.abi3.so": "k/8WxKvZBX+RZvst0LdlBT6o1L1p3ctRS0/D5t5WuxA=",
|
| 21 |
"weight_only_ffn/__init__.py": "DFYPlrhXwYjEqCl/8n0SmWGZV8NFml5DPhMjKfv98GY="
|
| 22 |
}
|
| 23 |
},
|
|
|
|
| 28 |
"dirty": false
|
| 29 |
},
|
| 30 |
"kernel": {
|
| 31 |
+
"sha": "38d37b67acfe940c99d7b3db721fa67163b82cec",
|
| 32 |
"dirty": false
|
| 33 |
}
|
| 34 |
}
|