Uploaded using `kernel-builder`.
Browse files- benchmarks/benchmark.py +5 -5
- benchmarks/benchmark_bf16_entry.py +31 -16
- build/torch211-cxx11-cu128-x86_64-linux/{_flashrt_fp8_ffn_cuda_6ec89eb.abi3.so → _flashrt_fp8_ffn_cuda_f1d4adb.abi3.so} +2 -2
- build/torch211-cxx11-cu128-x86_64-linux/_ops.py +3 -3
- build/torch211-cxx11-cu128-x86_64-linux/metadata.json +1 -1
- build/torch211-cxx11-cu130-x86_64-linux/{_flashrt_fp8_ffn_cuda_6ec89eb.abi3.so → _flashrt_fp8_ffn_cuda_f1d4adb.abi3.so} +2 -2
- build/torch211-cxx11-cu130-x86_64-linux/_ops.py +3 -3
- build/torch211-cxx11-cu130-x86_64-linux/metadata.json +1 -1
- build/torch211-cxx11-rocm71-x86_64-linux/{_flashrt_fp8_ffn_rocm_6ec89eb.abi3.so → _flashrt_fp8_ffn_rocm_f1d4adb.abi3.so} +1 -1
- build/torch211-cxx11-rocm71-x86_64-linux/_ops.py +3 -3
- build/torch211-cxx11-rocm71-x86_64-linux/metadata.json +1 -1
- build/torch211-cxx11-rocm72-x86_64-linux/{_flashrt_fp8_ffn_rocm_6ec89eb.abi3.so → _flashrt_fp8_ffn_rocm_f1d4adb.abi3.so} +1 -1
- build/torch211-cxx11-rocm72-x86_64-linux/_ops.py +3 -3
- build/torch211-cxx11-rocm72-x86_64-linux/metadata.json +1 -1
- build/torch212-cxx11-cu130-x86_64-linux/_flashrt_fp8_ffn_cuda_6ec89eb.abi3.so +0 -3
- build/torch212-cxx11-cu130-x86_64-linux/_flashrt_fp8_ffn_cuda_f1d4adb.abi3.so +3 -0
- build/torch212-cxx11-cu130-x86_64-linux/_ops.py +3 -3
- build/torch212-cxx11-cu130-x86_64-linux/metadata.json +1 -1
- build/torch212-cxx11-cu132-x86_64-linux/_flashrt_fp8_ffn_cuda_6ec89eb.abi3.so +0 -3
- build/torch212-cxx11-cu132-x86_64-linux/_flashrt_fp8_ffn_cuda_f1d4adb.abi3.so +3 -0
- build/torch212-cxx11-cu132-x86_64-linux/_ops.py +3 -3
- build/torch212-cxx11-cu132-x86_64-linux/metadata.json +1 -1
- build/torch212-cxx11-rocm71-x86_64-linux/{_flashrt_fp8_ffn_rocm_6ec89eb.abi3.so → _flashrt_fp8_ffn_rocm_f1d4adb.abi3.so} +1 -1
- build/torch212-cxx11-rocm71-x86_64-linux/_ops.py +3 -3
- build/torch212-cxx11-rocm71-x86_64-linux/metadata.json +1 -1
- build/torch212-cxx11-rocm72-x86_64-linux/{_flashrt_fp8_ffn_rocm_6ec89eb.abi3.so → _flashrt_fp8_ffn_rocm_f1d4adb.abi3.so} +1 -1
- build/torch212-cxx11-rocm72-x86_64-linux/_ops.py +3 -3
- build/torch212-cxx11-rocm72-x86_64-linux/metadata.json +1 -1
benchmarks/benchmark.py
CHANGED
|
@@ -126,14 +126,14 @@ class SourceOps:
|
|
| 126 |
up_w_scale,
|
| 127 |
hidden_scale,
|
| 128 |
down_w_scale,
|
| 129 |
-
|
| 130 |
hidden_fp8=None,
|
| 131 |
out=None,
|
| 132 |
):
|
| 133 |
-
if
|
| 134 |
-
|
| 135 |
if hidden_fp8 is None:
|
| 136 |
-
hidden_fp8 = torch.empty_like(
|
| 137 |
if out is None:
|
| 138 |
out = torch.empty((x.shape[0], down_w.shape[0]), device=x.device, dtype=torch.bfloat16)
|
| 139 |
self._ops.fp8_gelu_mlp_bf16(
|
|
@@ -146,7 +146,7 @@ class SourceOps:
|
|
| 146 |
up_w_scale,
|
| 147 |
hidden_scale,
|
| 148 |
down_w_scale,
|
| 149 |
-
|
| 150 |
hidden_fp8,
|
| 151 |
out,
|
| 152 |
)
|
|
|
|
| 126 |
up_w_scale,
|
| 127 |
hidden_scale,
|
| 128 |
down_w_scale,
|
| 129 |
+
hidden_bf16=None,
|
| 130 |
hidden_fp8=None,
|
| 131 |
out=None,
|
| 132 |
):
|
| 133 |
+
if hidden_bf16 is None:
|
| 134 |
+
hidden_bf16 = torch.empty((x.shape[0], up_w.shape[0]), device=x.device, dtype=torch.bfloat16)
|
| 135 |
if hidden_fp8 is None:
|
| 136 |
+
hidden_fp8 = torch.empty_like(hidden_bf16, dtype=fp8_dtype())
|
| 137 |
if out is None:
|
| 138 |
out = torch.empty((x.shape[0], down_w.shape[0]), device=x.device, dtype=torch.bfloat16)
|
| 139 |
self._ops.fp8_gelu_mlp_bf16(
|
|
|
|
| 146 |
up_w_scale,
|
| 147 |
hidden_scale,
|
| 148 |
down_w_scale,
|
| 149 |
+
hidden_bf16,
|
| 150 |
hidden_fp8,
|
| 151 |
out,
|
| 152 |
)
|
benchmarks/benchmark_bf16_entry.py
CHANGED
|
@@ -5,6 +5,7 @@ from __future__ import annotations
|
|
| 5 |
|
| 6 |
import argparse
|
| 7 |
import json
|
|
|
|
| 8 |
from dataclasses import asdict, dataclass
|
| 9 |
from pathlib import Path
|
| 10 |
|
|
@@ -96,6 +97,24 @@ def midm_padded_rows(rows: int) -> int:
|
|
| 96 |
return rows
|
| 97 |
|
| 98 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 99 |
def make_case(M: int, K: int, H: int, N: int):
|
| 100 |
x = torch.randn((M, K), device="cuda", dtype=torch.bfloat16) * 0.25
|
| 101 |
up = torch.randn((H, K), device="cuda", dtype=torch.bfloat16) * (K**-0.5)
|
|
@@ -194,7 +213,7 @@ def run_shape(ops, name: str, shape, args) -> Result:
|
|
| 194 |
up_scale,
|
| 195 |
hidden_scale,
|
| 196 |
down_scale,
|
| 197 |
-
|
| 198 |
hidden_fp8=staged_hidden_fp8,
|
| 199 |
out=staged_out,
|
| 200 |
)
|
|
@@ -213,7 +232,7 @@ def run_shape(ops, name: str, shape, args) -> Result:
|
|
| 213 |
up_scale,
|
| 214 |
hidden_scale,
|
| 215 |
down_scale,
|
| 216 |
-
|
| 217 |
hidden_fp8=exact_hidden_fp8,
|
| 218 |
out=exact_out,
|
| 219 |
)[:M]
|
|
@@ -234,22 +253,17 @@ def run_shape(ops, name: str, shape, args) -> Result:
|
|
| 234 |
bf16_metrics = metrics(got, bf16_expected)
|
| 235 |
staged_compatible = quant_exact and staged_metrics["max_abs"] == 0.0
|
| 236 |
|
| 237 |
-
flashrt_us
|
|
|
|
|
|
|
| 238 |
graph = torch.cuda.CUDAGraph()
|
| 239 |
flashrt_call()
|
| 240 |
torch.cuda.synchronize()
|
| 241 |
with torch.cuda.graph(graph):
|
| 242 |
flashrt_call()
|
| 243 |
-
graph_us =
|
| 244 |
-
separate_us =
|
| 245 |
-
|
| 246 |
-
)
|
| 247 |
-
kernel_us = base.time_us(
|
| 248 |
-
staged_call, warmup=args.warmup, iters=args.iters
|
| 249 |
-
)
|
| 250 |
-
eager_us = base.time_us(
|
| 251 |
-
torch_bf16_reference, warmup=args.warmup, iters=args.iters
|
| 252 |
-
)
|
| 253 |
|
| 254 |
compile_us = None
|
| 255 |
compile_status = "not_requested"
|
|
@@ -268,9 +282,7 @@ def run_shape(ops, name: str, shape, args) -> Result:
|
|
| 268 |
f"{compiled_metrics['cosine']:.8f}"
|
| 269 |
)
|
| 270 |
else:
|
| 271 |
-
compile_us =
|
| 272 |
-
compiled, warmup=args.warmup, iters=args.iters
|
| 273 |
-
)
|
| 274 |
compile_status = "fullgraph-ok"
|
| 275 |
except Exception as exc: # noqa: BLE001
|
| 276 |
compile_status = f"failed: {type(exc).__name__}: {exc}"
|
|
@@ -337,6 +349,7 @@ def main() -> None:
|
|
| 337 |
parser.add_argument("--shapes", default="all")
|
| 338 |
parser.add_argument("--warmup", type=int, default=20)
|
| 339 |
parser.add_argument("--iters", type=int, default=100)
|
|
|
|
| 340 |
parser.add_argument("--compile-baseline", action="store_true")
|
| 341 |
parser.add_argument("--output", type=Path)
|
| 342 |
args = parser.parse_args()
|
|
@@ -385,6 +398,8 @@ def main() -> None:
|
|
| 385 |
"torch": torch.__version__,
|
| 386 |
"warmup": args.warmup,
|
| 387 |
"iters": args.iters,
|
|
|
|
|
|
|
| 388 |
"results": [asdict(result) for result in results],
|
| 389 |
}
|
| 390 |
if args.output:
|
|
|
|
| 5 |
|
| 6 |
import argparse
|
| 7 |
import json
|
| 8 |
+
import statistics
|
| 9 |
from dataclasses import asdict, dataclass
|
| 10 |
from pathlib import Path
|
| 11 |
|
|
|
|
| 97 |
return rows
|
| 98 |
|
| 99 |
|
| 100 |
+
def median_time_us(fn, args) -> float:
|
| 101 |
+
return statistics.median(
|
| 102 |
+
base.time_us(fn, warmup=args.warmup, iters=args.iters)
|
| 103 |
+
for _ in range(args.rounds)
|
| 104 |
+
)
|
| 105 |
+
|
| 106 |
+
|
| 107 |
+
def abba_time_us(a, b, args) -> tuple[float, float]:
|
| 108 |
+
a_samples = []
|
| 109 |
+
b_samples = []
|
| 110 |
+
for _ in range(args.rounds):
|
| 111 |
+
a_samples.append(base.time_us(a, warmup=args.warmup, iters=args.iters))
|
| 112 |
+
b_samples.append(base.time_us(b, warmup=args.warmup, iters=args.iters))
|
| 113 |
+
b_samples.append(base.time_us(b, warmup=args.warmup, iters=args.iters))
|
| 114 |
+
a_samples.append(base.time_us(a, warmup=args.warmup, iters=args.iters))
|
| 115 |
+
return statistics.median(a_samples), statistics.median(b_samples)
|
| 116 |
+
|
| 117 |
+
|
| 118 |
def make_case(M: int, K: int, H: int, N: int):
|
| 119 |
x = torch.randn((M, K), device="cuda", dtype=torch.bfloat16) * 0.25
|
| 120 |
up = torch.randn((H, K), device="cuda", dtype=torch.bfloat16) * (K**-0.5)
|
|
|
|
| 213 |
up_scale,
|
| 214 |
hidden_scale,
|
| 215 |
down_scale,
|
| 216 |
+
hidden_bf16=staged_hidden,
|
| 217 |
hidden_fp8=staged_hidden_fp8,
|
| 218 |
out=staged_out,
|
| 219 |
)
|
|
|
|
| 232 |
up_scale,
|
| 233 |
hidden_scale,
|
| 234 |
down_scale,
|
| 235 |
+
hidden_bf16=exact_hidden,
|
| 236 |
hidden_fp8=exact_hidden_fp8,
|
| 237 |
out=exact_out,
|
| 238 |
)[:M]
|
|
|
|
| 253 |
bf16_metrics = metrics(got, bf16_expected)
|
| 254 |
staged_compatible = quant_exact and staged_metrics["max_abs"] == 0.0
|
| 255 |
|
| 256 |
+
flashrt_us, eager_us = abba_time_us(
|
| 257 |
+
flashrt_call, torch_bf16_reference, args
|
| 258 |
+
)
|
| 259 |
graph = torch.cuda.CUDAGraph()
|
| 260 |
flashrt_call()
|
| 261 |
torch.cuda.synchronize()
|
| 262 |
with torch.cuda.graph(graph):
|
| 263 |
flashrt_call()
|
| 264 |
+
graph_us = median_time_us(graph.replay, args)
|
| 265 |
+
separate_us = median_time_us(separate_quant_call, args)
|
| 266 |
+
kernel_us = median_time_us(staged_call, args)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 267 |
|
| 268 |
compile_us = None
|
| 269 |
compile_status = "not_requested"
|
|
|
|
| 282 |
f"{compiled_metrics['cosine']:.8f}"
|
| 283 |
)
|
| 284 |
else:
|
| 285 |
+
compile_us = median_time_us(compiled, args)
|
|
|
|
|
|
|
| 286 |
compile_status = "fullgraph-ok"
|
| 287 |
except Exception as exc: # noqa: BLE001
|
| 288 |
compile_status = f"failed: {type(exc).__name__}: {exc}"
|
|
|
|
| 349 |
parser.add_argument("--shapes", default="all")
|
| 350 |
parser.add_argument("--warmup", type=int, default=20)
|
| 351 |
parser.add_argument("--iters", type=int, default=100)
|
| 352 |
+
parser.add_argument("--rounds", type=int, default=5)
|
| 353 |
parser.add_argument("--compile-baseline", action="store_true")
|
| 354 |
parser.add_argument("--output", type=Path)
|
| 355 |
args = parser.parse_args()
|
|
|
|
| 398 |
"torch": torch.__version__,
|
| 399 |
"warmup": args.warmup,
|
| 400 |
"iters": args.iters,
|
| 401 |
+
"rounds": args.rounds,
|
| 402 |
+
"primary_order": "A-B-B-A median",
|
| 403 |
"results": [asdict(result) for result in results],
|
| 404 |
}
|
| 405 |
if args.output:
|
build/torch211-cxx11-cu128-x86_64-linux/{_flashrt_fp8_ffn_cuda_6ec89eb.abi3.so → _flashrt_fp8_ffn_cuda_f1d4adb.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 sha256:6902410bce97b0fb3a232f91473ef33cda9967597f090753cfeb2bd361490c9e
|
| 3 |
+
size 677320
|
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 _flashrt_fp8_ffn_cuda_f1d4adb
|
| 3 |
+
ops = torch.ops._flashrt_fp8_ffn_cuda_f1d4adb
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_flashrt_fp8_ffn_cuda_f1d4adb::{op_name}"
|
build/torch211-cxx11-cu128-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
+
"id": "_flashrt_fp8_ffn_cuda_f1d4adb",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
build/torch211-cxx11-cu130-x86_64-linux/{_flashrt_fp8_ffn_cuda_6ec89eb.abi3.so → _flashrt_fp8_ffn_cuda_f1d4adb.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 sha256:98d0cc9e54be58c795341ae8c7a7f8e375a6feb208b119df2ea8fe639023d4b3
|
| 3 |
+
size 647000
|
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 _flashrt_fp8_ffn_cuda_f1d4adb
|
| 3 |
+
ops = torch.ops._flashrt_fp8_ffn_cuda_f1d4adb
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_flashrt_fp8_ffn_cuda_f1d4adb::{op_name}"
|
build/torch211-cxx11-cu130-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
+
"id": "_flashrt_fp8_ffn_cuda_f1d4adb",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
build/torch211-cxx11-rocm71-x86_64-linux/{_flashrt_fp8_ffn_rocm_6ec89eb.abi3.so → _flashrt_fp8_ffn_rocm_f1d4adb.abi3.so}
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 167328
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:183b410023ba80d7c45a0814561e3906b98417c9becd7868cb5a63643307ab02
|
| 3 |
size 167328
|
build/torch211-cxx11-rocm71-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 _flashrt_fp8_ffn_rocm_f1d4adb
|
| 3 |
+
ops = torch.ops._flashrt_fp8_ffn_rocm_f1d4adb
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_flashrt_fp8_ffn_rocm_f1d4adb::{op_name}"
|
build/torch211-cxx11-rocm71-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
+
"id": "_flashrt_fp8_ffn_rocm_f1d4adb",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
build/torch211-cxx11-rocm72-x86_64-linux/{_flashrt_fp8_ffn_rocm_6ec89eb.abi3.so → _flashrt_fp8_ffn_rocm_f1d4adb.abi3.so}
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 165760
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:5f267f74dd6cf68b8cd8f378376439a5cf5772f2a230378eaed7f6b927716dca
|
| 3 |
size 165760
|
build/torch211-cxx11-rocm72-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 _flashrt_fp8_ffn_rocm_f1d4adb
|
| 3 |
+
ops = torch.ops._flashrt_fp8_ffn_rocm_f1d4adb
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_flashrt_fp8_ffn_rocm_f1d4adb::{op_name}"
|
build/torch211-cxx11-rocm72-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
+
"id": "_flashrt_fp8_ffn_rocm_f1d4adb",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
build/torch212-cxx11-cu130-x86_64-linux/_flashrt_fp8_ffn_cuda_6ec89eb.abi3.so
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:e6d9361c58816342dc79636e88c950840bb1fbdbeb5ccf690ea125c838851ae7
|
| 3 |
-
size 376296
|
|
|
|
|
|
|
|
|
|
|
|
build/torch212-cxx11-cu130-x86_64-linux/_flashrt_fp8_ffn_cuda_f1d4adb.abi3.so
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:79f128ef6afaca3f310b60e4b850522093cebbb9538f6d3e22a4434198660ac6
|
| 3 |
+
size 646904
|
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 _flashrt_fp8_ffn_cuda_f1d4adb
|
| 3 |
+
ops = torch.ops._flashrt_fp8_ffn_cuda_f1d4adb
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_flashrt_fp8_ffn_cuda_f1d4adb::{op_name}"
|
build/torch212-cxx11-cu130-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
+
"id": "_flashrt_fp8_ffn_cuda_f1d4adb",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
build/torch212-cxx11-cu132-x86_64-linux/_flashrt_fp8_ffn_cuda_6ec89eb.abi3.so
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:602c56052c4e10d26275d245c2325b57dd4ebb5f845fb560985363b279ddaaa6
|
| 3 |
-
size 376296
|
|
|
|
|
|
|
|
|
|
|
|
build/torch212-cxx11-cu132-x86_64-linux/_flashrt_fp8_ffn_cuda_f1d4adb.abi3.so
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9358cb7cd89212d3b073568ccc3f725d97bdb7af999da8bbf1559d1949459452
|
| 3 |
+
size 651000
|
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 _flashrt_fp8_ffn_cuda_f1d4adb
|
| 3 |
+
ops = torch.ops._flashrt_fp8_ffn_cuda_f1d4adb
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_flashrt_fp8_ffn_cuda_f1d4adb::{op_name}"
|
build/torch212-cxx11-cu132-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
+
"id": "_flashrt_fp8_ffn_cuda_f1d4adb",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
build/torch212-cxx11-rocm71-x86_64-linux/{_flashrt_fp8_ffn_rocm_6ec89eb.abi3.so → _flashrt_fp8_ffn_rocm_f1d4adb.abi3.so}
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 167224
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:3360caf36ad6fa3ec422873fb01078d469eed47196cc01182fb8b655f0e5abbd
|
| 3 |
size 167224
|
build/torch212-cxx11-rocm71-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 _flashrt_fp8_ffn_rocm_f1d4adb
|
| 3 |
+
ops = torch.ops._flashrt_fp8_ffn_rocm_f1d4adb
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_flashrt_fp8_ffn_rocm_f1d4adb::{op_name}"
|
build/torch212-cxx11-rocm71-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
+
"id": "_flashrt_fp8_ffn_rocm_f1d4adb",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
build/torch212-cxx11-rocm72-x86_64-linux/{_flashrt_fp8_ffn_rocm_6ec89eb.abi3.so → _flashrt_fp8_ffn_rocm_f1d4adb.abi3.so}
RENAMED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 165656
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ee057225282f3f2dc74037b47d83be93464eccf1bbe7af81e3592d918a5ae5d8
|
| 3 |
size 165656
|
build/torch212-cxx11-rocm72-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 _flashrt_fp8_ffn_rocm_f1d4adb
|
| 3 |
+
ops = torch.ops._flashrt_fp8_ffn_rocm_f1d4adb
|
| 4 |
|
| 5 |
def add_op_namespace_prefix(op_name: str):
|
| 6 |
"""
|
| 7 |
Prefix op by namespace.
|
| 8 |
"""
|
| 9 |
+
return f"_flashrt_fp8_ffn_rocm_f1d4adb::{op_name}"
|
build/torch212-cxx11-rocm72-x86_64-linux/metadata.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
-
"id": "
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|
|
|
|
| 1 |
{
|
| 2 |
"name": "flashrt-fp8-ffn",
|
| 3 |
+
"id": "_flashrt_fp8_ffn_rocm_f1d4adb",
|
| 4 |
"version": 1,
|
| 5 |
"license": "Apache-2.0",
|
| 6 |
"python-depends": [],
|