remove torchao dependency, quantize entirely via linear
Browse files- float8_quantize.py +31 -25
- requirements.txt +0 -1
float8_quantize.py
CHANGED
@@ -1,11 +1,6 @@
|
|
1 |
from loguru import logger
|
2 |
import torch
|
3 |
import torch.nn as nn
|
4 |
-
from torchao.float8.float8_utils import (
|
5 |
-
amax_to_scale,
|
6 |
-
tensor_to_amax,
|
7 |
-
to_fp8_saturated,
|
8 |
-
)
|
9 |
from torch.nn import init
|
10 |
import math
|
11 |
from torch.compiler import is_compiling
|
@@ -200,42 +195,55 @@ class F8Linear(nn.Module):
|
|
200 |
def quantize_weight(self):
|
201 |
if self.weight_initialized:
|
202 |
return
|
203 |
-
amax =
|
204 |
-
scale = amax_to_scale(amax, self.
|
205 |
-
self.float8_data = to_fp8_saturated(
|
206 |
-
|
207 |
-
self.
|
208 |
-
self.scale_reciprocal = self.scale.reciprocal()
|
209 |
self.weight.data = torch.zeros(
|
210 |
1, dtype=self.weight.dtype, device=self.weight.device, requires_grad=False
|
211 |
)
|
|
|
212 |
|
213 |
def set_weight_tensor(self, tensor: torch.Tensor):
|
214 |
self.weight.data = tensor
|
215 |
self.weight_initialized = False
|
216 |
self.quantize_weight()
|
217 |
|
|
|
|
|
|
|
|
|
|
|
|
|
218 |
def quantize_input(self, x: torch.Tensor):
|
219 |
if self.input_scale_initialized:
|
220 |
-
return to_fp8_saturated(x
|
|
|
|
|
221 |
elif self.trial_index < self.num_scale_trials:
|
222 |
-
|
|
|
|
|
223 |
self.input_amax_trials[self.trial_index] = amax
|
224 |
self.trial_index += 1
|
225 |
-
self.input_scale = amax_to_scale(
|
226 |
-
self.input_amax_trials[: self.trial_index].max(),
|
227 |
-
self.input_float8_dtype,
|
228 |
-
self.weight.dtype,
|
229 |
)
|
230 |
self.input_scale_reciprocal = self.input_scale.reciprocal()
|
231 |
-
return to_fp8_saturated(x
|
|
|
|
|
232 |
else:
|
233 |
-
self.input_scale = amax_to_scale(
|
234 |
-
self.input_amax_trials.max(), self.
|
235 |
)
|
236 |
self.input_scale_reciprocal = self.input_scale.reciprocal()
|
237 |
self.input_scale_initialized = True
|
238 |
-
return to_fp8_saturated(x
|
|
|
|
|
239 |
|
240 |
def reset_parameters(self) -> None:
|
241 |
if self.weight_initialized:
|
@@ -263,10 +271,8 @@ class F8Linear(nn.Module):
|
|
263 |
|
264 |
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
265 |
if self.input_scale_initialized or is_compiling():
|
266 |
-
x = (
|
267 |
-
|
268 |
-
.clamp(min=-self.input_max_value, max=self.input_max_value)
|
269 |
-
.type(self.input_float8_dtype)
|
270 |
)
|
271 |
else:
|
272 |
x = self.quantize_input(x)
|
|
|
1 |
from loguru import logger
|
2 |
import torch
|
3 |
import torch.nn as nn
|
|
|
|
|
|
|
|
|
|
|
4 |
from torch.nn import init
|
5 |
import math
|
6 |
from torch.compiler import is_compiling
|
|
|
195 |
def quantize_weight(self):
|
196 |
if self.weight_initialized:
|
197 |
return
|
198 |
+
amax = torch.max(torch.abs(self.weight.data)).float()
|
199 |
+
self.scale = self.amax_to_scale(amax, self.max_value)
|
200 |
+
self.float8_data = self.to_fp8_saturated(
|
201 |
+
self.weight.data, self.scale, self.max_value
|
202 |
+
).to(self.float8_dtype)
|
203 |
+
self.scale_reciprocal = self.scale.reciprocal()
|
204 |
self.weight.data = torch.zeros(
|
205 |
1, dtype=self.weight.dtype, device=self.weight.device, requires_grad=False
|
206 |
)
|
207 |
+
self.weight_initialized = True
|
208 |
|
209 |
def set_weight_tensor(self, tensor: torch.Tensor):
|
210 |
self.weight.data = tensor
|
211 |
self.weight_initialized = False
|
212 |
self.quantize_weight()
|
213 |
|
214 |
+
def amax_to_scale(self, amax, max_val):
|
215 |
+
return (max_val / torch.clamp(amax, min=1e-12)).clamp(max=max_val)
|
216 |
+
|
217 |
+
def to_fp8_saturated(self, x, scale, max_val):
|
218 |
+
return (x * scale).clamp(-max_val, max_val)
|
219 |
+
|
220 |
def quantize_input(self, x: torch.Tensor):
|
221 |
if self.input_scale_initialized:
|
222 |
+
return self.to_fp8_saturated(x, self.input_scale, self.input_max_value).to(
|
223 |
+
self.input_float8_dtype
|
224 |
+
)
|
225 |
elif self.trial_index < self.num_scale_trials:
|
226 |
+
|
227 |
+
amax = torch.max(torch.abs(x)).float()
|
228 |
+
|
229 |
self.input_amax_trials[self.trial_index] = amax
|
230 |
self.trial_index += 1
|
231 |
+
self.input_scale = self.amax_to_scale(
|
232 |
+
self.input_amax_trials[: self.trial_index].max(), self.input_max_value
|
|
|
|
|
233 |
)
|
234 |
self.input_scale_reciprocal = self.input_scale.reciprocal()
|
235 |
+
return self.to_fp8_saturated(x, self.input_scale, self.input_max_value).to(
|
236 |
+
self.input_float8_dtype
|
237 |
+
)
|
238 |
else:
|
239 |
+
self.input_scale = self.amax_to_scale(
|
240 |
+
self.input_amax_trials.max(), self.input_max_value
|
241 |
)
|
242 |
self.input_scale_reciprocal = self.input_scale.reciprocal()
|
243 |
self.input_scale_initialized = True
|
244 |
+
return self.to_fp8_saturated(x, self.input_scale, self.input_max_value).to(
|
245 |
+
self.input_float8_dtype
|
246 |
+
)
|
247 |
|
248 |
def reset_parameters(self) -> None:
|
249 |
if self.weight_initialized:
|
|
|
271 |
|
272 |
def forward(self, x: torch.Tensor) -> torch.Tensor:
|
273 |
if self.input_scale_initialized or is_compiling():
|
274 |
+
x = self.to_fp8_saturated(x, self.input_scale, self.input_max_value).to(
|
275 |
+
self.input_float8_dtype
|
|
|
|
|
276 |
)
|
277 |
else:
|
278 |
x = self.quantize_input(x)
|
requirements.txt
CHANGED
@@ -1,5 +1,4 @@
|
|
1 |
git+https://github.com/aredden/torch-cublas-hgemm.git@master
|
2 |
-
git+https://github.com/pytorch/ao.git@main
|
3 |
einops
|
4 |
PyTurboJPEG
|
5 |
pydantic
|
|
|
1 |
git+https://github.com/aredden/torch-cublas-hgemm.git@master
|
|
|
2 |
einops
|
3 |
PyTurboJPEG
|
4 |
pydantic
|