File size: 1,370 Bytes
937a94e
 
 
 
 
 
 
 
 
f9f24d7
937a94e
f9f24d7
 
937a94e
f9f24d7
 
937a94e
 
 
 
 
 
 
 
 
 
 
f5a3617
c5db835
 
 
 
 
f9f24d7
c5db835
 
 
f9f24d7
 
c5db835
f9f24d7
c5db835
 
968b96f
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
from typing import Any
from typing import Callable
from typing import ParamSpec
import spaces
import torch
from torch.utils._pytree import tree_map

P = ParamSpec('P')

TRANSFORMER_HIDDEN_DIM = torch.export.Dim('hidden', min=4096, max=8212)

# Specific to Flux. More about this is available in
# https://huggingface.co/blog/zerogpu-aoti
TRANSFORMER_DYNAMIC_SHAPES = {
    'hidden_states': {1: TRANSFORMER_HIDDEN_DIM},
    'img_ids': {0: TRANSFORMER_HIDDEN_DIM},
}

INDUCTOR_CONFIGS = {
    'conv_1x1_as_mm': True,
    'epilogue_fusion': False,
    'coordinate_descent_tuning': True,
    'coordinate_descent_check_all_directions': True,
    'max_autotune': True,
    'triton.cudagraphs': True,
}

def compile_transformer(pipeline: Callable[P, Any], *args: P.args, **kwargs: P.kwargs):
    @spaces.GPU(duration=1500)
    def f():
        with spaces.aoti_capture(pipeline.transformer) as call:
            pipeline(*args, **kwargs)

        dynamic_shapes = tree_map(lambda v: None, call.kwargs)
        dynamic_shapes |= TRANSFORMER_DYNAMIC_SHAPES

        exported = torch.export.export(
            mod=pipeline.transformer, 
            args=call.args, 
            kwargs=call.kwargs,
            dynamic_shapes=dynamic_shapes
        )
        return spaces.aoti_compile(exported, INDUCTOR_CONFIGS)
    
    compiled_transformer = f()
    return compiled_transformer