Skip to content

testing autoquant #114

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Open
wants to merge 4 commits into
base: gh/HDCharles/4/base
Choose a base branch
from
Open

testing autoquant #114

wants to merge 4 commits into from

Conversation

HDCharles
Copy link
Contributor

@HDCharles HDCharles commented Mar 1, 2024 •
edited
Loading

Stack from ghstack (oldest at bottom):

Summary:

improves runtime by 19.70 -> 19.76 img/sec

Test Plan: sh run.sh

Reviewers:

Subscribers:

Tasks:

Tags:

@HDCharles
testing autoquant
c44f0d0 
Summary:

improves runtime by 19.70 -> 19.76 img/sec

❯ one sh run.sh
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [06:32<00:00,  6.14s/it]
sam_model_type,batch_size,memory(MiB),memory(%),img_s(avg),batch_ms(avg)/batch_size,mIoU,use_compile,use_half,compress,epilogue_fusion_first,use_compile_decoder,use_nested_tensor,use_rel_pos,pad_input_image_batch,num_workers,num_batches,num_images,profile_path,memory_path
vit_h,16,14532,17,18.861125832244333,53.01910442113876,0.5865236891447146,max-autotune,torch.bfloat16,None,False,False,True,True,True,32,64,1024,None,None
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [07:08<00:00,  6.70s/it]
vit_h,16,14395,17,19.70834741975898,50.73992145061493,0.5875230894143607,max-autotune,torch.bfloat16,dynamic_quant,False,False,True,True,True,32,64,1024,None,None
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 3.850527899339795
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.3931088875979185
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.3931088875979185 3.190660197287798
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 4.768232116475701
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 3.8598313461989164
shape=(torch.Size([78400, 1280]), torch.Size([3840, 1280]), torch.Size([3840])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 1.4865157660096884
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.8800818361341953
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.8800818361341953 1.179535873234272
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 1.7427184619009497
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 1.4965661568567157
shape=(torch.Size([78400, 1280]), torch.Size([1280, 1280]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQFloatLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 4.215262923389673
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.661373794078827
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.661373794078827 3.485689079388976
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 5.220260447822511
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 4.2220821138471365
shape=(torch.Size([65536, 1280]), torch.Size([5120, 1280]), torch.Size([5120])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 4.666170105338097
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.113288130611181
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.113288130611181 2.626298717223108
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 4.855024302378297
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 4.674202110618353
shape=(torch.Size([65536, 5120]), torch.Size([1280, 5120]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 3.2269158866256475
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 3.7462301552295685
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 3.7462301552295685 2.6572815608233213
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 3.9978391956537966
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 3.2370124012231827
shape=(torch.Size([65536, 1280]), torch.Size([3840, 1280]), torch.Size([3840])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 1.2530277017503977
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.5717314090579748
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.5717314090579748 0.9894231799989939
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 1.5166664496064186
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 1.2606457574293017
shape=(torch.Size([65536, 1280]), torch.Size([1280, 1280]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQFloatLinearWeight'>
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [02:15<00:00,  2.12s/it]
vit_h,16,14463,17,19.76190752324237,50.602402567863464,0.5875653903095147,max-autotune,torch.bfloat16,auto_quant,False,False,True,True,True,32,64,1024,None,None

Test Plan:

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
HDCharles added a commit that referenced this pull request Mar 1, 2024
@HDCharles
testing autoquant
d5c1472 
Summary:

improves runtime by 19.70 -> 19.76 img/sec

❯ one sh run.sh
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [06:32<00:00,  6.14s/it]
sam_model_type,batch_size,memory(MiB),memory(%),img_s(avg),batch_ms(avg)/batch_size,mIoU,use_compile,use_half,compress,epilogue_fusion_first,use_compile_decoder,use_nested_tensor,use_rel_pos,pad_input_image_batch,num_workers,num_batches,num_images,profile_path,memory_path
vit_h,16,14532,17,18.861125832244333,53.01910442113876,0.5865236891447146,max-autotune,torch.bfloat16,None,False,False,True,True,True,32,64,1024,None,None
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [07:08<00:00,  6.70s/it]
vit_h,16,14395,17,19.70834741975898,50.73992145061493,0.5875230894143607,max-autotune,torch.bfloat16,dynamic_quant,False,False,True,True,True,32,64,1024,None,None
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 3.850527899339795
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.3931088875979185
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.3931088875979185 3.190660197287798
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 4.768232116475701
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 3.8598313461989164
shape=(torch.Size([78400, 1280]), torch.Size([3840, 1280]), torch.Size([3840])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 1.4865157660096884
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.8800818361341953
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.8800818361341953 1.179535873234272
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 1.7427184619009497
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 1.4965661568567157
shape=(torch.Size([78400, 1280]), torch.Size([1280, 1280]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQFloatLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 4.215262923389673
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.661373794078827
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.661373794078827 3.485689079388976
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 5.220260447822511
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 4.2220821138471365
shape=(torch.Size([65536, 1280]), torch.Size([5120, 1280]), torch.Size([5120])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 4.666170105338097
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.113288130611181
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.113288130611181 2.626298717223108
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 4.855024302378297
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 4.674202110618353
shape=(torch.Size([65536, 5120]), torch.Size([1280, 5120]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 3.2269158866256475
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 3.7462301552295685
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 3.7462301552295685 2.6572815608233213
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 3.9978391956537966
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 3.2370124012231827
shape=(torch.Size([65536, 1280]), torch.Size([3840, 1280]), torch.Size([3840])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 1.2530277017503977
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.5717314090579748
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.5717314090579748 0.9894231799989939
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 1.5166664496064186
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 1.2606457574293017
shape=(torch.Size([65536, 1280]), torch.Size([1280, 1280]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQFloatLinearWeight'>
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [02:15<00:00,  2.12s/it]
vit_h,16,14463,17,19.76190752324237,50.602402567863464,0.5875653903095147,max-autotune,torch.bfloat16,auto_quant,False,False,True,True,True,32,64,1024,None,None

Test Plan:

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: ac0ddc1
Pull Request resolved: #114
@facebook-github-bot facebook-github-bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Mar 1, 2024
@HDCharles
Update on "testing autoquant"
36d43be 
Summary:

improves runtime by 19.70 -> 19.76 img/sec

❯ one sh run.sh
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [06:32<00:00,  6.14s/it]
sam_model_type,batch_size,memory(MiB),memory(%),img_s(avg),batch_ms(avg)/batch_size,mIoU,use_compile,use_half,compress,epilogue_fusion_first,use_compile_decoder,use_nested_tensor,use_rel_pos,pad_input_image_batch,num_workers,num_batches,num_images,profile_path,memory_path
vit_h,16,14532,17,18.861125832244333,53.01910442113876,0.5865236891447146,max-autotune,torch.bfloat16,None,False,False,True,True,True,32,64,1024,None,None
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [07:08<00:00,  6.70s/it]
vit_h,16,14395,17,19.70834741975898,50.73992145061493,0.5875230894143607,max-autotune,torch.bfloat16,dynamic_quant,False,False,True,True,True,32,64,1024,None,None
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 3.850527899339795
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.3931088875979185
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.3931088875979185 3.190660197287798
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 4.768232116475701
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 3.8598313461989164
shape=(torch.Size([78400, 1280]), torch.Size([3840, 1280]), torch.Size([3840])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 1.4865157660096884
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.8800818361341953
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.8800818361341953 1.179535873234272
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 1.7427184619009497
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 1.4965661568567157
shape=(torch.Size([78400, 1280]), torch.Size([1280, 1280]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQFloatLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 4.215262923389673
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.661373794078827
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.661373794078827 3.485689079388976
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 5.220260447822511
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 4.2220821138471365
shape=(torch.Size([65536, 1280]), torch.Size([5120, 1280]), torch.Size([5120])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 4.666170105338097
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.113288130611181
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.113288130611181 2.626298717223108
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 4.855024302378297
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 4.674202110618353
shape=(torch.Size([65536, 5120]), torch.Size([1280, 5120]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 3.2269158866256475
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 3.7462301552295685
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 3.7462301552295685 2.6572815608233213
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 3.9978391956537966
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 3.2370124012231827
shape=(torch.Size([65536, 1280]), torch.Size([3840, 1280]), torch.Size([3840])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 1.2530277017503977
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.5717314090579748
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.5717314090579748 0.9894231799989939
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 1.5166664496064186
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 1.2606457574293017
shape=(torch.Size([65536, 1280]), torch.Size([1280, 1280]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQFloatLinearWeight'>
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [02:15<00:00,  2.12s/it]
vit_h,16,14463,17,19.76190752324237,50.602402567863464,0.5875653903095147,max-autotune,torch.bfloat16,auto_quant,False,False,True,True,True,32,64,1024,None,None

Test Plan:

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
HDCharles added a commit that referenced this pull request Mar 5, 2024
@HDCharles
testing autoquant
19bd380 
Summary:

improves runtime by 19.70 -> 19.76 img/sec

❯ one sh run.sh
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [06:32<00:00,  6.14s/it]
sam_model_type,batch_size,memory(MiB),memory(%),img_s(avg),batch_ms(avg)/batch_size,mIoU,use_compile,use_half,compress,epilogue_fusion_first,use_compile_decoder,use_nested_tensor,use_rel_pos,pad_input_image_batch,num_workers,num_batches,num_images,profile_path,memory_path
vit_h,16,14532,17,18.861125832244333,53.01910442113876,0.5865236891447146,max-autotune,torch.bfloat16,None,False,False,True,True,True,32,64,1024,None,None
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [07:08<00:00,  6.70s/it]
vit_h,16,14395,17,19.70834741975898,50.73992145061493,0.5875230894143607,max-autotune,torch.bfloat16,dynamic_quant,False,False,True,True,True,32,64,1024,None,None
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 3.850527899339795
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.3931088875979185
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.3931088875979185 3.190660197287798
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 4.768232116475701
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 3.8598313461989164
shape=(torch.Size([78400, 1280]), torch.Size([3840, 1280]), torch.Size([3840])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 1.4865157660096884
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.8800818361341953
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.8800818361341953 1.179535873234272
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 1.7427184619009497
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 1.4965661568567157
shape=(torch.Size([78400, 1280]), torch.Size([1280, 1280]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQFloatLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 4.215262923389673
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.661373794078827
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.661373794078827 3.485689079388976
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 5.220260447822511
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 4.2220821138471365
shape=(torch.Size([65536, 1280]), torch.Size([5120, 1280]), torch.Size([5120])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 4.666170105338097
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.113288130611181
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.113288130611181 2.626298717223108
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 4.855024302378297
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 4.674202110618353
shape=(torch.Size([65536, 5120]), torch.Size([1280, 5120]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 3.2269158866256475
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 3.7462301552295685
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 3.7462301552295685 2.6572815608233213
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 3.9978391956537966
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 3.2370124012231827
shape=(torch.Size([65536, 1280]), torch.Size([3840, 1280]), torch.Size([3840])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 1.2530277017503977
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.5717314090579748
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.5717314090579748 0.9894231799989939
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 1.5166664496064186
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 1.2606457574293017
shape=(torch.Size([65536, 1280]), torch.Size([1280, 1280]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQFloatLinearWeight'>
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [02:15<00:00,  2.12s/it]
vit_h,16,14463,17,19.76190752324237,50.602402567863464,0.5875653903095147,max-autotune,torch.bfloat16,auto_quant,False,False,True,True,True,32,64,1024,None,None

Test Plan:

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: a9134d6
Pull Request resolved: #114
@HDCharles HDCharles mentioned this pull request Mar 5, 2024
Merged
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 5, 2024
@HDCharles
Update on "Autoquant"
c6d59e5 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
(pytorch-labs/segment-anything-fast#114,
huggingface/diffusion-fast@176e85f)

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 5, 2024
@HDCharles
Autoquant
7995409 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
(pytorch-labs/segment-anything-fast#114,
huggingface/diffusion-fast@176e85f)

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: 3986099
Pull Request resolved: #38
@HDCharles
Update on "testing autoquant"
b7e8439 
Summary:

improves runtime by 19.70 -> 19.76 img/sec

❯ one sh run.sh
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [06:32<00:00,  6.14s/it]
sam_model_type,batch_size,memory(MiB),memory(%),img_s(avg),batch_ms(avg)/batch_size,mIoU,use_compile,use_half,compress,epilogue_fusion_first,use_compile_decoder,use_nested_tensor,use_rel_pos,pad_input_image_batch,num_workers,num_batches,num_images,profile_path,memory_path
vit_h,16,14532,17,18.861125832244333,53.01910442113876,0.5865236891447146,max-autotune,torch.bfloat16,None,False,False,True,True,True,32,64,1024,None,None
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [07:08<00:00,  6.70s/it]
vit_h,16,14395,17,19.70834741975898,50.73992145061493,0.5875230894143607,max-autotune,torch.bfloat16,dynamic_quant,False,False,True,True,True,32,64,1024,None,None
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 3.850527899339795
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.3931088875979185
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.3931088875979185 3.190660197287798
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 4.768232116475701
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 3.8598313461989164
shape=(torch.Size([78400, 1280]), torch.Size([3840, 1280]), torch.Size([3840])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 1.4865157660096884
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.8800818361341953
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.8800818361341953 1.179535873234272
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 1.7427184619009497
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 1.4965661568567157
shape=(torch.Size([78400, 1280]), torch.Size([1280, 1280]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQFloatLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 4.215262923389673
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.661373794078827
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.661373794078827 3.485689079388976
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 5.220260447822511
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 4.2220821138471365
shape=(torch.Size([65536, 1280]), torch.Size([5120, 1280]), torch.Size([5120])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 4.666170105338097
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.113288130611181
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 4.113288130611181 2.626298717223108
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 4.855024302378297
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 4.674202110618353
shape=(torch.Size([65536, 5120]), torch.Size([1280, 5120]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 3.2269158866256475
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 3.7462301552295685
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 3.7462301552295685 2.6572815608233213
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 3.9978391956537966
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 3.2370124012231827
shape=(torch.Size([65536, 1280]), torch.Size([3840, 1280]), torch.Size([3840])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'>
<class 'torchao.quantization.autoquant.AQFloatLinearWeight'> 1.2530277017503977
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.5717314090579748
<class 'torchao.quantization.autoquant.AQInt8DynamicallyQuantizedLinearWeight'> 1.5717314090579748 0.9894231799989939
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight'> 1.5166664496064186
<class 'torchao.quantization.autoquant.AQWeightOnlyQuantizedLinearWeight3'> 1.2606457574293017
shape=(torch.Size([65536, 1280]), torch.Size([1280, 1280]), torch.Size([1280])), dtype=torch.bfloat16, best_cls=<class 'torchao.quantization.autoquant.AQFloatLinearWeight'>
  0%|                                                                                                                                                                                                              | 0/64 [00:00<?, ?it/s]/home/cdhernandez/local/pytorch/torch/nested/__init__.py:166: UserWarning: The PyTorch API of nested tensors is in prototype stage and will change in the near future. (Triggered internally at /home/cdhernandez/local/pytorch/aten/src/ATen/NestedTensorImpl.cpp:177.)
  return _nested.nested_tensor(
100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 64/64 [02:15<00:00,  2.12s/it]
vit_h,16,14463,17,19.76190752324237,50.602402567863464,0.5875653903095147,max-autotune,torch.bfloat16,auto_quant,False,False,True,True,True,32,64,1024,None,None

Test Plan:

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
HDCharles added a commit that referenced this pull request Mar 19, 2024
@HDCharles
testing autoquant
8f28b42 
Summary:

improves runtime by 19.70 -> 19.76 img/sec

Test Plan: sh run.sh

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: 9c6098e
Pull Request resolved: #114
@HDCharles
Update on "testing autoquant"
bbd94ac 
Summary:

improves runtime by 19.70 -> 19.76 img/sec

Test Plan: sh run.sh

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
HDCharles added a commit that referenced this pull request Mar 19, 2024
@HDCharles
testing autoquant
1ae9a9a 
Summary:

improves runtime by 19.70 -> 19.76 img/sec

Test Plan: sh run.sh

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: 9c6098e
Pull Request resolved: #114
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Update on "Autoquant"
97733c2 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Autoquant
a37928a 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: 0dbb2ff
Pull Request resolved: #38
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Update on "Autoquant"
29214a9 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Autoquant
71300c3 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: fddbaf2
Pull Request resolved: #38
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Update base for Update on "Autoquant"
2ae74d3 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Update on "Autoquant"
bc2deb7 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Autoquant
fbda746 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: f268031
Pull Request resolved: #38
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Update base for Update on "Autoquant"
490c7c1 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: [D55103983](https://our.internmc.facebook.com/intern/diff/D55103983)

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Update on "Autoquant"
4aae2a3 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: [D55103983](https://our.internmc.facebook.com/intern/diff/D55103983)

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Autoquant
f66419e 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: 94089f7
Pull Request resolved: #38
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Update base for Update on "Autoquant"
69f8abb 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: [D55103983](https://our.internmc.facebook.com/intern/diff/D55103983)

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Update on "Autoquant"
8dcfafe 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: [D55103983](https://our.internmc.facebook.com/intern/diff/D55103983)

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Autoquant
57003a5 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: 94089f7
Pull Request resolved: #38
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Update base for Update on "Autoquant"
7fd7aac 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: [D55103983](https://our.internmc.facebook.com/intern/diff/D55103983)

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Update on "Autoquant"
55cce68 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: [D55103983](https://our.internmc.facebook.com/intern/diff/D55103983)

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Autoquant
f71c9cf 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: 94089f7
Pull Request resolved: #38
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Update base for Update on "Autoquant"
886ac77 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: [D55103983](https://our.internmc.facebook.com/intern/diff/D55103983)

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Update on "Autoquant"
5f5bc8e 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: [D55103983](https://our.internmc.facebook.com/intern/diff/D55103983)

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 19, 2024
@HDCharles
Autoquant
9c0388e 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: 3768385
Pull Request resolved: #38
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 25, 2024
@HDCharles
Update base for Update on "Autoquant"
b79c5bf 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: [D55103983](https://our.internmc.facebook.com/intern/diff/D55103983)

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 25, 2024
@HDCharles
Autoquant
46e940d 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: 3c1199d
Pull Request resolved: #38
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 25, 2024
@HDCharles
Update on "Autoquant"
17c670a 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: [D55103983](https://our.internmc.facebook.com/intern/diff/D55103983)

[ghstack-poisoned]
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 25, 2024
@HDCharles
Autoquant
8bce92f 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
@HDCharles HDCharles mentioned this pull request Mar 25, 2024
Merged
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 25, 2024
@HDCharles
Autoquant
190b3ea 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

ghstack-source-id: 3c1199d
Pull Request resolved: #81
@HDCharles HDCharles mentioned this pull request Mar 25, 2024
Merged
HDCharles added a commit to pytorch/ao that referenced this pull request Mar 25, 2024
@HDCharles
Autoquant (#82)
8119319 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
cpuhrsch pushed a commit to pytorch/ao that referenced this pull request Apr 1, 2024
@HDCharles @cpuhrsch
Autoquant (#82)
a10d5b1 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
dbyoung18 pushed a commit to dbyoung18/ao that referenced this pull request Jul 31, 2024
@HDCharles
Autoquant (pytorch#82)
22ed8b9 
Summary: Adding autoquantization functionality, using hte do_quant api
we can test kernel speeds and pick the best quantization type (or no
quantization) for each layer.

Test Plan: python test/test.py -k "autoquant"

also tested on SAM and SDXL
pytorch-labs/segment-anything-fast#114
HDCharles/sdxl-fast@8d9942a

Reviewers:

Subscribers:

Tasks:

Tags:

[ghstack-poisoned]
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers
No reviews
Assignees
No one assigned
Labels
CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed.
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
2 participants
@HDCharles @facebook-github-bot