Skip to content
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

FP16 Execution produces non accurate results #41

Closed
narendasan opened this issue Apr 23, 2020 · 0 comments · Fixed by #44
Closed

FP16 Execution produces non accurate results #41

narendasan opened this issue Apr 23, 2020 · 0 comments · Fixed by #44
Assignees
@narendasan
Labels
bug Something isn't working component: execution Issues re: Execution of engines priority: high

Comments

@narendasan
Copy link
Collaborator

Right now the same workflow to run FP32 models when used in FP16 gives bad results.
@narendasan narendasan added bug Something isn't working component: execution Issues re: Execution of engines labels Apr 23, 2020
@narendasan narendasan self-assigned this Apr 23, 2020
@narendasan narendasan mentioned this issue Apr 23, 2020
Merged
narendasan added a commit to narendasan/TRTorch that referenced this issue Apr 24, 2020
@narendasan
feat(/cpp/api): Working INT8 Calibrator, also resolves pytorch#41
5c0d737 
- Now creates output tensors of the correct type to accept data
- There still may be a data race in the creation of the dataloader
iterator
- Quantization and Dynamic Shape right now don't play well together,
potential subsequent release of TRT may address this

Signed-off-by: Naren Dasan <naren@narendasan.com>
Signed-off-by: Naren Dasan <narens@nvidia.com>
frank-wei pushed a commit that referenced this issue Jun 4, 2022
@tissue3
[fx2trt][bootcamp] Add support for 2 ops: torch.logical_or, torch.log…
5da146d 
…ical_xor (#41)

Summary:
Pull Request resolved: pytorch/fx2trt#41

fx2trt is a tool we use to create a TensorRT engine from a PyTorch model. The lowering is composed of 1) start the Pytorch model 2) frace model with acc tracer in acc_ops 3) Use TRT Interpreter to create a TensorRT engine

Here I:
1. Add corresponding acc ops
2. Add a converter for the acc op to acc_ops_converters.py.
3. Add a unit test for the converter in fbcode/deeplearning/trt/fx2trt_oss/test/converters/acc_op/test_logical_or/xor.py

Reviewed By: frank-wei

Differential Revision: D35237918

fbshipit-source-id: 82720b764f0c886749aafea84584cdcb5172d206
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees

@narendasan narendasan

Labels
bug Something isn't working component: execution Issues re: Execution of engines priority: high
Projects
None yet
Milestone
No milestone
Development

Successfully merging a pull request may close this issue.

Post training quantization support in TRTorch narendasan/TRTorch
1 participant
@narendasan