Published October 27, 2017
| Version v2.3.0
Software
Open
NervanaSystems/neon: CPU optimizations bringing ~7x gain in Deep Speech 2 performance and gains in Alexnet, VGG
Authors/Creators
- Scott Leishman
- Alex Park1
- Anil Thomas
- Jennifer Myers2
- yinyinl3
- nervetumer
- Urs Köster4
- hanlin-nervana
- Scott Gray5
- Augustus Odena
- wconstab
- Zach Dwiel
- Wei Wang6
- John Co-Reyes
- Peng Zhang7
- Arjun Bansal3
- Sébastien Arnold8
- Andy
- DawnStone
- Jeevan Shankar9
- kulig
- Tyler Lee
- Scott Cyphers10
- ruby-nervana
- Kai Arulkumaran
- Santi Villalba
- Xin Wang6
- Steven Robertson11
- Gabriel Pereyra12
- aravind-nrv
- 1. Cerebras Systems
- 2. @NervanaSystems
- 3. Nervana Systems
- 4. Cerebras
- 5. OpenAI
- 6. Intel Corporation
- 7. Intel
- 8. University of Southern California
- 9. Facebook
- 10. Intel Nervana
- 11. Intel - Nervana
- 12. Student
Description
- Optimized DeepSpeech2 MKL backend performance (~7X improvement over the CPU backend)
- Fused convolution and bias layer which significantly boosted AlexNet and VGG performance on Intel architectures with MKL backend
- Made SSD and Faster-RNN use VGG weight files in new format
- Fixed use of reset_cells hyperparameter
- Fixed MKL backend bug for GAN and Faster-RCNN models
Files
NervanaSystems/neon-v2.3.0.zip
Files
(3.3 MB)
| Name | Size | Download all |
|---|---|---|
|
md5:d9783f0467abb2491dd07bfb930a18a8
|
3.3 MB | Preview Download |
Additional details
Related works
- Is supplement to
- https://github.com/NervanaSystems/neon/tree/v2.3.0 (URL)