[habana] Misc fixes for vision nets #2838
Conversation
bertmaher
commented
May 1, 2019
bertmaher
commented
- Introduced LRN, fixed Add message when images are missing for CIFAR10 test #144
- Added Concat node to the list of supported quantized nodes
- Added RescaleQuantized node, replaced quantized MatMul with GEMM
| params->pWend = pad[0]; | ||
| params->pHbegin = pad[1]; | ||
| params->pHend = pad[1]; | ||
| params->pHbegin = pad[0]; |
|
|
||
| chk(synCreateGenericNode(&tensors[NI->getInput()].get(), | ||
| &tensors[NI].get(), 1, 1, (void *)params.get(), | ||
| "lrn_f32", NI->getName().str().c_str(), nullptr, |
There was a problem hiding this comment.
Is the datatype intentionally hardcoded? Does this operator not exist for other datatypes? It makes sense that it wouldn't, but I just want to confirm.
There was a problem hiding this comment.
Good eye - LRN does indeed only exist for f32. We could assert that here, although it'll already be caught by isOpSupported.
|
Also, it's interesting that we need LRN because I haven't heard of it being used for months now. |
|
It's used in inception_v1 and bvlc_alexnet. Unless it usually gets optimized away somehow? |
|
Ah, those two are 4-8 years old at this point. This PR made me think some recent models use them. |
* Introduced LRN, fixed pytorch#144 * Added Concat node to the list of supported quantized nodes * Added RescaleQuantized node, replaced quantized MatMul with GEMM
|
LRN indeed aren't used in modern topologies. I introduced it more like an exe4cise and to enable inception_v1. Besides, Habana supports it so why not? Indeed it is only available for fp32. |
