Ishitori opened a new issue #12700: Add support of dot product (GEMM) for 
float16 on CPU
URL: https://github.com/apache/incubator-mxnet/issues/12700
 
 
   ## Description
   Currently dot product for float16 is supported only on GPU (was added 
https://github.com/apache/incubator-mxnet/issues/10531), but is not supported 
on CPU. It would be great to add it to CPU for fast inference.
   
   
   
   Package used (Python/R/Scala/Julia):
   Python, mxnet 1.3.0
   
   ## Error Message:
   ```
   Traceback (most recent call last):
     File "<stdin>", line 1, in <module>
     File 
"/home/anaconda3/lib/python3.6/site-packages/mxnet/ndarray/ndarray.py", line 
189, in __repr__
       return '\n%s\n<%s %s @%s>' % (str(self.asnumpy()),
     File 
"/home/anaconda3/lib/python3.6/site-packages/mxnet/ndarray/ndarray.py", line 
1972, in asnumpy
       ctypes.c_size_t(data.size)))
     File "/home/anaconda3/lib/python3.6/site-packages/mxnet/base.py", line 
252, in check_call
       raise MXNetError(py_str(_LIB.MXGetLastError()))
   mxnet.base.MXNetError: [13:42:46] src/operator/tensor/./dot-inl.h:83: Check 
failed: outputs[0].type_flag_ == kFloat32 || outputs[0].type_flag_ == kFloat64 
|| (outputs[0].type_flag_ == kFloat16 && ctx.run_ctx.ctx.dev_mask() == 
mshadow::gpu::kDevMask) dot only supports float32/float64 for CPU, and 
float16/float32/float64 for GPU
   ```
   
   ## Minimum reproducible example
   ```
   a = mx.nd.array([1,2,3], dtype="float16")
   b = mx.nd.array([1,2,3], dtype="float16")
   mx.nd.dot(a, b.T)
   ```
   ## Steps to reproduce
   
   1. Run the Minimum reproducible example and see the exception
   
   ## What have you tried to solve it?
   
   1. There is nothing can be done, only either work with GPU or do float32 
math.

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
[email protected]


With regards,
Apache Git Services

Reply via email to