rickzx opened a new pull request, #16982:
URL: https://github.com/apache/tvm/pull/16982

   For workloads with a mixture of symbolic shape and concrete shape as batch 
sizes, we cannot directly use `int()` to obtain the batch size. Instead, we can 
use `arith.Analyzer` to check equality.
   
   For example:
   ```
   permute_dims1: R.Tensor((batch_size, 12, seq_len, 64), dtype="float16") = 
R.permute_dims(split_0, axes=[0, 2, 1, 3])
   permute_dims2: R.Tensor((batch_size, 12, seq_len, 64), dtype="float16") = 
R.permute_dims(split_1, axes=[0, 2, 1, 3])
   permute_dims3: R.Tensor((batch_size, 12, seq_len, 64), dtype="float16") = 
R.permute_dims(split_2, axes=[0, 2, 1, 3])
   permute_dims4: R.Tensor((batch_size, 12, 64, seq_len), dtype="float16") = 
R.permute_dims(permute_dims2, axes=[0, 1, 3, 2])
   matmul1: R.Tensor((batch_size, 12, seq_len, seq_len), dtype="float16") = 
R.matmul(permute_dims1, permute_dims4, out_dtype="float16")
   ``` 


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]

Reply via email to