CharlieFRuan opened a new pull request, #16220: URL: https://github.com/apache/tvm/pull/16220
This is motivated by supporting models like GPTNeoX that have the field `ffn_out_dtype` in config, making us set it as `out_dtype`. After the `matmul`, `x` would have type `out_dtype` (`x = op.matmul(x, w, out_dtype=self.out_dtype`). When `bias=True`, we want `bias` to be of type `out_dtype`, even if we call `to()`; otherwise, we may have a type mismatch when computing `x + self.bias`. -- 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]
