CharlieFRuan opened a new pull request, #16294: URL: https://github.com/apache/tvm/pull/16294
Currently when loading the params, we try to decode from bf16 to f32 regardless of the dtype of the param, since all the params are stored with "format=f32-to-bf16" in the record regardless of the dtype. We solve it by checking the dtype as well, just like the C++ counter part: https://github.com/apache/tvm/blob/4e66690a4d033af912f5051c0e5a16c9c10691d9/src/runtime/relax_vm/ndarray_cache_support.cc#L168-L172 -- 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]
