[TVM Discuss] [Questions] Question about use autotvm for quantization

Fri, 10 Apr 2020 06:59:48 -0700 


HI,I see the code vta/tutorials/autotvm/tune_relay_vta.py and other‘s 
discussions on the forum.

usually quantize a model and create auto-tuning task by 
**autotvm.task.extract_from_program**

    tasks = **autotvm.task.extract_from_program**(mod,  params=**params**,
                                                  
ops=(relay.op.get("nn.conv2d"),),
                                                  target=target,
                                                  target_host=env.target_host)


The ‘params’ here are from a float model.Why can it create a task used to 
quanzited model? 

I think ***extract_from_program’s*** parameter(***params***) should be a 
quantized params.

But how can we get this ‘params’ easily? 
   Thanks.





---
[Visit 
Topic](https://discuss.tvm.ai/t/question-about-use-autotvm-for-quantization/6323/1)
 to respond.

You are receiving this because you enabled mailing list mode.

To unsubscribe from these emails, [click 
here](https://discuss.tvm.ai/email/unsubscribe/b7d3d33f91515366c9404b868dd24b0cc91ab46fe4661d52775baa9936f80bac).

Reply via email to