Can I compress a IR model from FP32 to FP16?

[rrobin]

I have models in FP32 that I would like to compress to FP16. Is there a way to do this?

[Zulkifli_Intel]

Hello Robin,

Thank you for reaching out to us.

Model Optimizer can convert all floating-point weights to FP16 data type. To compress the model, use the --data_type option:

mo --input_model INPUT_MODEL --data_type FP16

Sincerely,

Zulkifli 

[rrobin]

Thanks for the answer. Yes I know about that option. But o would like to convert the IR model. I don't have access to the models previous format, only the IR.

[Zulkifli_Intel]

Hello Robin,

Sorry for the misinterpretation, I'm looking into this and will get back to you soon.

Sincerely,

Zulkifli

[Zulkifli_Intel · Accepted solution]

Hello Robin,

It is not possible to further transform an IR precision from FP32 to FP16 at the moment. However, our developer is working to enable this feature in future releases. Please refer to the Release Notes for updates.

Sincerely,

Zulkifli 

[Zulkifli_Intel]

Thank you for your question. If you need any additional information from Intel, please submit a new question as this thread is no longer being monitored.

Sincerely,

Zulkifli