Skip to content

[TFLite] TFLite FP16 Post Quantization Support #5823

Description

@FrozenGene

TensorFlow Lite now supports converting weights to 16-bit floating point values during model conversion from TensorFlow to TensorFlow Lite's flat buffer format. This results in a 2x reduction in model size.

However, this will insert new dequantize for ops (like conv2d) used for dequantize fp16 weight to fp32. Like this:
image

TVM doesn't support this behavior. List the things we mainly should to do:

  • Support float16 type inside tflite parser
  • Extend dequantize to support fp16 to fp32

Related issue:#5774

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions