README.md

metadata

language: en
tags:
  - Android
  - tflite
  - blenderbot
license: apache-2.0

Model Card

blenderbot-small-tflite is a tflite version of blenderbot-small-90M I converted for my UTA CSE3310 class. See the repo at https://github.com/kmosoti/DesparadosAEYE and the conversion process here.

You have to right pad your user and model input integers to make them [32,]-shaped. Then indicate te true length with the 3rd and 4th params.

display(interpreter.get_input_details())
display(interpreter.get_output_details())

[{'dtype': numpy.int32,
  'index': 0,
  'name': 'input_tokens',
  'quantization': (0.0, 0),
  'quantization_parameters': {'quantized_dimension': 0,
   'scales': array([], dtype=float32),
   'zero_points': array([], dtype=int32)},
  'shape': array([32], dtype=int32),
  'shape_signature': array([32], dtype=int32),
  'sparsity_parameters': {}},
 {'dtype': numpy.int32,
  'index': 1,
  'name': 'decoder_input_tokens',
  'quantization': (0.0, 0),
  'quantization_parameters': {'quantized_dimension': 0,
   'scales': array([], dtype=float32),
   'zero_points': array([], dtype=int32)},
  'shape': array([32], dtype=int32),
  'shape_signature': array([32], dtype=int32),
  'sparsity_parameters': {}},
 {'dtype': numpy.int32,
  'index': 2,
  'name': 'input_len',
  'quantization': (0.0, 0),
  'quantization_parameters': {'quantized_dimension': 0,
   'scales': array([], dtype=float32),
   'zero_points': array([], dtype=int32)},
  'shape': array([], dtype=int32),
  'shape_signature': array([], dtype=int32),
  'sparsity_parameters': {}},
 {'dtype': numpy.int32,
  'index': 3,
  'name': 'decoder_input_len',
  'quantization': (0.0, 0),
  'quantization_parameters': {'quantized_dimension': 0,
   'scales': array([], dtype=float32),
   'zero_points': array([], dtype=int32)},
  'shape': array([], dtype=int32),
  'shape_signature': array([], dtype=int32),
  'sparsity_parameters': {}}]
[{'dtype': numpy.int32,
  'index': 3112,
  'name': 'Identity',
  'quantization': (0.0, 0),
  'quantization_parameters': {'quantized_dimension': 0,
   'scales': array([], dtype=float32),
   'zero_points': array([], dtype=int32)},
  'shape': array([], dtype=int32),
  'shape_signature': array([], dtype=int32),
  'sparsity_parameters': {}}]