Original Paper: arXiv
| Code Source | Link | License |
|---|---|---|
| Original Source | GitHub | Apache V2.0 (LICENSE.google.bert) |
| Reference Source | HuggingFace | Apache V2.0 (LICENSE.generic.apache2) |
Extracting weights from the reference (requires HF transformers):
$ python -m bert.ref.extract_weights| Example | Description |
|---|---|
| SQuAD Inference | Runs a question-answer inference on a predefined context and question |
| Single Forward | Runs a single forward pass of batch size 1, seqlen 512 through the encoder |
How to run:
# SQuAD Inference
$ python -m bert.examples.squad_infer
Question: What is the focus of this paper?
My best guess:
isa - aware mapping problem
# Single Forward Pass
$ python -m bert.examples.single_fwdLevel 3: The model in this repository has been verified to produce numerically exact output given identical input and weight values.