[Good First Issue][TFL FE]: Support GELU operation for TensorFlow Lite

[e-nugmanova]

Context

OpenVINO component responsible for support of TensorFlow Lite models is called as TensorFlow Lite Frontend (TFL FE). TFL FE converts a model represented in TensorFlow Lite opset to a model in OpenVINO opset.

TensorFlow Lite opset is specified via

• reference implementation: link
• attributes definition in schema: link

What needs to be done?

For GELU operation support, you need to implement:

• common gelu loader directory link and register it in common operation table.

OutputVector translate_gelu_op(const NodeContext& node) {
    default_op_checks(node, 1, {"GELU"});
    auto x = node.get_input(0);

    // update these lines for best translation
    auto approximate = node.get_attribute<element::boolean>("approximate");
    auto res = make_shared<v7::Gelu>(x, ...);
    //

    set_node_name(node.get_name(), res);
    return res->outputs();
}

• TFL specific loader directory link

OutputVector gelu(const ov::frontend::tensorflow_lite::NodeContext& node) {
    const auto& decoder = get_decoder(node);
    std::map<std::string, ov::Any> attrs{
        {"approximate", decoder->get_attribute(&GeluOptions::approximate)},
    };
    return attribute_helper(node, attrs, ov::frontend::tensorflow::op::translate_gelu_op);
}

• register loader into the dictionary of Loaders: header, source.

  • header: TFL_OP_CONVERTER(gelu);
  • source: {"GELU", DEQUANTIZE_INPUTS(gelu)},

• implement layer test for your solution: link

One loader is responsible for conversion (or decomposition) of one type of TensorFlow Lite operation.

Useful links for TFL GELU support:

• TFL Reference implementation
• TFL Attributes
• OpenVINO operation to use link
• OpenVINO Gelu specification link

Resources

• Contribution guide - start here!
• Intel DevHub Discord channel - engage in discussions, ask questions and talk to OpenVINO developers

Contact points

@jane-intel

[kartik912]

.take

[github-actions[bot]]

Thank you for looking into this issue! Please let us know if you have any questions or require any help.

[koookieee]

.take

[github-actions[bot]]

Thanks for being interested in this issue. It looks like this ticket is already assigned to a contributor. Please communicate with the assigned contributor to confirm the status of the issue.

[rkazants]

Hi @kartik912,

Please share your update on this task.

Best regards, Roman

[p-wysocki]

I'm reopening the issue due to current assignee's inactivity.

[Shubham-Sahoo]

Hi @p-wysocki ! Would love to contribute to this issue. Can you assign it to me?

[mlukasze]

so it's yours, have fun :)