Error while using cache branch in profiling models

[MatteoFasulo]

Hi,
I was trying to profile the Whisper decoder model (tiny) using the ORT Web Perf tool. However, when I run the profiler it gives me an error: Error: input 'cache_position' is missing in 'feeds'.

The following code is the reference in ort-perf.html:

if (gen == "whisper-decoder") {
    feed["input_ids"] = fillTensor([1, seqlen], "int64", 1n);
    feed["encoder_hidden_states"] = fillTensor([1, 1500, 384], "float32");
    // past_key_values.0.encoder.value[FLOAT, batch_sizex6xencoder_sequence_length_outx64]
    // past_key_values.1.decoder.key[FLOAT, batch_sizex6xpast_decoder_sequence_lengthx64]
    const decoder_shape = [1, 6, seqlen, 64];
    const encoder_shape = [1, 6, 1500, 64];
    for (var k in inputNames) {
        const v = inputNames[k];
        if (v.startsWith("past_key_values.")) {
            if (v.includes("decoder")) {
                feed[v] = fillTensor(decoder_shape, "float32", 1);
            } else if (v.includes("encoder")) {
                feed[v] = fillTensor(encoder_shape, "float32", 1);
            }
        }
    }
    feed['use_cache_branch'] = new ort.Tensor("bool", [0], [1]);
    return feed;
}

Removing the following line seems to work:

// feed['use_cache_branch'] = new ort.Tensor("bool", [0], [1]);