liuwei-git

add phi3 support

3

Make phi3 as an explicit model to support in llama.

Add phi3 128K model support

1

The only difference between phi3 4k and 128k model is from the rotary embedding. 128k model adds long/short rope scaling factors (freq_factors) and an attn factor to each hidden dimension....

### Issue to fix: The static OnnxLoader::loader object may be destroyed by the system before the customer's cleanup code (e.g., TRITONSERVER_ServerDelete) runs. This premature destruction causes an assertion failure inside...