Implement runner for phi-3-mini (#4500)
Summary: This PR mainly does the following things: - implement runner for phi-3-mini - replace phi-3-mini to use bpe tokenizer, which is shared for all LLMs now - fix a small bug in bpe tokenizer Pull Request resolved: https://github.com/pytorch/executorch/pull/4500 Test Plan: ``` ./build/phi_3_mini_runner --model_path phi-3-mini-kv-128.pte --tokenizer_path tokenizer.bin --prompt "Tell me a story" --temperature 0 Prefilling tokens ... 24948 592 263 5828 Generating tokens ... about a time when you had to overcome a challenge. I remember when I was in high school, I had to prepare for a big exam that would determine my future. I had to study hard, but I also had to balance my schoolwork, my hobbies, and my social life. It was not easy, but I managed to do it. I made a study schedule, set goals, and rewarded myself for my achievements. I also asked for help from my teachers, friends, and family. I faced many difficulties, but I never gave up. I passed the exam with flying colors and``` Reviewed By: larryliu0820 Differential Revision: D60609165 Pulled By: helunwencser fbshipit-source-id: 9abab0ba8ea8e50559272c6001fa868e49f40a96
L
Lunwen He committed
864e0b0a3c1dcac1bf14937ea64030118203248c
Parent: 14c2473
Committed by Facebook GitHub Bot <facebook-github-bot@users.noreply.github.com>
on 8/5/2024, 5:54:13 PM