Add a skycastle job for ExecuTorch running llama
Summary: As titled. This CI job covers the e2e workflow: 1. Download language_llama checkpoint, params.json and tokenizer. 2. Export *.pte file using `export_llama`. 3. Generate tokenizer from the tokenizer artifact using `tokenizer.py` 4. Run the pte file using `llama2/runner.cpp` 5. Verify the result is correct. For starter, focus on fp32 + kv cache ~~+ xnnpack~~ combo (xnnpack runs into out of memory issue rn :() Reviewed By: JacobSzwejbka Differential Revision: D53580453 fbshipit-source-id: fd2b91808f436852092bd87deba00a0d6270fdc1
M
Mengwei Liu committed
c3c198967e862b9f22a0448df27712509200ea43
Parent: 3164688
Committed by Facebook GitHub Bot <facebook-github-bot@users.noreply.github.com>
on 2/9/2024, 10:01:24 AM