Fix the issues with latest transformers, add previously removed function to compute usable past KV length for cache compatibility.
#13
by
Great-Luso
- opened
The code to fix the issue is copied from another fix pull request: https://huggingface.co/it-just-works/stella_en_1.5B_v5_bf16/commit/03aedd040580357ec688f3467f1109af5e053249