Updates transformers library and fixes missing Seq2SeqTrainerArguments

clcarwin / alpaca-weight

Train llama with lora on one 4090 and merge weight of lora to work as stanford alpaca.

MIT License

50 stars 4 forks source link

Closed bruvduroiu closed 1 year ago

bruvduroiu commented 1 year ago

Addresses:

https://github.com/clcarwin/alpaca-weight/issues/2
transformers changelogs changing LLaMATokenizer to LlamaTokenizer, etc.
pins transformers and peft libraries in requirements.txt to avoid feature drift in the future