loubnabnl / santacoder-finetuning

Fine-tune SantaCoder for Code/Text Generation.
Apache License 2.0
179 stars 22 forks source link

Make FIM optional #11

Closed loubnabnl closed 1 year ago

loubnabnl commented 1 year ago

This PR by @Stillerman added support to FIM for santacoder fine-tuning:rocket: But since this code is used not just for SantaCoder this PR changes default fim rates to zero, and handles case where user specifies fim_rate>0 and tokenizer doesn't have fim tokens.

Stillerman commented 1 year ago

Great catch! Looks good