Commit History
-
Switched safetensor base back to swe-dk-finetune-step-1000 (SOTA model)
Verified signature06a3857
-
Replaced EOT with end of text in accordance with LLM2Vec
Verified signature30a6c96 -
Changed EOS token to end_of_text to follow LLM2Vec
Verified signatureNot entirely sure if this is legit.. But I am guessing it is because we are finetuning not for a conv style where EOS is typically used, but for embedding purposes closer to the base model training where a sentence ends using EOS end of text. Not sure how much it matters.447ce2b -
Updated padding side to "left"
Verified signaturefc6f018 -
Model save
Verified signatureb2c679a -
Training in progress, step 3000
Verified signatured28e6a8 -
Training in progress, step 2500
Verified signaturedad4e05 -
Training in progress, step 2000
Verified signature457aed2 -
Training in progress, step 1500
Verified signatureb5b0df0 -
Training in progress, step 1000
Verified signature9e573ac -
Training in progress, step 500
Verified signature5d79455 -
Model save
Verified signaturea7fbb4a -
Training in progress, step 4
Verified signature12cb039 -
Training in progress, step 2
Verified signatureeaf4fcb
-
Updated EOS token to left padding and <end_of_text>
Verified signature5096219 -
tokenizer config from llama-swe last training
Verified signature3f022f7 -
Added custom config in line with llm2vec
Verified signature78b61c2
-
Add adapter config and model to main branch
d351b35 -
initial commit
Verified signature882b7c6