firstly I finetuned a NousResearch/CodeLlama-7b-hf on my custom dataset, using PEFT, Qlora, and trl SFT trainer.
secondly here is the training and evaluation loss.
0.58 after 5 epochs
After training, i push the model to huggingface_hub and these are the files which is uploaded to huggingface_hub:
After that i loaded the fine-tuned model from the hugging face and tested for some queries, when i test it gives not the correct response, so I fine-tuned my model on Pinescript programming language and it gave the response in Python language (I also gives the system message that you must gives an answer in pine script language)
so how to solve this problem please give me any idea.
and also check the files in the above screenshot that i have adapter.satetensor file and adaptor_config,json file and i haven’t the adapter.bin file in my fine-tuned model,