Finetune a state-of-the-art code-focused LLM such as IBM Granite with the pairs “beginner” (prompt) and “content” (code), “average” (prompt) and “content” (code), and “expert” (prompt) and “content” (code) from the new dataset (~300k rows). Evaluate the resulting model by manually comparing its output with the base model and the other models finetuned with the new dataset.
danielbrdz
commented
3 days ago
Description
Finetune a state-of-the-art code-focused LLM such as IBM Granite with the pairs “beginner” (prompt) and “content” (code), “average” (prompt) and “content” (code), and “expert” (prompt) and “content” (code) from the new dataset (~300k rows). Evaluate the resulting model by manually comparing its output with the base model and the other models finetuned with the new dataset.