{"slug": "base-or-instruct-version-of-llm-for-fine-tuning", "title": "Base or Instruct version of LLM for fine tuning?", "summary": "In a Q&A on fine-tuning large language models, Sahar advises using the Instruct version of models like Llama 3.1 8B, Gemma 2 9B, and Mistral 7B for most tasks, including multiple-text classification, because base versions lack prompting and require few-shot techniques. The choice depends on resources and desired behavior, but Instruct versions are generally recommended unless domain adaptation is needed.", "body_md": "Can I please ask if Base version or Instruct version is right choice for fine tuning LLM model?\n\nFor example, I use the LLM model for multiple text classification. Should I choose Llama3.1 8b or Llama3.1 8b Instruct?\n\nAnd can this be generalized to other models? Like for Gemma2 9b vs Gemma2 9b Instruct, Mistral 7b vs Mistral 7b Instruct, Mistral-Nemo-Base-2407 vs Mistral-Nemo-Instruct-2407 and so on…It’s not clear to me from the documentation…\n\nTL;DR Use the instructional version of the models.\n\nAs Always - It depends; Mostly on your resources and expectations.\n\nIn most cases, ppl who don’t require domain adaptation or significant differences in alignment - will want to work with the Instruct version, for LLaMA or any other LLM.\n\nFor example,\n\n- Prompting isn’t a thing on the base version, and you’ll need to use few-shot and other techniques to get the model to understand what you want.\n- Using Instruct versions allowing you to achieve almost anything with a prompt, but you get stuck in this “ping-pong UX”.\n\n**In your case, for multiple-text classification,**\n\n- Create a test set.\n- Start with prompting a model.\n- Do the same with different models.\n- Fine-tune the best-performing model. Test the FT model.\n\nGood luck.\n\n[Sahar](https://www.linkedin.com/in/sahar-millis/)\n\nWhether you fine-tune a base model or an instruct model depends on the behavior you want your final model to have and the structure of your data.\n\nSometimes, it makes more sense to start with a base model rather than further specializing a model that has already undergone instruction tuning. Other times, starting from an instruct model is better if you want to preserve its instruction-following behavior while adapting it to your specific task.", "url": "https://wpnews.pro/news/base-or-instruct-version-of-llm-for-fine-tuning", "canonical_source": "https://discuss.huggingface.co/t/base-or-instruct-version-of-llm-for-fine-tuning/100838#post_4", "published_at": "2026-08-31 19:06:03+00:00", "updated_at": "2026-08-31 19:23:46.254048+00:00", "lang": "en", "topics": ["large-language-models", "ai-tools"], "entities": ["Llama 3.1 8B", "Gemma 2 9B", "Mistral 7B", "Sahar"], "alternates": {"html": "https://wpnews.pro/news/base-or-instruct-version-of-llm-for-fine-tuning", "markdown": "https://wpnews.pro/news/base-or-instruct-version-of-llm-for-fine-tuning.md", "text": "https://wpnews.pro/news/base-or-instruct-version-of-llm-for-fine-tuning.txt", "jsonld": "https://wpnews.pro/news/base-or-instruct-version-of-llm-for-fine-tuning.jsonld"}}