A user is seeking guidance on how to transform a Qwen9B model, which has undergone continued pre-training (CPT) on Polish legal data, into a functional chat assistant capable of reasoning and function calls. The user has completed three CPT stages focusing on legal commentaries, legislation, and court decisions, and has a variety of other data sources including Q&A forums, study materials, and existing harness conversations. The primary questions revolve around the optimal next steps, specifically whether to continue CPT or move to Supervised Fine-Tuning (SFT), and how to best utilize the available data for reasoning and tool use, considering the roles of SFT and Reinforcement Learning (RL). AI
IMPACT Guidance on fine-tuning and data utilization for domain-specific LLMs.
RANK_REASON User-generated question seeking technical advice on model development.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
