Moe_llama3_8bx2_Rag / mergekit_moe_config.yml
S-miguel's picture
Upload 12 files
0d3207d verified
base_model: X:/FOUNDATIONAL-OPEN-LLMS/Meta-Llama-3-8B-Instruct
gate_mode: cheap_embed
dtype: bfloat16
experts:
- source_model: X:/FOUNDATIONAL-OPEN-LLMS/Llama3-ChatQA-1.5-8B
positive_prompts:
- "This model specializes in general questions, vector database retrieval, and data queries."
- source_model: X:/FOUNDATIONAL-OPEN-LLMS/Meta-Llama-3-8B-Instruct
positive_prompts:
- "This model specializes in validating information, deep thinking, and assembling detailed high-grade responses."