Training - Nexus Knowledge Base

Training Approach: - LoRA (Low-Rank Adaptation) for efficient fine-tuning - PEFT library (not Unsloth - compatibility issues) - JSON datasets: [{"prompt": "...", "response": "..."}]

Key Datasets: - lars_identity.json - Core identity and personality - nexus_knowledge.json - Nexus system understanding - tool_use.json - MCP tool patterns

Training Command:

python ~/corlera-training/scripts/train.py \
  --model /data/models/huggingface/qwen2.5-7b-abliterated \
  --dataset ~/corlera-training/datasets/lars_identity.json \
  --output ~/corlera-training/outputs/lars-lora

Track Projects: - LARS Training System (0dd041be) - Corlera Training System (dc4eeb11) - Training MCP Server (a6a12130)

Contents