🧠 BAU Small 1.5B (GGUF Q4_K_M)
BAU Small 1.5B (bau-small-1.5b.gguf) is a specialized on-device intelligence copilot model fine-tuned on Qwen2.5-1.5B-Instruct for African retail, pharmacy, supermarket, and wholesale SMEs.
🎯 Key Capabilities
- Tool Calling & Structured Retrieval: Generates precise
<tool_call>{"name": "queryStoreData", "arguments": {...}}</tool_call>commands to query live POS sales, inventory batches, expiring pharmaceuticals, cashier voids, and expense ledgers. - Generative UI JSON Cards: Synthesizes structured widget cards (
GENERATIVE_CHART,RED_FLAG_ALERT,SHIFT_SCHEDULE,AUTO_TASK) directly into its response. - African SME Financial Intelligence: Deep domain knowledge of Nigerian Naira (₦), GMROI, Cash Conversion Cycle (CCC), EOQ, blind cashier drawer reconciliation, and ATS cold-chain protocols.
- Ultra-Fast On-Device Footprint: Quantized in Q4_K_M (~941 MB) for sub-second offline inference on CPU & GPU (Electron, llama.cpp, node-llama-cpp, Ollama).
🚀 Quickstart with node-llama-cpp / llama.cpp
System Prompt
You are Neurons AI, an intelligent business copilot for African retail, wholesale, and pharmacy SMEs. You have access to the tool `queryStoreData(resource, query, filter)` to inspect sales, inventory, staff, expenses, invoices, and tasks. Always call tools when store numbers are needed. Use Naira (₦) for currency.
Direct CLI Inference
llama-cli -m bau-small-1.5b.gguf -p "<|im_start|>system\nYou are Neurons AI, an intelligent business copilot for African retail and SME stores.<|im_end|>\n<|im_start|>user\nShow me today's sales breakdown by payment method.<|im_end|>\n<|im_start|>assistant\n" -c 2048 --temp 0.3
Model Details
- Base Architecture: Qwen2.5-1.5B-Instruct
- Parameters: 1.54B (LoRA Rank 16, Alpha 32)
- Quantization: Q4_K_M (4-bit Medium)
- File Size: ~941 MB
- Context Length: 2048 tokens