YouLearn-Granite4-H1B-V11-GGUF

YouLearn Granite 4.0 H 1B V11 - GGUF

Fine-tuned IBM Granite 4.0 H 1B (Hybrid Mamba-2/Transformer architecture) for the YouLearn offline study assistant Android app. Quantized to Q4_K_M for mobile deployment.

Model Details

• Base Model: ibm-granite/granite-4.0-h-1b-instruct
• Architecture: Hybrid Mamba-2 / Transformer (Extremely low KV-Cache pressure for long documents)
• Fine-Tuning: YouLearn V11 instructional dataset, 564 training steps
• Quantization: Q4_K_M (4-bit medium)
• Primary Use Case: Standard/Premium tier Android devices — long document QA, structured outputs, and learning tasks.

Purpose

This model is intended to provide superior long-context handling compared to pure transformer models (like MiniCPM5) on mobile devices. The Mamba architecture prevents memory explosion when users upload large PDFs or notes.

Capabilities

• Advanced study Q&A and tutoring
• Document-grounded answers (RAG) with low RAM usage
• Structured output (key points, flashcards, mind maps)
• Strong instruction following (JSON/Mermaid formats)

Usage (llama.cpp)

Note: Ensure your llama.cpp version supports the Mamba-2 hybrid architecture.

./llama-cli -m YouLearn-Granite4-H1B-V11-FT564-Q4_K_M.gguf \
  -p "<|start_of_role|>system<|end_of_role|>YouLearn is a private offline study tutor.<|end_of_text|><|start_of_role|>user<|end_of_role|>Explain photosynthesis.<|end_of_text|><|start_of_role|>assistant<|end_of_role|>" \
  -n 512 \
  --top-p 0.85 --temp 0.35 --min-p 0.05