A fine-tuned Gemma 3 1B (INT4) for context-aware text prediction on Android, quantized to GGUF Q4 K M for on-device inference via llama.cpp. Model Details Base model: google/gemma-3-1b-pt Fine-tuning: LoRA via Unsloth on a single RTX 3060 Quantization: Q4 K M via llama.cpp convert hf to gguf.py Target device: Android (HeliBoard keyboard fork) Training Data
--- license: apache-2.0 base_model: google/gemma-3-1b-pt tags: - gemma - keyboard - on-device - android - gguf - lora - unsloth language: - en --- # Gemma 3 1B — Context-Aware Keyboard (Q4_K_M GGUF) A fine-tuned Gemma 3 1B (INT4) for context-aware text prediction on Android, quantized to GGUF Q4_K_M for on-device inference via llama.cpp. ## Model Details - **Base model:** [google/gemma-3-1b-pt](https://huggingface.co/google/gemma-3-1b-pt) - **Fine-tuning:** LoRA via Unsloth on a single RTX 3060 - **Quantization:** Q4_K_M via llama.cpp `convert_hf_to_gguf.py` - **Target device:** Android (HeliBoard keyboard fork) ## Training Data ~3B token conversational corpus from five sources with per-source cleaning, thread reconstruction, and synthetic typo augmentation: - Reddit conversations - OpenSubtitles - Enron email corpus - Hacker News comments - MS MARCO queries ## Prompt Format The model uses custom special tokens to condition predictions on app context: ``` messaging [on-screen conversation history] ``` - ` ` — app category (messaging, email, search, social, etc.) - ` ` — marks start of on-screen conversational context - ` ` — marks the cursor position; model predicts the next tokens ## Evaluation | Metric | Gemma (LLM) | N-gram baseline | |--------|-------------|-----------------| | MRR (top-5) | 0.3772 | 0.0509 | | Hit@5 | 52.4% | 9.9% | | Win rate | 92.2% | 4.5% | Evaluated on 1,865 examples. ## Usage Designed for integration with [HeliBoard](https://github.com/Helium314/HeliBoard) via llama.cpp's Android NDK build. Load the GGUF, feed the structured prompt, and use top-k token predictions as keyboard suggestions. ## Limitations - English-only - Trained on informal/conversational text; may underperform on formal or technical writing - ~800MB model size requires download on first app launch
A fine-tuned Gemma 3 1B (INT4) for context-aware text prediction on Android, quantized to GGUF Q4 K M for on-device inference via llama.cpp. Model Details Base model: google/gemma-3-1b-pt Fine-tuning: LoRA via Unsloth on a single RTX 3060 Quantization: Q4 K…
