Downloads and installation

Application

The public chat page is a visual preview with no model backend. Download the application and the Gemma model to chat locally.

Download the TagaLLM application ZIP (includes the chat UI and local server) · GitHub source repository.

• Local setup and inference instructions
• Gemma system prompt and cultural references
• Visual preview of the chat interface

Research artifacts

Browse the TagaLLM collection on Hugging Face.

TagaLLM Gemma 4 E2B model card · download Q4_K_M GGUF (3.43 GB).

Original 2K tokenizer and 4K research tokenizer.

Scratch checkpoint inventory and hashes.

• TagaLLM 3.75M-256 · download weights
• TagaLLM 15.5M-256 · download weights
• TagaLLM 15.52M-512 · download weights

Dataset card on Hugging Face. Dataset provenance and historical split information. The redacted candidate is documented there. Dialogue files are not listed on the public Hugging Face main branch.

Application architecture

1. GitHub Pages serves the custom UI and saves conversation history in the visitor's browser.
2. For local chat, the UI forwards the current conversation to the local Gemma service.
3. Gemma 4 E2B generates the answer using Tagalog instructions and relevant cultural references. The evaluated research setup also used SEA-LION for language and scope checks.
4. The service returns the reply to the UI.