nmarafo/local-gemma4-rag

Update Ollama help card to include Hugging Face PNA instructions

Add local Ollama backend and Gemma 4 12B MTP support

Add hover tooltips to source citations

Add version indicator v1.2.0 to UI headers

Optimize Orama similarity, WebGPU dtype, ONNX CDN paths, and parallelize PDF text extraction

fix: remove similarity threshold from Orama vector search to allow non-semantic summary queries to retrieve context

feat: add visual progress indicator for document vectorization to provide immediate UI feedback

feat: add capability to rename and delete documents from the RAG panel

fix: inject RAG context into user prompt instead of system prompt to satisfy Gemma 4 instruction tuning

fix: improve system prompt to prevent model from requesting document when context is provided

fix: make the standalone html the primary entrypoint for huggingface

feat: integrate structural RAG from desktop app to standalone html

Fix: replaced blocked native alert with custom showError modal & fixed RAG hanging on error

UI/UX: Major overhaul - Ported Desktop App system (Workspaces, Structural RAG, Premium Citations)

Deployment: set standalone multimodal version as official index.html for HF Space

Docs: add gemma-4-E4B to models and update title

UI/UX: Premium Multimodal Standalone redesign with Vision & Audio support

Fix: standalone worker hanging and character escaping

Stability: explicit WebGPU memory cleanup on page reload

Fix: robust reasoning filter to prevent UI hang in main app

UI: added Clear Chat and Copy Output functionality

Feature: hybrid RAG/general knowledge and document deletion support

Feature: dynamic model selection (Gemma 4 E2B vs E4B) with VRAM warnings

Feature: add full Markdown support with code syntax highlighting (Highlight.js) and custom CSS

Docs: move models metadata to frontmatter

Docs: update model metadata in README

UI: add creator credits and social links for Norberto Martín Afonso

Fix: use official Chat Template via apply_chat_template to prevent prompt echoing

Fix: implement Gemma 4 reasoning filtering, turn format, and stop tokens

Perf: optimize WebGPU speed with warmup and greedy decoding, plus show-sources UI

Feature: add PDF support, premium Dark UI with document list, and fix RAG prompt echo

Fix: update Orama search syntax to v2 required format

Update: upgrade to Transformers.js v4.0.1 with worker/library cache busting (v401) to fix Gemma 4 support

Fix: upload full set of ONNX Runtime v1.24.3 modules including asyncify to resolve all fetch errors

Upgrade: migrate to Transformers.js v4.0.0 and ONNX Runtime 1.22.0 for native Gemma 4 support

Fix: update local WASM files to 1.22.0-dev to match Transformers.js 3.8.1 requirement

Update: upgrade Transformers.js to v3.8.1 for native Gemma 4 architecture support

Fix: use matching ONNX Runtime dev version for Transformers.js 3.3.0 to resolve t.getValue error

Fix: use absolute URL for local WASM modules to ensure correct fetch in workers

Fix: move WASM modules to local hosting to bypass CDN fetch issues

Fix: use explicit mapping for verified stable WASM modules (v1.20.1)

Fix: point to onnxruntime-web CDN for missing WASM modules

Fix: use ONNX version of Gemma 4 for Transformers.js compatibility

Complete Local RAG implementation with Gemma 4 and WebGPU

Merge from Hugging Face with our changes

Initial sync with Antigravity footer

Add files via upload

Update index.html

Update README.md

Update README.md