docs: update model sizes in README

docs: add demo link and social post section to README

docs: add 'offgrid' achievement to README

feat: add dataset upload script, evaluation pipeline, and fine-tuning support

refactor: add Llama singleton caching and remove guarded beat stream logic

docs: simplify README short description

docs: expand README with detailed app overview and usage guide

feat: enhance Llama.cpp mode with styled disclaimer

feat: add support for Llama.cpp mode with GGUF inference

refactor: implement guarded beat generation to ensure proper Llama cleanup

refactor: optimize Llama instance reuse and remove redundant GC calls

refactor: improve JSON extraction and GGUF model loading flow

chore: update Python and PyTorch versions, add CUDA index URL

fix: update LD_LIBRARY_PATH handling to ensure llama.cpp resolves CUDA libraries with PyTorch 2.x

fix: ensure llama.cpp loads cudart globally to resolve linking issues with PyTorch libraries

chore: remove unused index URLs from requirements.txt

refactor: replace GGUF loading implementation with streamlined handling and contextual loading

feat: add support for GGUF via llama.cpp and anthropic integration

fix: specify "no text" requirement in illustration prompts to avoid label or lettering issues

feat: increase max token limit for story generation to 1024

fix: include raw skeleton input in parse failure logs for better debugging

feat: refine prompts for unique character names and enhance story generation sampling behavior

fix: restrict choice visibility after max beats

fix: ensure consistent visibility updates for narration-related UI elements

feat: enhance LLM call tracing and data export for SFT training

feat: implement user-triggered narration and decouple audio generation

feat: add story trace building and asynchronous push support

feat: add character appearance support for consistent visual references across illustrations

feat: add narrative skeleton generation for interactive stories

feat: rename from story sprout to storygrove

feat: integrate image generation with visual profile consistency and custom seeds

feat: add parallel image and audio generation with VRAM-aware handling

chore: image lands before narration

feat: book-style UI with cover page and alternating page layou

feat: add FLUX.2-klein-4B image generation per story beat

fix: Remove unnecessary package dependencies from requirements.txt for hf spaces compatibility

Merge pull request #1 from TheMLChannelAI/feat/add_voice

feat: Add narrator and VRAM management modules for story audio creation and efficient model handling.

feat: Interactive kids story generator using gradio and local LLM loaded via transformers library.