Refresh progress every 5 tokens to reduce bar flicker

Hide token stream from Output box; show complete text only at end (no flicker)

Yield placeholders between progress stages so the bar actually moves

Real progress: track HF download tqdm + update bar during token streaming

Tune llama-cpp-python for HF Spaces 2-vCPU: n_ctx=512, n_threads=2, n_batch=8

Trigger rebuild to pick up merges-corrected GGUFs

Switch large models to GGUF via llama-cpp-python 0.2.62

Stream output token-by-token + add progress bar for model load

Drop GGUF path, keep bf16 + preload + new model entries

Pin llama-cpp-python to direct manylinux wheel URL (0.2.62 cp311)

Pin llama-cpp-python==0.2.90 with --prefer-binary for cp311 wheels

Use abetlen pre-built CPU wheels for llama-cpp-python

Pin Python to 3.11 for llama-cpp-python pre-built wheels

Add 3B-Chat-v2, switch large models to GGUF, preload default

Switch 1B model to s2_hf loader (use HF transformers + safetensors)

Force Space rebuild - verify 2B model deployment

Fix 1B model ID: gpuburnout-1b -> GPUburnout-1B-160K (correct HF repo name)

Add accelerate dependency for device_map support

Update Gradio SDK version to 5.50.0 for build compatibility

Trigger Space rebuild

Add GPUburnout-2B model and update subtitle to 2 billion parameters

Thicker borders (2px) on all boxes

Add visible borders to all boxes for better contrast on dark bg

Fix dark theme - remove invalid input_text_color variable

Use Gradio theme system for dark mode instead of CSS overrides

Force dark theme with pure CSS overrides (#08080d)

Add weights_only=True to S1 torch.load for safety

Dark theme matching blog background (#08080d)

Fix 1B description: 11.8B tokens trained (not 30B dataset size)

Rename all models to GPUburnout-* branding, default to 134M for faster first run

Rename Llama 1B to GPUburnout-1B to match blog branding

Use CPU-only PyTorch to speed up Space startup

Add hf_hub_download for 1B model weights

feat: add Tiny (3.2M) and GPT-2 Small (134M) model weights

feat: add app code, configs, and tokenizers

initial commit