chore: suppress Qwen config warnings in greedy generation

swap: Qwen2.5-0.5B-Instruct (faster, reliable #### format) + fix dp2 detection bug

feat: pre-load model at startup to avoid 200s first-request wait

fix: handle None gold_prob/prod_prob in steps display

chore: gitignore __pycache__

fix: context-aware Step token detection for TinyLlama (handle multiple token IDs)

Restore top_p_presence_next field in TimestepArtifacts to match research code

Fix: remove top_p_presence_next arg not in TimestepArtifacts

Pin huggingface_hub<0.26.0 for gradio 5 compat

Fix Python 3.13 compat: add audioop-lts for gradio, bump to gradio 5

Switch to TinyLlama-1.1B for CPU, remove bitsandbytes

HF Space: Reasoning Trajectory Demo