LeoLakshman/ai-human-text-detector

Fix README: correct LSTM/CNN file size, fix LLM download-timing contradiction, add live Space link

Fix stale TinyLlama comment in requirements.txt

Don't redundantly show 'Verdict: X' as the reason when the LLM gives no separate reason line

Update README for subprocess-isolation fix and smaller SmolLM2 explainer

Run LLM calls in isolated subprocesses (fixes TF/PyTorch in-process deadlock) and switch explainer to smaller SmolLM2-360M-Instruct (cuts total LLM size ~3.1GB -> ~1.7GB)

Pre-download LLM weights at Docker build time instead of on first live request

Revert to local LLM loading (remote Inference Providers requires paid third-party billing for chat models)

Pin Inference Providers client to featherless-ai instead of auto-routing

Update README for remote Inference Providers architecture

Switch LLM cross-check + explanation to HF Inference Providers API (remote) instead of local model loading

Revert legacy-Keras loader attempt; generically strip stray quantization_config from layer configs instead

Fix LSTM/CNN .h5 loading by forcing legacy Keras 2 H5 loader

Project 2: add Qwen2.5-0.5B-Instruct + TinyLlama-1.1B-Chat LLM integration