Support
This is a CS224N course project, not a production service — but issues and questions are genuinely welcome.
Why does it say “Offline demo” instead of “Live model”?
No Hugging Face API token is configured for this deployment, so the API route returns a small set of canned sample responses instead of calling the model. See the Docs page for how to add a token.
Is this using the actual fine-tuned model from the notebook?
Not by default — that checkpoint lives only in the training notebook's session and isn't published anywhere the live site can reach. The playground calls the public base model, surajp/gpt2-hindi, with the same decoding logic. See the About page for the full explanation.
Predictions look slow the first time I try them.
Hugging Face's hosted Inference API spins model instances down when idle, so the first request after a quiet period can take several seconds to “warm up.” Subsequent requests are much faster.
Can I use this for languages other than Hindi?
The tokenizer and fine-tuning corpus are Hindi-specific, so results for other languages will be unreliable. Swapping HF_MODEL_ID to a different base model would need matching changes to the corpus and preprocessing.