Spaces:
Paused
Paused
Commit History
-
feat: HF Inference API + cooldowns (v0.5)
Hermes Bot +1: NanoboozhooReplace local llama-cpp-python GGUF with serverless HF Inference API. New shared/inference_client.py with project-scoped cooldowns to protect credit budget (6s/10s/12s for tinybard/focusfriend/crittercalm). Build fixes: - Removed llama-cpp-python from requirements (was causing BUILD_ERROR on HF Spaces due to the heavy C++ compile on CPU tier). Iteration: - Models: Qwen/Qwen2.5-1.5B-Instruct default (1.5B sweet spot, free tier). Override via INFERENCE_MODEL, or per-project TINYBARD_MODEL / FOCUSFRIEND_MODEL / CRITTERCALM_MODEL. - Always-fallback: every LLM call falls back to procedural/template output if inference fails or is in cooldown. UX never blocks. - TinyBard footer polls /api/model_status every 2s to show model name + cooldown countdown. HF Spaces are the primary dev/test surface; this commit is what the Spaces will pick up on next deploy. Co-Authored-By: Nanoboozhoo <nanoboozhoo@git-aaaki>e4b2617
-
Initial release: CritterCalm — AI Voice Cloning Animal Soother
Hermes Bot +1: NanoboozhooClone your voice, then generate calming audio sessions for your pets. - OmniVoice (0.6B) zero-shot voice cloning from 3-10 sec audio - Dolphin-X1-8B script generation with 48 pre-written template fallback - Kokoro TTS (82M) built-in soothing voices - 3-tab Gradio interface: Clone Voice, Calm Your Pet, About & Science - 6 animal types × 8 situations, fully local, no cloud APIs Built for Build Small Hackathon 2026 — Backyard AI track. Badges: Off the Grid, Well-Tuned, Field Notes, Sharing is Caring Co-Authored-By: Nanoboozhoo <nanoboozhoo@git-aaaki>144ed35