convaiinnovations/laya-demo

fix: work on laya 0.3.2 too (PyPI index had not propagated 0.3.3)

fix: work on laya 0.3.2 too (PyPI index had not propagated 0.3.3)

fix: work on laya 0.3.2 too (PyPI index had not propagated 0.3.3)

perf: preload routing at boot, sharing the checkpoint the Space already builds

perf: preload routing at boot, sharing the checkpoint the Space already builds

perf: preload routing at boot, sharing the checkpoint the Space already builds

fix: import spaces before torch so ZeroGPU can attach a GPU

fix: fall back to CPU when ZeroGPU cannot attach a GPU

fix: routing tab runs on CPU, no background thread, no GPU fork boundary

fix: routing tab runs on CPU, no background thread, no GPU fork boundary

fix: warm checkpoints on a background thread so start-up is never blocked

fix: warm checkpoints on a background thread so start-up is never blocked

perf: preload all checkpoints at boot so routing never pays a model load

perf: preload all checkpoints at boot so routing never pays a model load

perf: preload all checkpoints at boot so routing never pays a model load

feat: multilingual routing tab (additive, existing tabs untouched)

feat: multilingual routing tab (additive, existing tabs untouched)

feat: multilingual routing tab (additive, existing tabs untouched)

feat: multilingual routing tab (additive, existing tabs untouched)

Remove credit card example from guardrails tab

Fix model routing thresholds and add domain criteria descriptions

Rename space branding to Laya Demo

Sort answers table by risk/severity descending across all themes

Moderation: explicit rubric levels, threat question, severity composed in code

One shared GPU entry point, batched RAG scoring, per-call timing in logs

Preload and warm the model at start-up so the first request is fast

RL Agent demo: triage, email, guardrails, RAG filtering, moderation, routing, playground

initial commit