I'm working on a domain-specific NLP task and noticed catastrophic forgetting when applying LoRA adapters. Has anyone found a reliable checkpoint strategy or loss weighting approach...
The current streaming API is fantastic, but adding direct speech-to-speech translation would save massive latency for our customer support bot. Would love to see WebSocket support for audio...
Sharing our architecture using NexusAI's AutoML pipeline + Kafka. We reduced false positives by 62% and cut inference costs by half using the new batch prediction endpoints. Diagram inside...
The new context window is huge, but I'm seeing degraded output quality when stuffing >80K tokens. Are there any official guidelines on chunking, system prompt placement, or temperature tuning...
We're opening early access to the improved AutoML workflow with built-in drift detection and one-click rollback. Sign up below to get an invite link. Looking for feedback on the UI...
Getting 429 errors at ~800 RPM even though docs say 1200 RPM is allowed. Happens mostly on the EU-West endpoint. Anyone else experiencing this or is it a regional throttling issue...