How to design a storage and search system for 10,000 Telegram channels with 300,000 posts per day with latency less than 2 seconds?
Middle+
192
Machine Learning / AI
3611
0
If the worker is synchronous, will it have batching?
Middle+
188
Can you consider embeddings synchronously directly in the API endpoint? What will happen if you process embeddings for a large stream of documents in a FastAPI endpoint?
Middle+
176
How to design a system for automatic content generation for Telegram channels in the style of the channel's author?
Middle+
170
When to choose LoRA and when to retrieve by similar posts? How much data is needed for each approach?
Middle+
164
How did you validate LoRA fine-tuning?
Middle+
160
Have you used neural networks as a quality assessment metric?
Middle+
159
What rank did you set for LoRA and what does it depend on?
Middle+
156
Tell us about your experience with RAG systems and what you did in your last project.
Middle+
156
Did you do all this yourself (RAG, LoRA, working with neural networks) or did you help someone?
Middle+
152
In which layers did you insert LoRA and what is q_proj?
Middle+
143
Have you deployed neural networks on your servers or in the cloud?
Middle+
143
What metrics did you use to evaluate system quality and how did you achieve a 40% improvement?
Middle+
142
Why can LoRA overfit even with a small number of parameters? Does the choice of rank affect overfitting?
Middle+
134
Why is a vector database needed in this architecture?
Middle+
133