Fire Video! ❤ Would love to see a rag+ system that can be safely kept at sync with changing data. For example, loading docs to rag is trivial, but if i want to change some value in a doc, its not trivial which chunks i can disregard and delete from the vdb before re-embedding... Let's say i have a stock, prices change, availability changes, etc. A Synced RAG is needed. Who's got this? Thanks & All the best
Thx for the great video, but i dont really understand the difference between advanced rag and modular rag. doesnt modular only provide ready-to-use modules for the identified processes in the advanced rag (isnt this just some kind of service collection for the ideas of the advanced rag) ?
Hello, excellent video. I'm curious about how the MLX framework demonstrates superiority over GGUF format, especially considering GGUF's capabilities in large language model applications. I am currently running Mixtral-8x7b and its clones at 8-bit precision on i512400 with a 128GB Linux box, offloading some layers to an RTX 4090 and 3090, and the performance is acceptable.
Fire Video! ❤
Would love to see a rag+ system that can be safely kept at sync with changing data.
For example, loading docs to rag is trivial, but if i want to change some value in a doc, its not trivial which chunks i can disregard and delete from the vdb before re-embedding... Let's say i have a stock, prices change, availability changes, etc. A Synced RAG is needed.
Who's got this? Thanks &
All the best
Thx for the great video, but i dont really understand the difference between advanced rag and modular rag. doesnt modular only provide ready-to-use modules for the identified processes in the advanced rag (isnt this just some kind of service collection for the ideas of the advanced rag) ?
Hello, excellent video. I'm curious about how the MLX framework demonstrates superiority over GGUF format, especially considering GGUF's capabilities in large language model applications. I am currently running Mixtral-8x7b and its clones at 8-bit precision on i512400 with a 128GB Linux box, offloading some layers to an RTX 4090 and 3090, and the performance is acceptable.
🎉🎉