-
-
InfluxDB
Power Real-Time Data Analytics at Scale. Get real-time insights from all types of time series data with InfluxDB. Ingest, query, and analyze billions of data points in real-time with unbounded cardinality.
Hello everyone, new to LLMs. I am working on my thesis project. The whole idea is to create a mixed reality voice assistant that can control some devices in a room and you can have with it a more intelligent conversation compared to other voice assistants(Alexa,Google, etc.). I thought initially to use wit.ai for the extraction of commands and if it's not a recognized command to send a request to a chatgpt API. Then I realized that wit.ai wasn't pretty accurate when I added multiple intents, for example distinguishing creating list from adding item to list or removing item from removing reminder etc, even though I added enough training data. Sometimes it identifies the entities correctly, but the intent is wrong even though there are no intents with these entities in the existing data. If user does not specify correctly a command I will have to handle a lot of scenarios like not specifying time or date or both and asking user to give them again and managing to extract the values correctly after. I thought that chatgpt could actually do both if I give some instructions with the formats of prompts and responses I want. I tried it in the web version and seemed to do what I want. Then I realized how expensive it is to use their API just to have context on a conversation. So my next thought was to use a local llm. So could someone recommend me a small model that can be used for such case or what adjustments should I do in order to work? In my university I can use a machine with 32GΒ Ram and Nvidia Α4000 GPU. Thank you :)