Error creating collection.
mxcoldhit opened this issue · comments
I have got this somewhat working locally now using the oobabooga api with a local model and using Chroma for storage. I am getting an error when it restarts the while loop of main.py because it is trying to make a collection called tasks again. Chroma is throwing an error that says collection with name tasks already exists. Have you run into this issue? I can delete the index and start the process over from the beginning just fine and it will create the collection but I will just run into the error again. Im still looking into the solution. It looks like it might be on the chromadb side.
HI @mxcoldhit ,
It sounds like it's not properly deleting the tasks table before updating. I would need to have hands on to determine exactly what's going on, but you can add some print() functions to the agents modules to help determine why this is occurring.
I suspect the source of the error is the Prioritization Agent. We delete the Tasks table before loading the new tasks. This is to prevent building a large table of every task list created. If you have enabled persistent storage with Chroma, that could be the cause as well. We have not tested persistent storage yet.
Keep in mind, this project is moving. We have a new update coming in a few hours that may break some of the changes or functionality that you added. I would be interested to hear details about how you got Oobabooga API working. Was it an update on their end, or did you change something in our code to make it work. Also, which LLM are you using with Ooobabooga?
Thanks for the response @DataBassGit . I will look into the agents and check to see if persistent storage is enabled. I updated your code. I had to edit the oobabooga api and the if statements in the agents to point to the oobabooga api by default. There might be some other changes but those were the major two changes. I just forked your repo so I will upload them to my fork and see what all changes were made. I am using the ggml-vicuna-7b-4bit model.
Ah. We have clearly broken some of the config parsing now that I look back. It was getting in the way of our progress while we refactored the agents and database calls. I'm still interested to see where you needed to make changes, though. Thanks for your feedback!
@DataBassGit I uploaded the changes to my fork. You should be able to see them. The new persona was not used so it's not important. I first changed the config file for the variables, then edited the oobabooga api after reading the example api in the oobabooga repo. I then changed the api that the agents were looking for.
This is great. I think next weekend we will add config parsing for the LM back to the model. When we do that, I will ask you clone the repo and push your update oobabooga_api.py to main so that you can get credit for your work!
Did you make any additional changes to the startweb_ui.bat file besides what we had document?
@DataBassGit Yes. I had to specify the model that I wanted to use. python server.py --auto-devices --listen --no-stream --model ggml-vicuna-7b-4bit
I have the latest updates working with a local model using oobabooga api. Still working on the collection problem though. I'm not using persistence. it's coming from when the task creation agent tries to save the task and add it to the collection. i think the error is created when it tries to determine if the type of the target is an embedding. It's looking for a list but I don't think the list exists.
That may be an issue purely on the DB side. If you're able, can you add a try/except and see if you can catch the exception? We're not able to reproduce the issue, so anything we tell you at this point is a guess. I'd start by verifying your version of ChromaDB matches the requirements file.
I would also recommend commenting out the other agents. Any logs you can get us would be helpful.
Also, we are going to clean up the agent code today for readability. It will probably be late before we post it, but that might illuminate the issue.
I just wanted to follow up with you here. We've made some pretty significant overhauls. Are you still experiencing issues?