gputopia-worker-0.3.1

• connects with model lists so the queen can schedule better
• supports imagegen via sdxl
• download https://gputopia.s3.us-east-2.amazonaws.com/bin/gputopia-worker-cuda-torch-linux-64.tar.gz to support image generation and fine tuning
• eventually we will make that release easier to install

gputopia-worker-0.3.0

• extended support for /v1/embeddings
  • use model "fastembed:BAAI/bge-base-en-v1.5" for the most recent and fastest models
  • llama model embeddings work as expected
• download https://gputopia.s3.us-east-2.amazonaws.com/bin/gputopia-worker-cuda-torch-linux-64.tar.gz to support fine tuning

gputopia-worker-0.2.3

• fix for another possible worker shutdown when websocket connection is unstable

gputopia-worker-0.2.2

• fixed reconnect bug

latest torch fine tuning release is here:

https://gputopia.s3.us-east-2.amazonaws.com/bin/gputopia-worker-cuda-torch-linux-64.tar.gz

gputopia-worker-0.2.1

• fixed bug when a ~/.config dir was missing.
• version includes torch support
• workers sign requests using their public key
• worker public key can be searched for in the worker-detail endpoint

gputopia-worker-0.1.9

ver

gputopia-worker-0.1.6

.

gputopia-worker-0.1.3

Merge branch 'main' of github.com:ArcadeLabsInc/ai-worker

gputopia-worker-0.1.2

fixed problem where websocket would drop during some longer inferences

gputopia-worker-0.1.1

• uses the wss://queenbee.gputopia.ai/worker
• user has to include their LNURL to get paid or an auth-key containing uid:
• user has to have an account with gputopia with a matching LNURL or uid
• default is volunteer/unpaid
• currently only supports huggingface gguf or ggml models