Trying out the h2ogpt locally to chat with documents.

If using the one-click installer, in the step where I mentioned to copy paste the text in the Target textbox, I forgot to mention to replace "pythonw" with "python" in that text before running it a command window.

Github page: [ Ссылка ]

Manual installation steps: [ Ссылка ]

Issue loading model from one click installer: [ Ссылка ]

For the issue mentioned loading TheBloke's model through the app from the one click installer, the resolution is:
In the Models tab, in the textbox "Choose Base Model" select "llama" and more options will appear in the UI, and it auto populates the fields on the bottom to:

Choose LLaMa.cpp Model Path/URL (for Base Model: llama): [ Ссылка ]
LLaMa.cpp Num. GPU Layers Offloaded: 100
LLaMa.cpp Batch Size: 1024
LLaMa.cpp Num. Group Query Attention (8 for 70B LLaMa2): 0

Then you can click "Load-Unload Model/LORA" on the top right.

Voice made with Eleven Labs

свернуть