cocktailpeanut / dalai

The simplest way to run LLaMA on your local machine
https://cocktailpeanut.github.io/dalai
13.09k stars 1.43k forks source link

Docker compose never responds #455

Open oscadev opened 1 year ago

oscadev commented 1 year ago

I have ubuntu with a 12GB 3080, 10700k, and 32GB. I installed this by clonging the repo, then doing the few docker commands listed. I have both alpaca 7 and 13 installed. Using the general template, I use this prompt

Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.

### Instruction:
What is 1 plus 1?

### Response:

and it just keeps spinning with the "cancel" button, and it never answers. This is where it stays:


dalai-dalai-1  | exec: /root/dalai/alpaca/main --seed -1 --threads 4 --n_predict 200 --model models/7B/ggml-model-q4_0.bin --top_k 40 --top_p 0.9 --temp 0.8 --repeat_last_n 64 --repeat_penalty 1.3 -p "Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
dalai-dalai-1  | 
dalai-dalai-1  | ### Instruction:
dalai-dalai-1  | What is 1 plus 1?
dalai-dalai-1  | 
dalai-dalai-1  | ### Response:
dalai-dalai-1  | " in /root/dalai/alpaca
DanielNobbe commented 1 year ago

Experiencing the same on Ubuntu, not seeing any threads show high usage either

okellotechconsults commented 1 year ago

Same issue, trying to deploy on vps, I tried all methods, started with the npx methods, then docker but they all behave the same way, no response is ever returned, it's a 2CPU ARM64 with 16GB RAM

cmerther commented 12 months ago

Also getting the same on a windows 10 machine with the 13B model.

fdyhgs commented 11 months ago

Tick "debug" on top left and run again. I'd expect an error message to show.

mirek190 commented 11 months ago

what do you expect from dead project?

Stop using that ancient dead project and go to llamacpp or koboldcpp ... Also download models from https://huggingface.co/TheBloke ggml versions