error in process filter: Wrong number of arguments: #<subr F616e6f6e796d6f75732d6c616d626461_anonymous_lambda_27>, 1

s-kostyaev / ellama

Ellama is a tool for interacting with large language models from Emacs.

GNU General Public License v3.0

492 stars 30 forks source link

error in process filter: Wrong number of arguments: #<subr F616e6f6e796d6f75732d6c616d626461_anonymous_lambda_27>, 1 #13

Closed artelse closed 10 months ago

artelse commented 11 months ago

Title says it all. This is my first install of Ollama and Ellama. Running latest version 20231105.1147

s-kostyaev commented 11 months ago

Please M-x toggle-debug-on-error, reproduce the error and paste here debugger window content. Which version of emacs do you use?

artelse commented 11 months ago

Enabled the debugger and found it couldn't find the model. Fixed this in my init and now I don't get the error. However, when I try to summarize a region of text, it causes 100% cpu and the summary is printed extremely slow. This is on Linux and Emacs 30.0.50, not nightly but from a few weeks ago.

s-kostyaev commented 10 months ago

Which hardware do you have? Cpu, GPU, ram. And which model do you use? Default one?

artelse commented 10 months ago

Hardware might be the problem: Intel Core i7-5600, HD Graphics 5500 So it might just run on cpu with 16GB RAM. Model is zephyr, perhaps switch to a lighter model; any recommendation? Intended use is coding and general text use.

s-kostyaev commented 10 months ago

You can try:

lower quantization of zephyr (like 7b-beta-q2_K, 7b-beta-q3_K_S, 7b-beta-q3_K_M, 7b-beta-q3_K_L
lighter models, like orca-mini:3b-q4_0 or starcoder:1b, starcoder:3b for coding
deploy ollama on more powerful machine and connect ellama to it (on cloud provider, for example)
use OpenAI of Google API and connect ellama to it

artelse commented 10 months ago

Thanks tested with orca-mini and it works. Still a bit slow, so might go for a server solution or get a better GPU. When I switch to an org-mode buffer and ask to write some code, it still assumes ``` markdown syntax for the code output.

s-kostyaev commented 10 months ago

Org syntax should be supported separately. I will think about it later.