chatgpt-shell takes over 7 seconds to run: ollama models are slow to validate

[proger]

chatgpt-shell took too long to run so i did (elp-instrument-package "chatgpt-shell") and saw this in (elp-results):

chatgpt-shell                                               1           7.587093      7.587093
chatgpt-shell-start                                         1           7.587086      7.587086
chatgpt-shell--resolved-model                               15          7.586741      0.5057827333
chatgpt-shell-model-version                                 330         7.5860030000  0.0229878878
chatgpt-shell-ollama--validate-command                      1320        7.5674309999  0.0057329022
chatgpt-shell-ollama--fetch-model-versions                  1320        7.5549990000  0.0057234840
chatgpt-shell--prompt-pair                                  4           6.073431      1.51835775
chatgpt-shell--shell-info                                   5           5.062926      1.0125852
chatgpt-shell--model-short-version                          5           5.062874      1.0125747999
chatgpt-shell--update-prompt                                1           4.578275      4.578275
chatgpt-shell--model-label                                  5           2.523899      0.5047798
chatgpt-shell--make-buffer-name                             1           1.513425      1.513425
chatgpt-shell-google--validate-command                      2640        0.0003690000  1.397...e-07
chatgpt-shell-anthropic--validate-command                   1980        0.0002900000  1.464...e-07
chatgpt-shell-openai--validate-command                      330         0.0001559999  4.727...e-07
chatgpt-shell-kagi--validate-command                        330         0.0001279999  3.878...e-07
chatgpt-shell-deepseek--validate-command                    660         0.0001209999  1.833...e-07
chatgpt-shell--add-menus                                    1           5e-05         5e-05
chatgpt-shell--shrink-system-prompt                         10          4.7e-05       4.7e-06
chatgpt-shell--system-prompt-name                           5           3.500...e-05  7.000...e-06
chatgpt-shell--primary-buffer                               3           2.100...e-05  7.000...e-06
chatgpt-shell--shell-buffers                                3           1.499...e-05  4.999...e-06
chatgpt-shell-duplicate-map-keys                            1           4e-06         4e-06
chatgpt-shell--prompt-regexp                                4           0.0           0.0

ollama stuff takes lots of time and I don't even have it.

The mitigation is to restrict available models:

(setq chatgpt-shell-models (chatgpt-shell-openai-models))

[xenodium]

Hey thanks for reporting with details! Sorry for the delay here...

Can you share what OS you're running on? Are you online/offline when starting the shell? Are you running ollama on your machine? I'm on macOS, running chatgpt-shell with and without Ollama running in the background and haven't been able to reproduce just yet.

[proger]

I use macOS 14.5, I was online, not running ollama.

[strickinato]

I was hitting a lot of really bad slowness.

I believe the issue is because if you do not have chatgpt-shell-model-version set -- it tries to validate all the models to see which one to choose. There are a lot of models so this takes a long time and happens syncronously.

Setting chatgpt-shell-model-version to a specific model fixed the slowness for me.

@proger -- I wonder if you didn't have it set?

[strickinato]

Hmmm...

Actually - maybe that's not helpful. It seems you were wondering about the ollama models taking specifically long --

That I don't know... but setting the chatgpt-shell-model-version makes it not relevant 😄 🤷