bug: load model first time load meanwhile i cant access my SSD - TENSOR convert ???

[kalle07]

Jan version

0.5.11

Describe the Bug

i load a model the first time
nearly no CPU or GPU is used meanwhile and it need ~120 sec on my SSD (100% read/wite)
iv tested 10 models and they run fine in GPT4all (~120sec first time) but i can access my SSD while the model is loaded ...

while load the model with JAN it need also 120sec (first time) but meanwhile i cant acces my SSD !!!
btw what is happened that long time ? i can copy on same SSD 10GB in 20sec.
i think the model is converted to TENSOR-RT (iv installed) at least you can write please wait util convert

Steps to Reproduce

No response

Screenshots / Logs

No response

What is your OS?

• MacOS
• Windows
• Linux

[louis-jan]

Hi @kalle07. What model are you using?
Could you please assist in uploading the log files? We will investigate then.

[kalle07]

which log where i can find ?

even here no info where is the path

i mean i would help you but ^^

[louis-jan]

Hi @kalle07 Hi @kalle07, thanks for the help. You can click the System Monitor (bottom-right corner) to see the Open App Log button.

[kalle07]

the last one need 3min to load

and all in all ... all model talk such as much ... the overall settings on models are not that good ^^

cortex.log

[louis-jan]

Hi @kalle07 Could you please try Llama 3.2 1B or 3B to see if the models run fine? Also, would you mind sharing your computer specs here? Like:

• CPU
• GPU
• RAM
• SSD model

[kalle07]

that seems ok ...

log-txt.txt

but
https://huggingface.co/bartowski/Llama-3.1-SauerkrautLM-8b-Instruct-GGUF/tree/main
SauerkrautLM-Llama-3.1-8b-Instruct.Q5_K_M.gguf
need 3min
log2.txt

but if i exit jan, start again same model load its fast ~5sec

intel 8core 4GHZ
rtx4660
64GB
SSD (standart 500MB/s - not fast NVM)

[louis-jan]

@kalle07 Is it possible that other processes are also using SSD? There's also a background running inference. Does the Activity Monitor report app usage at 100%?

[kalle07]

no its repeatable ... and NO GPU or CPU usage while this 3min