fmaclen
TypeError: Cannot read properties of undefined (reading 'content')
Starting from 0.13.1, while having a conversation with deepseek-coder-v2:16b-lite-instruct-q6_K (maybe with other models too, haven't tested) such an exception gets thrown:
deepseek-coder-v2:16b-lite-instruct-q6_K
Flow to reproduce it on my machine:
- Send a big message (Code of razor component, 600+ lines)
- Get response from the model
- Send another message
- Error.
The funny things is that the response can be shown for some time, but then it gets delete and replaced with "Sorry". And I end up with an error message and then the response
This error message is indicative of the context window of the model being exceeded.
We made some changes to the way we interact with Ollama a few months ago (way before 0.13.1) which was causing a similar error message to appear.
Unfortunately I'm traveling right now and it's impossible for me to download such a big model on hotel wifi, but I'll try to take a look at it in the next few weeks.
In the meantime, what version of Ollama are you using?
Thank you for the response! At the time the error appeared, I had been using 0.13.1. Then I updated to the most recent one (0.14.x), but the error did not disappear
@EduardKononov I tried this model a bunch, with long conversations and I'm unable to replicate this issue. Is it possible your Ollama server is running out of memory?
Can you provide the specs of the Ollama server environment?
You can run npx envinfo --system --npmPackages svelte,rollup,webpack --binaries --browsers.
I can replicate this on a system with 8GB VRAM + 64GB RAM, so ollama is not running out of memory, here is some more info. I'm using ollama 0.3.12 with CUDA (ollama-cuda) and model is partialy on GPU, partially on CPU (since it does not fit within 8GB of VRAM)
NOTE: I can also reproduce this error while running this model with ollama run, so I don't think that the issue is related to Hollama.
When running with ollama run, followin error occurs after some time of chatting:
Error: an unknown error was encountered while running this model
NAME ID SIZE PROCESSOR UNTIL
deepseek-coder-v2:16b 63fb193b3a9b 10 GB 22%/78% CPU/GPU 2 minutes from now
For debugging purposes I have prepared a following conversation that should trigger this (each line is a new prompt in the same session):
Generate me a curl command that makes a POST request to https://moodle.example.com/login/index.php with from data "username" and "password" and saves cookies output to a file
I need to use form data not query params
I also need a logintoken from moodle, how can I obtain it programatically with curl
What does -d parameter do?
Can you use -F instead of -d?
What do -oP flags in grep do?
What does \K mean in regular expression?
Write this using awk and regular expressions capture groups instead of \K
