Large files cause freeze when embedding

[rjmacarthy]

Describe the bug
Large files cause freeze when embedding

To Reproduce
Embed a file > 500 lines or so.

Expected behavior
It should work

Screenshots
na

Logging

API Provider
Ollama

Chat or Auto Complete?
Embedding

Model Name
All models

Desktop (please complete the following information):

• OS: Ubuntu

Additional context
Nothing else

[gpinkham]

I was noticing this but I also am noticing that I can't seem to get Ollama to keep the embeddings model in memory. 4 mins into the embedding process the model disappears and the Twinny process hangs there. I thought it was large files but I'm now thinking its the model being unloaded. I have set the env variable to tell it to keep models in memory for 60m but that hasn't impacted it. (guessing I didn't set it properly).

[gpinkham]

some additional debugging info in case it helps.
I ran ollama with a keep alive of 10h and debug turned on.. verified the model is in memory and ollama ps says its there for 10 hours.
I clicked on the "embed documents".. I see the twinny dialog in vscode.. it shows progress and then stops at 8.20%. the file its on is about 350 lines of Ruby code.

in the ollama debug I see this.

time=2024-12-02T22:32:56.359-05:00 level=DEBUG source=sched.go:575 msg="evaluating already loaded" model=/Users/gpinkham/.ollama/models/blobs/sha256-797b70c4edf85907fe0a49eb85811256f65fa0f7bf52166b147fd16be2be4662 time=2024-12-02T22:32:56.362-05:00 level=DEBUG source=runner.go:752 msg="embedding request" content="�\x02 REMOVED CONTENTS �\x01�\x01" time=2024-12-02T22:32:56.362-05:00 level=DEBUG source=cache.go:104 msg="loading cache slot" id=0 cache=71 prompt=8 used=0 remaining=8 [GIN] 2024/12/02 - 22:32:56 | 200 | 12.461333ms | 127.0.0.1 | POST "/api/embed" time=2024-12-02T22:32:56.371-05:00 level=DEBUG source=sched.go:407 msg="context for request finished" time=2024-12-02T22:32:56.371-05:00 level=DEBUG source=sched.go:339 msg="runner with non-zero duration has gone idle, adding timer" modelPath=/Users/gpinkham/.ollama/models/blobs/sha256-797b70c4edf85907fe0a49eb85811256f65fa0f7bf52166b147fd16be2be4662 duration=10h0m0s time=2024-12-02T22:32:56.371-05:00 level=DEBUG source=sched.go:357 msg="after processing request finished event" modelPath=/Users/gpinkham/.ollama/models/blobs/sha256-797b70c4edf85907fe0a49eb85811256f65fa0f7bf52166b147fd16be2be4662 refCount=0

10 minutes later the twinny dialog has not moved beyond that file and the 8.20%. and the model is still in memory (per ollama ps) so its not ollama unloading the model causing the hang in twinny.

I also stopped the ollama server and the twinny dialog is still running. (bar moving across the bottom of the dialog but percentage still not changing)