[Question]: I'm getting a generic response when using Ollama.

[edbyford]

What is your question?

Hi folks, I am using fabric with Olama and phi3:medium. I have a brand new MacBook Pro M4 Max 36GB. I installed fabric using Go and used the script to generate aliases for each pattern.

I think I'm getting a generic response because when I use extract_wisdom it doesn't use the structure given in the pattern. (IDEAS, QUOTES, etc)

I expected to get a generic response if my GPU is not powerful enough. So, I tried piping it through a summary. However, my syntax was incorrect and I just used "summary" rather than "create_summary" - and then it used the correct structure.

So I'm trying to debug why it seems to give me the correct structure when I pipe it through something incorrectly.

---

Here's what I used:

yt https://www.youtube.com/watch\?v\=5x4s2d3YWak | extract_wisdom

Gave me a generic response, with some details from the transcript but no structure as given in the prompt.

Then I tried:

yt https://www.youtube.com/watch\?v\=5x4s2d3YWak | summary | fabric -sp extract_wisdom

This gives me a result closer to what I expected, with the structure.

Then:

yt https://www.youtube.com/watch\?v\=5x4s2d3YWak | summary
zsh: command not found: summary

This gives an error.

Here's the result of --listmodels:

fabric --listmodels

Available models:

Ollama

	[1]	phi3:medium
	[2]	gemma2:latest
	[3]	gemma:7b
	[4]	dolphin-mixtral:latest
	[5]	llama3.2:latest

[edbyford]

I've made some progress. So I created a new model file based on phi3:medium-128K and gave it a context of 32768 based on the instructions in this YouTube video.

I then selected, via fabric --setup, the new model and told fabric that the context was 32768.

However, whilst the #summary and #ideas I get are better, there are no other elements to the structure, things like references or habits.

What am I missing here?