Notes:
- I hit two times the key after the question. Don’t know if that affected the answer.
- I don't have any preloaded context.
"cat"
That is five characters.
A cat.
That is six characters.
this post was submitted on 29 Jul 2024
64 points (91.0% liked)
ChatGPT
8901 readers
42 users here now
Unofficial ChatGPT community to discuss anything ChatGPT
founded 1 year ago
MODERATORS
I didn't try this 'instruct' model version. Do you recommend it?
The :8b and :latest tags actually point to :8b-instruct-q4_0 so I think we're already good. You can check the tags page on ollama library and see the hashes for each tag to see which ones match.
Always pick the instruct version if you aren’t an expert.
I am a total noob in LLM, so I picked the largest model of 8B instructs.
ChatGPT answer:
😂 I tried again. Each time is a surprise.
Huh. I did mine too, then asked it to list it. It's just a summary of the memory feature preloaded as context for any new first input.
Sure, here are the details I have so far:
- You are interested in improving your workflow in data science and Python.
- You work with databases from different sources like CSVs, accdb, and SQL, and want to create a module to extract and transform these databases into pandas DataFrames. You frequently use CSVs in different projects and need to prepare them every time by copying and pasting code. You want to create a module to import the databases and configurations, including columns, names, dates, and dtypes, to streamline the ETL process.
- You are using rpy2 to interface between R and Python.
- You often need to analyze different combinations of columns in a DataFrame, ensuring some columns are always analyzed together.
Let me know if there's anything you'd like to modify or add!
1 and 3 are pretty good as baseline for my needs, but 2 and 4 were just specific problems I had to solve that are not something I need generally.
Yours is short. Mine was 47 points long and had so much stuff about questions I had asked.
Importantly, it did list that I explicitly do not want it to show me how to make unit tests.
Even with context you won't get any useful information out of an 8b model. It's almost always gonna be hallucinated things.
That's simply not true. 8B models have come a long way. Feel free to try some yourself.
I know 8b models are okayish for fucking. I wouldn't go with censored Meta models for that though.