Small models similar to reader-lm?

Posted by noellarkin@reddit | LocalLLaMA | View on Reddit | 5 comments

https://ollama.com/library/reader-lm Just came across this and it's a perfect fit for my workflow - - I was using a few-shot prompt to achieve this. I love that it works so well as a 1.5b model. Are there any other "niche use case" models that are under 2B, and reliable enough to be used in workflows? I tested the new LLaMa 1B models and they're not all that useful, IMO once you go below 3B it makes more sense to have models trained on specific use cases.

Reply to Post

5 Comments

[-]

cpdomina@reddit

check out llmware's models [https://huggingface.co/llmware](https://huggingface.co/llmware) the SLIM and BLING categories. they train small models for very specific tasks

[-]

noellarkin@reddit (OP)

thanks! yep this is exactly what I wanted to check out :)

[-]

RandiyOrtonu@reddit

have you tried gemma2 2b and qwen2.5 1.5b?

[-]

noellarkin@reddit (OP)

yep - - I'm more curious about specialist models tbh, like reader-lm or something similar to https://ollama.com/library/nuextract (its 3B+ but similar approach ie fine-tuned for one task)

[-]

RandiyOrtonu@reddit

hm I understood u want like base models for specific tasks only rather than fine-tuned models