11 lines
762 B
Markdown
11 lines
762 B
Markdown
---
|
|
license: apache-2.0
|
|
---
|
|
|
|
Llamafyd version of Qwen .5B further fine tuned on wiki, math, science, and chat datasets. Based on Cinder data.
|
|
This model should be fine tuned on further rag, function calling, programing, or assistant datasets for best performance.
|
|
Next model will have a focus on rag.
|
|
|
|
This model is ok at rag. It is very verbose from being trained on wikipedia Q and A with a whole article as the answer. Tiny-textbooks and Cosmopedia 100k, all very long responses.
|
|
It was also trained with normal RAG datasets, as well as a medical rag dataset I put together. Most of the common math chat datasets. Conversation datasets like hermes 1, fastchat, synthia, capybara, cinder, puffin, ect.
|
|
I will work on putting together the full list and posting. |