RWKV world 32k model fooocus on coding ability.

This model finetuned from RWKV world 7B with context 32k, focus on multi turn coding.

Trainning details

4*A800 27hours with 1B tokens

datasets

mainly tiny codes and add a lots of long context multi turn datasets. only finetuend in User: xxx\n\nAssistant: xxx\n format

Showcases

other

if using RWKV runner to run this model, need to wait for updates in chat mode, as default chat using Question: xxx\n\nAnswer: xxx and have a default system prompt so far.

completions mode is more suitable for this model

How

easiest way to use this model by using RWKV runner https://github.com/josStorer/RWKV-Runner, now RWKV runner support AMD GPUs in config webGPU

or you can use https://github.com/cgisky1980/ai00rwkvserver as well

BTW, runner support GPT-format like API