RWKV-7 trained on the Pile w/ "20b tokenizer" (332115325534 tokens)
0.1B = L12-D768, lr 8e-4 to 3e-5 cosine decay, wd 0.1, bsz 8x30x4096
0.4B = L24-D1024, lr 6e-4 to 2e-5 cosine decay, wd 0.1, bsz 8x30x4096
1.5B = L24-D2048, lr 5e-4 to 1.5e-5 cosine decay, wd 0.1, bsz 8x45x4096
Check https://github.com/BlinkDL/RWKV-LM for details.
How to run it:
https://pypi.org/project/rwkv/
or
Inference Providers
NEW
This model is not currently available via any of the supported Inference Providers.
The model cannot be deployed to the HF Inference API:
The model has no library tag.