RWKV-Notebooks

Uses jsonl and binidx instead of the garbage RAM-intensive scripts.

Example.jsonl file contents:

{"text": "This is a sentence"}
{"text": "Instruction: a\n\nInput: b\n\nResponse: c"}
{"text": "Question: a\n\nAnswer: b"}

The tokenizer will combine all jsonl files inside your dataset folder into two files the train script will read. Read the outputs of the cells to know what to do.

---

RWKV-infctx:

Only tested 0.4B-World

RWKV-LorA:

Soon™