ed
/
rwkv.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Update README.md

This commit is contained in:
saharNooby 2023-03-31 19:06:31 +04:00
parent 01d667f066
commit 02c9946b57
1 changed files with 2 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
README.md
View File

@ -4,13 +4,12 @@ This is a port of [BlinkDL/RWKV-LM](https://github.com/BlinkDL/RWKV-LM) to [gger
**WORK IN PROGRESS: NOTHING WORKS YET!** If you know C/C++/ggml, please help! **WORK IN PROGRESS: NOTHING WORKS YET!** If you know C/C++/ggml, please help!
Inference code runs, but outputs all `NaN`s in logits, most probably due to missing operators. Values are correct up to `ln0`: result of `ln0` matches with reference implementation. Inference code runs and outputs some correctly-looking numbers in logits. Values are checked to be correct at least up to `ln0`, they match with reference implementation.
## Plan ## Plan
1. Make FP32 inference work 1. Make FP32 inference work
1. Implement and use element-wise `max`, `exp`, `sigmoid` 1. Compare vectors step-by-step with reference implementation
2. Compare vectors step-by-step with reference implementation
2. Validate states and logits against [reference implementation](https://github.com/BlinkDL/ChatRWKV/blob/main/RWKV_in_150_lines.py) by creating a testing script 2. Validate states and logits against [reference implementation](https://github.com/BlinkDL/ChatRWKV/blob/main/RWKV_in_150_lines.py) by creating a testing script
3. Heavily refactor code; optimize where possible 3. Heavily refactor code; optimize where possible
4. Make FP16 inference work 4. Make FP16 inference work