BlinkDL commited on
Commit
8411e7f
1 Parent(s): 0134aed

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +7 -13
README.md CHANGED
@@ -20,27 +20,21 @@ tags:
20
  license: apache-2.0
21
  datasets:
22
  - EleutherAI/pile
23
- - togethercomputer/RedPajama-Data-1T
24
  ---
25
 
26
  # RWKV-5 World (Training in Progress)
27
 
28
- RWKV-5 Details: https://twitter.com/BlinkDL_AI/status/1685230712247795713
29
 
30
- Use rwkv pip package 0.8.7+ for RWKV-5 inference.
 
 
31
 
32
  ## Model Description
33
 
34
  RWKV-5 trained on 100+ world languages (70% English, 15% multilang, 15% code).
35
 
36
- World = Some_Pile + Some_RedPajama + Some_OSCAR + All_Wikipedia + All_ChatGPT_Data_I_can_find
37
-
38
- RWKV-5 training: set --my_testing "r" ("r2" is better) in latest RWKV-LM
39
-
40
- RWKV-5 inference implementation reference: https://github.com/BlinkDL/ChatRWKV/blob/main/RWKV_v5_demo.py
41
-
42
- 0.1B world:
43
-
44
- RWKV-5 37% trained = LAMBADA ppl 18.1 acc 42.93%
45
 
46
- RWKV-4 100% trained = LAMBADA ppl 25.5 acc 36.29%
 
20
  license: apache-2.0
21
  datasets:
22
  - EleutherAI/pile
23
+ - cerebras/SlimPajama-627B
24
  ---
25
 
26
  # RWKV-5 World (Training in Progress)
27
 
28
+ Use rwkv pip package 0.8.14+ for RWKV-5 inference.
29
 
30
+ GUI: https://github.com/josStorer/RWKV-Runner (see Releases)
31
+
32
+ How it works: https://twitter.com/BlinkDL_AI/status/1685230712247795713
33
 
34
  ## Model Description
35
 
36
  RWKV-5 trained on 100+ world languages (70% English, 15% multilang, 15% code).
37
 
38
+ World = Some_Pile + Some_SlimPajama + Some_OSCAR + All_Wikipedia + All_ChatGPT_Data_I_can_find
 
 
 
 
 
 
 
 
39
 
40
+ RWKV-5 training: set --my_testing "2r4" in latest RWKV-LM