DigitalDaimyo
/

ASA-ASM-wikitext103-raw

addressed-state-attention

Model card Files Files and versions

Justin Brown commited on 27 days ago

Commit

c2f9ca0

·

verified ·

1 Parent(s): 6084694

Update README.md

Files changed (1) hide show

README.md +2 -1

README.md CHANGED Viewed

@@ -27,5 +27,6 @@ This repo contains a checkpoint for an **Addressed State Model (ASM)** built fro
 import torch
 ckpt = torch.load("ASA_ASM_wt103-rawv1_gpt2_T1024_L21_D384_H8_K16_M32_ropek1_alibi1_gamma1_step75000_best.pt", map_location="cpu")
 print(ckpt.keys())
-Notes
 This is a raw training artifact. A lightweight inference wrapper / conversion can be added later.

 import torch
 ckpt = torch.load("ASA_ASM_wt103-rawv1_gpt2_T1024_L21_D384_H8_K16_M32_ropek1_alibi1_gamma1_step75000_best.pt", map_location="cpu")
 print(ckpt.keys())
+## Notes
 This is a raw training artifact. A lightweight inference wrapper / conversion can be added later.