Remove FIM limitation note

README.md

@@ -101,10 +101,6 @@ next_logits = fim_forward(model, before_ids, after_ids, lang_id=LANG_ID)
 print(tok.decode([next_logits.argmax().item()]))
 ```
 
-> **Note**: These models were trained primarily for prefix completion. The `wne` embedding
-> (norm ~11 vs `wpe` ~110) is small, suggesting limited FIM training. Suffix-first FIM
-> (Method 1) leverages causal attention to incorporate suffix context without special tokens.
-
 ## Configuration
 
 | Field | Value |