fix(rope): YaRN attention_factor is (0.1*ln(factor)+1)*attn_factor, not the bare multiplier

config.json

@@ -41,7 +41,7 @@
       "original_max_position_embeddings": 4096,
       "beta_slow": 1.0,
       "beta_fast": 64.0,
-      "attention_factor": 1.0,
+      "attention_factor": 1.4158883083359672,
       "partial_rotary_factor": 1.0
     }
   },
@@ -202,4 +202,4 @@
   "decoder_sparse_step": 1,
   "norm_topk_prob": true,
   "sliding_window": 0
-}
+}