Updated config. Removed "use_cache": true and also "cache_implementation". They were interfering in some versions of transformers.

config.json

@@ -12,7 +12,6 @@
   "attention_dropout": 0.0,
   "attn_logit_softcapping": null,
   "bos_token_id": 2,
-  "cache_implementation": "hybrid",
   "classifier_dropout_prob": 0.0,
   "eos_token_id": 1,
   "final_logit_softcapping": null,
@@ -102,7 +101,7 @@
   "sliding_window_pattern": 6,
   "torch_dtype": "float32",
   "transformers_version": "4.51.3",
-  "use_cache": true,
+  "use_cache": false,
   "use_non_causal_attention": true,
   "vocab_size": 262146
 }