fix: use existing model config

src/axolotl/cli/convert_linear_attention.py

@@ -49,12 +49,9 @@ def do_linearize(cfg: DictDefault, cli_args: TrainerCliArgs) -> None:
     for p in model.parameters():
         p.requires_grad = False
 
-    # load config
-    base_config = load_model_config(cfg)
-
     # convert to linear llama
     linear_llama_config = LinearLlamaConfig.from_llama(
-        base_config, cfg.attention_config
+        model.config, cfg.attention_config
     )
     model = LinearLlamaForCausalLM.from_llama(
         model, config=linear_llama_config, train_attention=True