Merge pull request #2945 from marko1616/bugfix/lora-model-merge

修复了在 transformers > 4.36.2 版本中部分模型合并 Lora 模型时因生成配置校验而导致的崩溃问题

Former-commit-id: 49f9dbb4b168b0e9f72e2233271aca130fca55e7

src/llmtuner/train/tuner.py

@@ -64,6 +64,15 @@ def export_model(args: Optional[Dict[str, Any]] = None):
         for param in model.parameters():
             param.data = param.data.to(output_dtype)
 
+    # Configuration check and fix
+    config = model.generation_config
+    if not config.do_sample and (
+        (config.temperature is not None and config.temperature != 1.0) or
+        (config.top_p is not None and config.top_p != 1.0) or
+        (config.typical_p is not None and config.typical_p != 1.0)
+    ):
+        config.do_sample = True
+
     model.save_pretrained(
         save_directory=model_args.export_dir,
         max_shard_size="{}GB".format(model_args.export_size),