update gguf and support for CompressedLinear (#950)

Author: n1ck-guo

auto_round/__main__.py

@@ -789,7 +789,7 @@ def run_eval():
     if args.eval_task_by_task:
         eval_task_by_task(
             model=args.model,
-            device=args.device,
+            device=args.device_map,
             tasks=args.tasks,
             batch_size=args.eval_bs,
             trust_remote_code=not args.disable_trust_remote_code,

auto_round/export/export_to_gguf/convert.py

@@ -412,7 +412,7 @@ def prepare_tensors(cls):
             skip = False
             for tensor_info in cls.gguf_writer.tensors:
                 if new_name in tensor_info:
-                    print("new_name already add to gguf_writer, skip")
+                    logger.warning(f"{new_name} already add to gguf_writer, skip")
                     skip = True
                     break
             if skip: