support to export gguf q4_0 and q4_1 format

auto_round/__main__.py

@@ -14,15 +14,15 @@
 import sys
 
 def run_eval():
-    if "--sequence" in sys.argv:
-        sys.argv.remove("--sequence")
-        from auto_round.script.llm import setup_eval_parser, eval_sequence
-        args = setup_eval_parser()
-        eval_sequence(args)
-    else:
+    if "--non_sequence" in sys.argv:
+        sys.argv.remove("--non_sequence")
         from auto_round.script.llm import setup_eval_parser, eval
         args = setup_eval_parser()
         eval(args)
+    else:
+        from auto_round.script.llm import setup_eval_parser, eval_sequence
+        args = setup_eval_parser()
+        eval_sequence(args)
 
 def run():
     if "--eval" in sys.argv:

auto_round/script/llm.py

@@ -313,8 +313,12 @@ def tune(args):
     from auto_round.utils import logger
 
     if args.format in ["gguf:q4_0", "gguf:q4_1"]:
+        args.bits = 4
+        if args.act_bits <= 8:
+            logger.warning(f"{args.format} not support for activation quantization. Reset act_bits to 16.")
+            args.act_bits = 16
         if args.group_size != 32:
-            logger.warning(f"{args.format} not support for group_size: {args.group_size}."
+            logger.warning(f"{args.format} not support for group_size: {args.group_size}. "
                 "Reset group_size to 32.")
             args.group_size = 32
         if args.format.endswith("_0"):