Modify gptq example logic (#1545)

Signed-off-by: YIYANGCAI <yiyang.cai@intel.com>
intel · Jan 17, 2024 · e951e7a · e951e7a
1 parent 2835bdb
commit e951e7a
Showing 1 changed file with 6 additions and 5 deletions.
diff --git a/...s/pytorch/nlp/huggingface_models/language-modeling/quantization/llm/run_clm_no_trainer.py b/...s/pytorch/nlp/huggingface_models/language-modeling/quantization/llm/run_clm_no_trainer.py
@@ -367,16 +367,17 @@ def eval_func(model):
     if args.ipex:
         user_model = load(os.path.abspath(os.path.expanduser(args.output_dir)))
     else:
-        user_model, _ = get_user_model()
-        kwargs = {'weight_only': True} if args.approach == 'weight_only' else {}
-        user_model = load(os.path.abspath(os.path.expanduser(args.output_dir)), user_model, **kwargs)
+        if args.gptq_debug:
+            user_model = torch.load(os.path.join(args.output_dir, "gptq_best_model.pt"))
+        else:
+            user_model, _ = get_user_model()
+            kwargs = {'weight_only': True} if args.approach == 'weight_only' else {}
+            user_model = load(os.path.abspath(os.path.expanduser(args.output_dir)), user_model, **kwargs)
 else:
     user_model, _ = get_user_model()
 
 if args.accuracy:
     user_model.eval()
-    if args.gptq_debug:
-        user_model = torch.load(os.path.join(args.output_dir, "gptq_best_model.pt"))
     if args.code_generation:
         from intel_extension_for_transformers.llm.evaluation.lm_code_eval import evaluate
         from transformers import AutoTokenizer