add down proj back for q4_0

intel-analytics · Dec 5, 2024 · 8948a43 · 8948a43
1 parent d8b0ced
commit 8948a43
Showing 1 changed file with 4 additions and 0 deletions.
diff --git a/python/llm/src/ipex_llm/transformers/npu_models/convert.py b/python/llm/src/ipex_llm/transformers/npu_models/convert.py
@@ -93,6 +93,10 @@ def replace_with_QuantizedLinear(layer, qtype, device, modules_to_not_convert,
             if (layer.in_features == 3584 and layer.out_features == 152064):
                 qtype = "sym_int8_rtn"
                 iqtype = ggml_tensor_qtype[qtype]
+        if qtype == "sym_int4_rtn":
+            if (layer.in_features == 18944 and layer.out_features == 3584):
+                qtype = "sym_int8_rtn"
+                iqtype = ggml_tensor_qtype[qtype]
         enable_scale_search = os.environ.get("IPEX_LLM_NPU_QUANTIZATION_OPT", "0") != "0"
         qweights, scale = ggml_convert_qtype(layer.weight.data.to(torch.float32),
                                              iqtype, device=device,