diff --git a/python/llm/src/ipex_llm/ggml/quantize.py b/python/llm/src/ipex_llm/ggml/quantize.py
index 98917d8c..a5e2cb91 100644
--- a/python/llm/src/ipex_llm/ggml/quantize.py
+++ b/python/llm/src/ipex_llm/ggml/quantize.py
@@ -45,7 +45,8 @@ ggml_tensor_qtype = {"sym_int4": 2,   # q4_0 in ggml
                      "q2_k": 23,
                      "gguf_iq1_s": 24,
                      "gguf_iq1_m": 25,
-                     "q6_k": 26}
+                     "q6_k": 26,
+                     "q4_k": 27}
 
 _llama_quantize_type = {"q4_0": 2,
                         "q4_1": 3,
diff --git a/python/llm/src/ipex_llm/transformers/low_bit_linear.py b/python/llm/src/ipex_llm/transformers/low_bit_linear.py
index 5ac7cb7b..534e077d 100644
--- a/python/llm/src/ipex_llm/transformers/low_bit_linear.py
+++ b/python/llm/src/ipex_llm/transformers/low_bit_linear.py
@@ -74,6 +74,7 @@ IQ2_XXS = ggml_tensor_qtype["gguf_iq2_xxs"]
 IQ2_XS = ggml_tensor_qtype["gguf_iq2_xs"]
 Q2_K = ggml_tensor_qtype["q2_k"]
 IQ1_S = ggml_tensor_qtype["gguf_iq1_s"]
+Q4_K = ggml_tensor_qtype["q4_k"]
 Q6_K = ggml_tensor_qtype["q6_k"]
 
 
@@ -215,7 +216,7 @@ def ggml_convert_qtype(tensor: torch.Tensor, qtype: int,
     if not convert_shape_only and device != 'meta':
         dst = ctypes.c_void_p(dst_tensor.data.data_ptr())
         hist = (ctypes.c_int64 * 16)()
-        if qtype not in [IQ2_XXS, IQ2_XS, Q2_K, IQ1_S, Q6_K]:
+        if qtype not in [IQ2_XXS, IQ2_XS, Q2_K, IQ1_S, Q4_K, Q6_K]:
             ggml.ggml_quantize_tensor(src, dst, qtype, n, k, hist)
         else:
             if imatrix is not None:
diff --git a/python/llm/src/ipex_llm/transformers/model.py b/python/llm/src/ipex_llm/transformers/model.py
index a5640fe2..34b4b0c0 100644
--- a/python/llm/src/ipex_llm/transformers/model.py
+++ b/python/llm/src/ipex_llm/transformers/model.py
@@ -119,7 +119,7 @@ class _BaseAutoModelClass:
                                 ``'sym_int5'``, ``'asym_int5'``, ``'sym_int8'``, ``'nf3'``,
                                 ``'nf4'``, ``'fp4'``, ``'fp8'``, ``'fp8_e4m3'``, ``'fp8_e5m2'``,
                                 ``'gguf_iq2_xxs'``, ``'gguf_iq2_xs'``, gguf_iq1_s'``,
-                                ``'fp16'``, ``'bf16'``, or ``'q6_k'``,
+                                ``'fp16'``, ``'bf16'``, ``'q4_k'`` or ``'q6_k'``,
                                 ``'sym_int4'`` means symmetric int 4, ``'asym_int4'`` means
                                 asymmetric int 4, ``'nf4'`` means 4-bit NormalFloat, etc.
                                 Relevant low bit optimizations will be applied to the model.
@@ -378,7 +378,7 @@ class _BaseAutoModelClass:
                           f"Unknown load_in_low_bit value: {q_k}, expected:"
                           f" sym_int4, asym_int4, sym_int5, asym_int5, sym_int8, nf3, nf4, "
                           f"fp4, fp8, fp8_e4m3, fp8_e5m2, fp16,  bf16, gguf_iq2_xxs, "
-                          f"gguf_iq2_xs, gguf_iq1_s, q2_k, q6_k, mixed_fp4 or mixed_fp8.")
+                          f"gguf_iq2_xs, gguf_iq1_s, q2_k, q4_k, q6_k, mixed_fp4 or mixed_fp8.")
         qtype = ggml_tensor_qtype[q_k]
 
         # In case it needs a second try,