Support fp6 save & load (#11034)

intel-analytics · May 15, 2024 · 686f603 · 686f603
1 parent ac384e0
commit 686f603
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/python/llm/src/ipex_llm/transformers/low_bit_linear.py b/python/llm/src/ipex_llm/transformers/low_bit_linear.py
@@ -268,7 +268,7 @@ def ggml_q_format_convet_xpu2cpu(tensor: torch.Tensor, num_elem: int, qtype: int
 
     src = ctypes.c_void_p(tensor.data.data_ptr())
 
-    if qtype in [SYM_INT4, ASYM_INT4, SYM_INT8, NF4, NF3, FP4, FP8E4, FP8E5]:
+    if qtype in [SYM_INT4, ASYM_INT4, SYM_INT8, NF4, NF3, FP4, FP6, FP8E4, FP8E5]:
         dst_tensor = torch.empty_like(tensor)
     elif qtype == ggml_tensor_qtype["sym_int5"]:
         QK = ggml.ggml_qk_size(ggml_tensor_qtype["asym_int5"])