ggml-hexagon: Implement true Q8_0 quantization on Hexagon NPU for more accurate mixed-precision matmul operations (#17977)

* feat: implement real Q8_0 * feat: adding cmake option for configuring FP32 quantize group size * typo: set() shall be used --------- Co-authored-by: ngdxzy <zhenyu_xu@uri.edu>
2025-12-19 12:42:28 -05:00
parent 14931a826e
commit ce734a8a2f
5 changed files with 147 additions and 2 deletions
@@ -22,6 +22,7 @@
            "GGML_LLAMAFILE":   "OFF",
            "GGML_OPENCL":      "ON",
            "GGML_HEXAGON":     "ON",
+            "GGML_HEXAGON_FP32_QUANTIZE_GROUP_SIZE": "128",
            "LLAMA_CURL":       "OFF"
        }
    },
@@ -36,6 +37,7 @@
            "GGML_LLAMAFILE":   "OFF",
            "GGML_OPENCL":      "ON",
            "GGML_HEXAGON":     "ON",
+            "GGML_HEXAGON_FP32_QUANTIZE_GROUP_SIZE": "128",
            "LLAMA_CURL":       "OFF"
        }
    },