Add ROCmFP4 CPU quantization support

ggml/include/ggml.h

@@ -437,7 +437,9 @@ extern "C" {
         GGML_TYPE_TQ3_1S  = 47, // TurboQuant 3-bit weight: WHT-rotated 8-level Lloyd-Max, block_size=32
         GGML_TYPE_TQ4_1S  = 48, // TurboQuant 4-bit weight: WHT-rotated 16-level Lloyd-Max, block_size=32
         GGML_TYPE_Q6_0    = 49,
-        GGML_TYPE_COUNT   = 50,
+        GGML_TYPE_Q4_0_ROCMFP4      = 50, // ROCmFP4 experimental UE4M3 scales + packed AMD FP4 blocks
+        GGML_TYPE_Q4_0_ROCMFP4_FAST = 51, // ROCmFP4 single-scale speed layout
+        GGML_TYPE_COUNT   = 52,
     };
 
     // precision
@@ -481,6 +483,8 @@ extern "C" {
         GGML_FTYPE_MOSTLY_MXFP4   = 25, // except 1d tensors
         GGML_FTYPE_MOSTLY_NVFP4   = 26, // except 1d tensors
         GGML_FTYPE_MOSTLY_Q1_0    = 27, // except 1d tensors
+        GGML_FTYPE_MOSTLY_Q4_0_ROCMFP4          = 45, // except 1d tensors
+        GGML_FTYPE_MOSTLY_Q4_0_ROCMFP4_FAST     = 46, // ROCmFP4 single-scale speed layout
     };
 
     // available tensor operations:

ggml/rocmfp4/README.md

@@ -0,0 +1,48 @@
+# ROCmFP4
+
+ROCmFP4 adds two experimental 4-bit GGUF tensor layouts intended for compact
+AMD-oriented inference formats:
+
+- `Q4_0_ROCMFP4`: 32 weights per block, packed 4-bit values, and two finite
+  unsigned E4M3 scale bytes, one scale per 16 weights. The block size is
+  18 bytes, or 4.50 bits per weight.
+- `Q4_0_ROCMFP4_FAST`: 32 weights per block, packed 4-bit values, and one
+  finite unsigned E4M3 scale byte for the full block. The block size is
+  17 bytes, or 4.25 bits per weight.
+
+The 4-bit values use a small signed codebook with levels up to `5.0` after
+the decoded scale is applied. Quantization searches all finite E4M3 scale
+candidates and keeps the lowest-error assignment. Invalid scale bytes are
+rejected during row validation so malformed tensors fail early.
+
+This directory contains the format-specific CPU reference helpers. This initial
+change adds the GGUF tensor types, row validation, quantization, and
+dequantization needed to create and inspect ROCmFP4 files from the normal
+`llama-quantize` workflow.
+
+- CPU reference quantization, dequantization, row validation, and
+  `llama-quantize` support.
+
+The feature is additive. Existing tensor types, file types, and backend
+dispatch paths are unchanged unless a tensor is explicitly stored as
+`Q4_0_ROCMFP4` or `Q4_0_ROCMFP4_FAST`.
+
+Accelerated ROCm/HIP and Vulkan execution paths can be added in follow-up
+changes after the tensor formats and CPU reference behavior are reviewed.
+
+Example quantization:
+
+```sh
+./llama-quantize model-f16.gguf model-rocmfp4.gguf Q4_0_ROCMFP4
+./llama-quantize model-f16.gguf model-rocmfp4-fast.gguf Q4_0_ROCMFP4_FAST
+```
+
+Importance matrices use the existing quantize interface:
+
+```sh
+./llama-quantize --imatrix imatrix.dat model-f16.gguf model-rocmfp4.gguf Q4_0_ROCMFP4
+```
+
+Advanced mixed recipes can be expressed with the existing
+`--tensor-type` and `--tensor-type-file` options instead of adding extra
+public file-type presets.