NVFP4 Emulation

run_fp4.py

@@ -0,0 +1,24 @@
+
+import numpy
+import torch
+
+from vllm import LLM, SamplingParams
+
+prompts = [
+    "The Swiss Alps are", "The president of the USA is",
+    "The Boston Bruins are"
+]
+
+# Create a sampling params object for greedy sampling
+sampling_params = SamplingParams(temperature=0.80, top_p=0.95, max_tokens=40, min_tokens=10)
+#llm  = LLM('nm-testing/Llama-3.1-8B-Instruct-FP4-Weight')
+#llm = LLM("/home/dsikka/llm-compressor/examples/quantization_w4a16_fp4/TinyLlama-1.1B-Chat-v1.0-FP4")
+#llm = LLM("/home/dsikka/llm-compressor/examples/quantization_w4a16_fp4/Llama-3.1-8B-Instruct-NVFP4A16")
+#llm = LLM("/home/dsikka/llm-compressor/examples/quantization_w4a16_fp4/Llama-3.1-8B-Instruct-NVFP4A16-MSE")
+#llm = LLM("nm-testing/Llama-3.3-70B-Instruct-NVFP4A16", max_model_len=4096)
+# Print the outputs.
+llm = LLM("nvidia/Llama-3.3-70B-Instruct-FP4", max_model_len=4096, quantization="nvfp4", enforce_eager=True)
+output = llm.generate(prompts, sampling_params)
+for o in output:
+    print(o.outputs[0].text)
+    print("\n")

vllm/model_executor/layers/quantization/compressed_tensors/compressed_tensors.py

@@ -24,7 +24,7 @@
     W4A16SPARSE24_SUPPORTED_BITS, WNA16_SUPPORTED_BITS, CompressedTensors24,
     CompressedTensorsScheme, CompressedTensorsW4A16Sparse24,
     CompressedTensorsW8A8Fp8, CompressedTensorsW8A8Int8,
-    CompressedTensorsW8A16Fp8, CompressedTensorsWNA16)
+    CompressedTensorsW8A16Fp8, CompressedTensorsWNA16, CompressedTensorsW4A4Fp4)
 from vllm.model_executor.layers.quantization.compressed_tensors.utils import (
     find_matched_target, is_activation_quantization_format,
     should_ignore_layer)
@@ -299,6 +299,13 @@
         # All conditions satisfied.
         return True
 
+    def _is_fp4_nvfp4(self, weight_quant: BaseModel, input_quant: BaseModel):
+        is_group_quant = weight_quant.strategy == QuantizationStrategy.GROUP.value
+        is_group_size_16 = weight_quant.group_size == 16
+        is_float_type = weight_quant.type == QuantizationType.FLOAT
+        is_4_bits = weight_quant.num_bits == 4
+        return is_group_quant and is_float_type and is_4_bits
+
     def _is_wNa16_group_channel(self, weight_quant: BaseModel,
                                 input_quant: BaseModel) -> bool:
         input_quant_none = input_quant is None
@@ -313,6 +320,8 @@
             self, weight_quant: BaseModel,
             input_quant: BaseModel) -> "CompressedTensorsScheme":
 
+        if self._is_fp4_nvfp4(weight_quant, input_quant):
+            return CompressedTensorsW4A4Fp4()
         # Detect If Mixed Precision
         if self._is_wNa16_group_channel(weight_quant, input_quant):
             if (self.quant_format == CompressionFormat.marlin_24.value

vllm/model_executor/layers/quantization/compressed_tensors/schemes/__init__.py

@@ -8,6 +8,7 @@
 from .compressed_tensors_w8a16_fp8 import CompressedTensorsW8A16Fp8
 from .compressed_tensors_wNa16 import (WNA16_SUPPORTED_BITS,
                                        CompressedTensorsWNA16)
+from .compressed_tensors_w4a4_nvfp4 import CompressedTensorsW4A4Fp4
 
 from .compressed_tensors_24 import CompressedTensors24  # isort: skip
 
@@ -16,5 +17,5 @@
     "CompressedTensorsW8A16Fp8", "CompressedTensorsW4A16Sparse24",
     "CompressedTensorsW8A8Int8", "CompressedTensorsW8A8Fp8",
     "WNA16_SUPPORTED_BITS", "W4A16SPARSE24_SUPPORTED_BITS",
-    "CompressedTensors24"
+    "CompressedTensors24", "CompressedTensorsW4A4Fp4"
 ]

vllm/model_executor/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a4_nvfp4.py

@@ -0,0 +1,113 @@
+# SPDX-License-Identifier: Apache-2.0
+
+from typing import Callable, List, Optional
+
+import torch
+import torch.nn.functional as F
+from torch.nn.parameter import Parameter
+
+from vllm.model_executor.layers.quantization.compressed_tensors.schemes import (
+    CompressedTensorsScheme)
+from vllm.model_executor.layers.quantization.utils.nvfp4_emulation_utils import (
+    dequantize_to_dtype)
+from vllm.model_executor.parameter import (GroupQuantScaleParameter,
+                                           ModelWeightParameter,
+                                           PerTensorScaleParameter)
+
+__all__ = ["CompressedTensorsW4A4Fp4"]
+
+
+class CompressedTensorsW4A4Fp4(CompressedTensorsScheme):
+
+    def __init__(self):
+        self.group_size = 16
+
+    @classmethod
+    def get_min_capability(cls) -> int:
+        # dont restrict as emulations
+        return 80
+
+    def create_weights(self, layer: torch.nn.Module,
+                       output_partition_sizes: List[int],
+                       input_size_per_partition: int,
+                       params_dtype: torch.dtype, weight_loader: Callable,
+                       **kwargs):
+
+        # Weight
+        self.output_partition_sizes = output_partition_sizes
+        self.params_dtype = params_dtype
+        weight = ModelWeightParameter(
+            data=torch.empty(
+                # 2 fp4 items are packed in the input dimension
+                sum(output_partition_sizes),
+                input_size_per_partition // 2,
+                dtype=torch.uint8),
+            input_dim=1,
+            output_dim=0,
+            weight_loader=weight_loader)
+        layer.register_parameter("weight_packed", weight)
+
+        # Global Weight Scale
+        weight_global_scale = PerTensorScaleParameter(
+            data=torch.empty(len(output_partition_sizes), dtype=torch.float32),
+            weight_loader=weight_loader)
+        layer.register_parameter("weight_global_scale", weight_global_scale)
+
+        # Per Group Weight Scale
+        weight_scale = GroupQuantScaleParameter(data=torch.empty(
+            sum(output_partition_sizes),
+            input_size_per_partition // self.group_size,
+            dtype=torch.float8_e4m3fn,
+        ),
+                                                input_dim=1,
+                                                output_dim=0,
+                                                weight_loader=weight_loader)
+
+        layer.register_parameter("weight_scale", weight_scale)
+
+    def swizzle_blockscale(self, scale: torch.tensor):
+        assert (scale.dtype == torch.float8_e4m3fn)
+        # Pad and blockwise interleave weight_scale
+        scale_ndim = scale.ndim
+        if scale.ndim == 2:
+            scale = scale.unsqueeze(0)
+        assert scale.ndim == 3
+        B, M, K = scale.shape
+        round_up_multiple = lambda x, m: (x + m - 1) // m * m
+        M_padded = round_up_multiple(M, 128)
+        K_padded = round_up_multiple(K, 4)
+        padded_scale = torch.zeros((B, M_padded, K_padded), dtype=scale.dtype)
+        padded_scale[:B, :M, :K] = scale
+        batches, rows, cols = padded_scale.shape
+        assert rows % 128 == 0
+        assert cols % 4 == 0
+        padded_scale = padded_scale.reshape(batches, rows // 128, 4, 32,
+                                            cols // 4, 4)
+        swizzled_scale = padded_scale.permute((0, 1, 4, 3, 2, 5))
+        swizzled_scale = swizzled_scale.contiguous().cuda()
+        return (swizzled_scale.reshape(M, K)
+                if scale_ndim == 2 else swizzled_scale.reshape(B, M, K))
+
+    def process_weights_after_loading(self, layer) -> None:
+        print(layer.weight_global_scale)
+        layer.weight_global_scale = Parameter(
+            layer.weight_global_scale.max().to(torch.float32),
+            requires_grad=False)
+        # Note: a post weight loading step but not required for the emulation
+        swizzled_weight_scale = self.swizzle_blockscale(layer.weight_scale)
+        layer.weight_scale_swizzled = Parameter(swizzled_weight_scale,
+                                                requires_grad=False)
+
+    def apply_weights(self,
+                      layer: torch.nn.Module,
+                      x: torch.Tensor,
+                      bias: Optional[torch.Tensor] = None) -> torch.Tensor:
+
+        w_fp4 = layer.weight_packed.data
+        w_global_scale = layer.weight_global_scale
+        w_blockscale = layer.weight_scale_swizzled.data
+        w_dq = dequantize_to_dtype(w_fp4, w_blockscale, w_global_scale,
+                                   x.dtype, x.device, self.group_size)
+        out = F.linear(x, w_dq)
+        del w_dq
+        return out

vllm/model_executor/layers/quantization/modelopt.py

@@ -3,6 +3,7 @@
 from typing import Any, Dict, List, Optional, Union
 
 import torch
+import torch._dynamo
 from torch.nn import Module
 from torch.nn.parameter import Parameter
 
@@ -13,116 +14,23 @@
 from vllm.model_executor.layers.quantization.base_config import (
     QuantizationConfig, QuantizeMethodBase)
 from vllm.model_executor.layers.quantization.kv_cache import BaseKVCacheMethod
+from vllm.model_executor.layers.quantization.utils.nvfp4_emulation_utils import (
+    dequantize_to_dtype, ref_nvfp4_quant)
 from vllm.model_executor.layers.quantization.utils.quant_utils import (
     is_layer_skipped)
 from vllm.model_executor.layers.quantization.utils.w8a8_utils import (
     Fp8LinearOp, requantize_with_max_scale)
 from vllm.model_executor.parameter import (ModelWeightParameter,
                                            PerTensorScaleParameter)
 from vllm.platforms import current_platform
-from vllm.scalar_type import scalar_types
+
+torch._dynamo.config.suppress_errors = True
 
 logger = init_logger(__name__)
 
 QUANT_ALGOS = ["FP8", "NVFP4"]
 KV_CACHE_QUANT_ALGOS = ["FP8"]
 
-FLOAT4_E2M1_MAX = scalar_types.float4_e2m1fn.max()
-
-kE2M1ToFloat = torch.tensor([0., 0.5, 1., 1.5, 2., 3., 4., 6.],
-                            dtype=torch.float32)
-
-
-def break_fp4_bytes(a, dtype):
-    assert a.dtype == torch.uint8
-    m, n = a.shape
-    # Vectorized nibble processing
-    a_flat = a.flatten()
-    high = (a_flat & 0xF0) >> 4  # Upper nibbles
-    low = a_flat & 0x0F  # Lower nibbles
-    # Combine nibbles for batch processing
-    combined = torch.stack((low, high), dim=1).flatten()
-    # Vectorized sign and magnitude extraction
-    signs = (combined & 0x08).to(torch.bool)  # Sign bits
-    abs_vals = (combined & 0x07).to(torch.long)
-    # Device-aware lookup and sign application
-    kE2M1 = kE2M1ToFloat.to(device=a.device)
-    values = kE2M1[abs_vals] * torch.where(signs, -1.0, 1.0)
-    # Reshape to final form
-    return values.reshape(m, n * 2).to(dtype=dtype)
-
-
-def convert_swizzled_to_linear(a_sf_swizzled: torch.Tensor, m, k, block_size):
-    m_tiles = (m + 128 - 1) // 128
-    f = block_size * 4
-    k_tiles = (k + f - 1) // f
-    tmp = torch.reshape(a_sf_swizzled, (1, m_tiles, k_tiles, 32, 4, 4))
-    tmp = torch.permute(tmp, (0, 1, 4, 3, 2, 5))
-    out = tmp.reshape(m_tiles * 128, k_tiles * f // block_size)
-    return out[0:m, 0:k]
-
-
-def dequantize_to_dtype(tensor_fp4,
-                        tensor_sf,
-                        global_scale,
-                        dtype,
-                        device,
-                        block_size=16):
-    """Dequantize the fp4 tensor back to high precision."""
-    # Two fp4 values are packed into one uint8.
-    assert tensor_fp4.dtype == torch.uint8
-    m, packed_k = tensor_fp4.shape
-    k = packed_k * 2
-    tensor_f32 = break_fp4_bytes(tensor_fp4, dtype)
-    tensor_f32 = tensor_f32.reshape(m, k // block_size, block_size)
-    tensor_sf = tensor_sf.view(torch.float8_e4m3fn)
-    tensor_sf = convert_swizzled_to_linear(tensor_sf, m, k, block_size)
-    tensor_sf_dtype = tensor_sf.to(torch.float32) / global_scale
-
-    # scale the tensor
-    out = (tensor_f32 * tensor_sf_dtype.unsqueeze(-1)).reshape(m, k)
-    return out.to(dtype)
-
-
-def cast_to_fp4(x):
-    sign = torch.sign(x)
-    x = torch.abs(x)
-    x[(x >= 0.0) & (x <= 0.25)] = 0.0
-    x[(x > 0.25) & (x < 0.75)] = 0.5
-    x[(x >= 0.75) & (x <= 1.25)] = 1.0
-    x[(x > 1.25) & (x < 1.75)] = 1.5
-    x[(x >= 1.75) & (x <= 2.5)] = 2.0
-    x[(x > 2.5) & (x < 3.5)] = 3.0
-    x[(x >= 3.5) & (x <= 5.0)] = 4.0
-    x[x > 5.0] = 6.0
-    return x * sign
-
-
-def get_reciprocal(x):
-    if isinstance(x, torch.Tensor):
-        return torch.where(x == 0, torch.tensor(0.0, dtype=x.dtype), 1.0 / x)
-    elif isinstance(x, (float, int)):
-        return 0.0 if x == 0 else 1.0 / x
-    else:
-        raise TypeError("Input must be a float, int, or a torch.Tensor.")
-
-
-def ref_nvfp4_quant(x, global_scale, block_size):
-    assert global_scale.dtype == torch.float32
-    assert x.ndim == 2
-    m, n = x.shape
-    x = torch.reshape(x, (m, n // block_size, block_size))
-    vec_max = torch.max(torch.abs(x), dim=-1,
-                        keepdim=True)[0].to(torch.float32)
-    scale = global_scale * (vec_max * get_reciprocal(FLOAT4_E2M1_MAX))
-    scale = scale.to(torch.float8_e4m3fn).to(torch.float32)
-    output_scale = get_reciprocal(scale * get_reciprocal(global_scale))
-
-    scaled_x = x.to(torch.float32) * output_scale
-    clipped_x = torch.clamp(scaled_x, -6.0, 6.0).reshape(m, n)
-    # both outputs are float32
-    return cast_to_fp4(clipped_x), scale.squeeze(-1)
-
 
 class ModelOptFp8Config(QuantizationConfig):
     """Config class for ModelOpt FP8."""
@@ -289,7 +197,7 @@
 
     @classmethod
     def get_min_capability(cls) -> int:
-        return 89
+        return 80
 
     @classmethod
     def get_config_filenames(cls) -> List[str]:
@@ -483,20 +391,11 @@
 
         # for input only the contracting dimension has a constraint.
         x_m, x_k = x.shape
-        w_n, w_k = layer.weight.shape
-        # print(f"{x.shape=}")
-        # print(f"{layer.weight.shape=}")
-        output_shape = [x_m, w_n]
-        block_size = 16
+        block_size = group_size = 16
 
         # quantize input to (FP4 and interleaved block scale)
-        # x_global_scale = layer.input_scale
         x_global_scale = 1 / layer.input_scale
-        # x_fp4, x_blockscale = scaled_fp4_quant(x, s_quant)
         x_fp4, x_blockscale = ref_nvfp4_quant(x, x_global_scale, block_size)
-        # x_blockscale = self.swizzle_blockscale(x_blockscale)
-        # print(f"{x_fp4.shape=}")
-        # print(f"{x_blockscale.shape=}")
 
         # dequantize input
         x_fp4 = x_fp4.reshape(x_m, x_k // block_size, block_size)
@@ -507,20 +406,11 @@
         # dequantize weight
         w_fp4 = layer.weight.data.view(torch.uint8)
         w_blockscale = layer.weight_scale_swizzled.data
-        w_global_scale = layer.weight_scale_2
-        # print(f"{w_fp4.shape=}")
-        # print(f"{w_blockscale.shape=}")
-        # print(f"{w_global_scale.shape=}")
+        w_global_scale = 1 / layer.weight_scale_2
         w_dq = dequantize_to_dtype(w_fp4, w_blockscale, w_global_scale,
-                                   output_dtype, x.device,
-                                   block_size).to(output_dtype)
-        # print(f"{w_dq.shape=}")
+                                   output_dtype, x.device, block_size)
 
         # matmul
         out = torch.matmul(x_dq, w_dq.t())
-        del x_dq, w_dq
-        # print(f"{out.shape=}")
-
-        if bias is not None:
-            out = out + bias
-        return out.view(*output_shape)
+        del w_dq, x_dq
+        return out