[Feature] Add support for MoE models in the calibration-free RTN-based quantization (#20766)

Signed-off-by: Alex Kogan <alex.kogan@oracle.com>

tests/quantization/test_rtn.py

@@ -8,7 +8,10 @@ import pytest
 
 from tests.quantization.utils import is_quant_method_supported
 
-MODELS = ["microsoft/Phi-3-mini-4k-instruct"]
+MODELS = [
+    "microsoft/Phi-3-mini-4k-instruct",  # dense model
+    "ai21labs/Jamba-tiny-dev",  # MoE model
+]
 
 
 @pytest.mark.skipif(not is_quant_method_supported("rtn"),

vllm/model_executor/layers/quantization/rtn.py

@@ -3,18 +3,19 @@
 # Copyright © 2025, Oracle and/or its affiliates.
 
 import os
-from typing import Any, Optional
+from typing import Any, Callable, Optional
 
 import torch
 import torch.nn.functional as F
 from torch.nn.parameter import Parameter
 
 from vllm.logger import init_logger
+from vllm.model_executor.layers.fused_moe import FusedMoE, FusedMoEMethodBase
 from vllm.model_executor.layers.linear import (LinearBase, LinearMethodBase,
                                                set_weight_attrs)
 from vllm.model_executor.layers.quantization import QuantizationMethods
 from vllm.model_executor.layers.quantization.base_config import (
-    QuantizationConfig)
+    QuantizationConfig, QuantizeMethodBase)
 
 logger = init_logger(__name__)
 """By default, use 8 bit as target precision, but it can be 
@@ -71,9 +72,11 @@ class RTNConfig(QuantizationConfig):
         return cls(weight_bits, group_size)
 
     def get_quant_method(self, layer: torch.nn.Module,
-                         prefix: str) -> Optional["RTNLinearMethod"]:
+                         prefix: str) -> Optional["QuantizeMethodBase"]:
         if isinstance(layer, LinearBase):
             return RTNLinearMethod(self)
+        elif isinstance(layer, FusedMoE):
+            return RTNMoEMethod(self)
         return None
 
 
@@ -94,11 +97,18 @@ class RTNTensor:
             self.data.narrow(dim, start // factor, length // factor),
             self.scale.narrow(dim, start, length), self.quant_config)
 
+    def __getitem__(self, key):
+        return RTNTensor(self.data[key], self.scale[key], self.quant_config)
+
     @property
     def shape(self):
         shape = self.data.shape
         factor = 1 if self.quant_config.weight_bits == 8 else 2
-        return torch.Size((shape[0] * factor, shape[1]))
+        batch_present = len(shape) == 3
+        if batch_present:
+            return torch.Size((shape[0], shape[1] * factor, shape[2]))
+        else:
+            return torch.Size((shape[0] * factor, shape[1]))
 
     def copy_(self, loaded_weight: torch.Tensor) -> None:
         qweight, weight_scale = rtn_quantize(loaded_weight.cuda(),
@@ -165,7 +175,7 @@ class RTNLinearMethod(LinearMethodBase):
         weight = RTNParameter(data=torch.empty(output_size_per_partition //
                                                factor,
                                                input_size_per_partition,
-                                               dtype=torch.int8),
+                                               dtype=torch.uint8),
                               scale=scale,
                               quant_config=self.quant_config)
 
@@ -180,18 +190,7 @@ class RTNLinearMethod(LinearMethodBase):
         layer.output_size_per_partition = output_size_per_partition
 
     def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
-        """torch.compile does not know how to deal with a Parameter subclass
-        (aka RTNParameter). As we don't really need RTNParameters for the
-        forward pass, we replace them with equivalent instances of Parameters.
-        """
-        old_weight = layer.weight
-        assert isinstance(old_weight, RTNParameter)
-        data = old_weight.data.data
-
-        delattr(layer, "weight")
-
-        new_weight = Parameter(data=data, requires_grad=False)
-        layer.register_parameter("weight", new_weight)
+        fix_weights(layer, "weight")
 
     def apply(self,
               layer: torch.nn.Module,
@@ -209,6 +208,128 @@ class RTNLinearMethod(LinearMethodBase):
         return out
 
 
+class RTNMoEMethod(FusedMoEMethodBase):
+
+    def __init__(self, quant_config: RTNConfig):
+        self.quant_config = quant_config
+
+    def create_weights(self, layer: torch.nn.Module, num_experts: int,
+                       hidden_size: int, intermediate_size_per_partition: int,
+                       params_dtype: torch.dtype, **extra_weight_attrs):
+
+        factor = 1 if self.quant_config.weight_bits == 8 else 2
+
+        # Fused gate_up_proj (column parallel)
+        num_groups_per_col = (hidden_size // self.quant_config.group_size
+                              if self.quant_config.group_size != -1 else 1)
+        w13_scale = Parameter(
+            torch.empty(num_experts,
+                        2 * intermediate_size_per_partition,
+                        num_groups_per_col,
+                        dtype=params_dtype),
+            requires_grad=False,
+        )
+        layer.register_parameter("w13_scale", w13_scale)
+
+        w13_weight = RTNParameter(data=torch.empty(
+            num_experts,
+            2 * intermediate_size_per_partition // factor,
+            hidden_size,
+            dtype=torch.uint8),
+                                  scale=w13_scale,
+                                  quant_config=self.quant_config)
+        layer.register_parameter("w13_weight", w13_weight)
+        set_weight_attrs(w13_weight, extra_weight_attrs)
+
+        # down_proj (row parallel)
+        num_groups_per_col = (intermediate_size_per_partition //
+                              self.quant_config.group_size
+                              if self.quant_config.group_size != -1 else 1)
+        w2_scale = Parameter(torch.zeros(num_experts,
+                                         hidden_size,
+                                         num_groups_per_col,
+                                         dtype=params_dtype),
+                             requires_grad=False)
+        layer.register_parameter("w2_scale", w2_scale)
+
+        w2_weight = RTNParameter(data=torch.empty(
+            num_experts,
+            hidden_size // factor,
+            intermediate_size_per_partition,
+            dtype=torch.uint8),
+                                 scale=w2_scale,
+                                 quant_config=self.quant_config)
+        layer.register_parameter("w2_weight", w2_weight)
+        set_weight_attrs(w2_weight, extra_weight_attrs)
+
+    def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
+        weight_bits = self.quant_config.weight_bits
+        fix_weights(layer, "w13_weight", weight_bits == 4)
+        fix_weights(layer, "w2_weight", weight_bits == 4)
+
+    def apply(
+        self,
+        layer: torch.nn.Module,
+        x: torch.Tensor,
+        router_logits: torch.Tensor,
+        top_k: int,
+        renormalize: bool,
+        use_grouped_topk: bool = False,
+        topk_group: Optional[int] = None,
+        num_expert_group: Optional[int] = None,
+        global_num_experts: int = -1,
+        expert_map: Optional[torch.Tensor] = None,
+        custom_routing_function: Optional[Callable] = None,
+        scoring_func: str = "softmax",
+        e_score_correction_bias: Optional[torch.Tensor] = None,
+        apply_router_weight_on_input: bool = False,
+        activation: str = "silu",
+        enable_eplb: bool = False,
+        expert_load_view: Optional[torch.Tensor] = None,
+        logical_to_physical_map: Optional[torch.Tensor] = None,
+        logical_replica_count: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        if enable_eplb:
+            raise NotImplementedError(
+                "EPLB not supported for `RTNMoEMethod` yet.")
+
+        from vllm.model_executor.layers.fused_moe import fused_experts
+
+        topk_weights, topk_ids = FusedMoE.select_experts(
+            hidden_states=x,
+            router_logits=router_logits,
+            use_grouped_topk=use_grouped_topk,
+            top_k=top_k,
+            renormalize=renormalize,
+            topk_group=topk_group,
+            num_expert_group=num_expert_group,
+            custom_routing_function=custom_routing_function,
+            scoring_func=scoring_func,
+            e_score_correction_bias=e_score_correction_bias)
+
+        weight_bits = self.quant_config.weight_bits
+        group_size = self.quant_config.group_size
+
+        ret = fused_experts(
+            x,
+            layer.w13_weight,
+            layer.w2_weight,
+            topk_weights=topk_weights,
+            topk_ids=topk_ids,
+            inplace=True,
+            activation=activation,
+            use_int4_w4a16=weight_bits == 4,
+            use_int8_w8a16=weight_bits == 8,
+            global_num_experts=global_num_experts,
+            w1_scale=layer.w13_scale,
+            w2_scale=layer.w2_scale,
+            apply_router_weight_on_input=apply_router_weight_on_input,
+            expert_map=expert_map,
+            block_shape=[0, group_size])
+
+        return ret
+
+
 def rtn_quantize(tensor: torch.Tensor, num_bits: int,
                  group_size: int) -> tuple[torch.Tensor, torch.Tensor]:
     """Quantize a tensor using per-group static scaling factor.
@@ -221,34 +342,44 @@ def rtn_quantize(tensor: torch.Tensor, num_bits: int,
                     If equal to -1, each row in the input tensor is treated
                     as one group.
     """
+    batch_present = len(tensor.shape) == 3
+    if not batch_present:
+        tensor = tensor.unsqueeze(0)
 
     q_range = 2**num_bits
-    num_groups = (tensor.shape[0] * tensor.shape[1] //
-                  group_size if group_size != -1 else tensor.shape[0])
+    num_groups = (tensor.shape[1] * tensor.shape[2] //
+                  group_size if group_size != -1 else tensor.shape[1])
     """Calculate a scaling factor per input group.
     """
-    input_flat = tensor.reshape(num_groups, -1)
-    input_min = torch.min(input_flat, dim=1, keepdim=True)[0]
-    input_max = torch.max(input_flat, dim=1, keepdim=True)[0]
+    input_flat = tensor.reshape(tensor.shape[0], num_groups, -1)
+    input_min = torch.min(input_flat, dim=2, keepdim=True)[0]
+    input_max = torch.max(input_flat, dim=2, keepdim=True)[0]
     input_max_abs = torch.max(input_min.abs(), input_max.abs())
     scale = (input_max_abs * 2.0 / (q_range - 1))
-    """Scale each input group, truncate and round to the nearest integer.
+    """Scale each input group, round to the nearest integer, shift 
+    the range and truncate.
     """
     scaled_input = input_flat / scale
-    scaled_input = scaled_input.clamp(-q_range // 2, q_range // 2 - 1)
     scaled_input = scaled_input.round()
+    scaled_input += q_range // 2
+    scaled_input = scaled_input.clamp(0, q_range - 1)
 
-    scale = scale.reshape(tensor.shape[0], -1).contiguous()
-    inputs_q = scaled_input.reshape(tensor.shape).to(torch.int8)
+    scale = scale.reshape(tensor.shape[0], tensor.shape[1], -1).contiguous()
+    inputs_q = scaled_input.reshape(tensor.shape).to(torch.uint8)
     inputs_q = inputs_q.contiguous()
 
     if num_bits == 4:
         """Pack two 4-bit values into each byte.
         """
-        inputs_q = (inputs_q[:, 1::2] << 4) | (inputs_q[:, ::2] & 0xf)
-        inputs_q = inputs_q.reshape(tensor.shape[0] // 2, tensor.shape[1])
+        inputs_q = (inputs_q[:, :, 1::2] << 4) | (inputs_q[:, :, ::2] & 0xf)
+        inputs_q = inputs_q.reshape(tensor.shape[0], tensor.shape[1] // 2,
+                                    tensor.shape[2])
         inputs_q = inputs_q.contiguous()
 
+    if not batch_present:
+        inputs_q = inputs_q.squeeze(0)
+        scale = scale.squeeze(0)
+
     return inputs_q, scale
 
 
@@ -259,31 +390,60 @@ def rtn_dequantize(tensor: torch.Tensor, scale: torch.Tensor) -> torch.Tensor:
         tensor: The input tensor.
         scale: The tensor with per-group scale factors.
     """
+    batch_present = len(tensor.shape) == 3
+    if not batch_present:
+        tensor = tensor.unsqueeze(0)
+        scale = scale.unsqueeze(0)
 
-    num_groups = scale.size(0) * scale.size(1)
-    input_dim, output_dim = tensor.shape
+    num_groups = scale.size(1) * scale.size(2)
+    batch, input_dim, output_dim = tensor.shape
 
-    num_bits = 8 if input_dim == scale.size(0) else 4
+    num_bits = 8 if input_dim == scale.size(1) else 4
+    q_range = 2**num_bits
     if num_bits == 4:
         input_dim *= 2
 
-    data = torch.empty((input_dim, output_dim),
+    data = torch.empty((batch, input_dim, output_dim),
                        dtype=scale.dtype,
                        device=tensor.device)
 
     if num_bits == 8:
         data.copy_(tensor)
+        data -= q_range // 2
     else:
         """Unpack two 4-bit values from each byte.
         """
-        tensor = tensor.reshape(input_dim, output_dim // 2)
+        tensor = tensor.reshape(batch, input_dim, output_dim // 2)
         for i in range(2):
-            data[:, i::2] = (tensor << 4 * (1 - i)) >> 4
+            data[:, :, i::2] = ((tensor << 4 *
+                                 (1 - i)) >> 4).to(torch.int8) - q_range // 2
     """Scale each input group with its scaling factor.
     """
-    scale = scale.reshape(num_groups, -1)
-    data = data.reshape(num_groups, -1)
+    scale = scale.reshape(batch, num_groups, -1)
+    data = data.reshape(batch, num_groups, -1)
     data = torch.mul(data, scale)
 
-    input_deq = data.reshape((input_dim, output_dim)).contiguous()
+    input_deq = data.reshape((batch, input_dim, output_dim)).contiguous()
+    if not batch_present:
+        input_deq = input_deq.squeeze(0)
+
     return input_deq
+
+
+def fix_weights(layer: torch.nn.Module,
+                param_name: str,
+                reshape: bool = False):
+    """torch.compile does not know how to deal with a Parameter subclass
+    (aka RTNParameter). As we don't really need RTNParameters for the
+    forward pass, we replace them with equivalent instances of Parameters.
+    """
+    old_weight = getattr(layer, param_name)
+    assert isinstance(old_weight, RTNParameter)
+    data = old_weight.data.data
+
+    delattr(layer, param_name)
+
+    if reshape:
+        data = data.reshape(old_weight.shape[0], old_weight.shape[1] * 2, -1)
+    new_weight = Parameter(data=data, requires_grad=False)
+    layer.register_parameter(param_name, new_weight)