[Quantization/NVFP4] Speed up TRTLLM NVFP4 MOE weight loading and fix K/V scale loading for MLA Attn (vllm-project#25968)

Signed-off-by: Pavani Majety <[email protected]>

Author: pavanimajety

vllm/model_executor/layers/quantization/kv_cache.py

@@ -86,7 +86,7 @@ def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
                 logger.warning_once(
                     "Checkpoint does not provide a q scaling factor. "
                     "Setting it to k_scale. This only matters for "
-                    "the flash-attn backend.")
+                    "FP8 Attention backends (flash-attn or flashinfer).")
                 layer._q_scale.copy_(k_scale)
                 layer._q_scale_float = k_scale
 
@@ -98,9 +98,9 @@ def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
             if (k_scale == 1.0 and v_scale == 1.0
                     and "e5m2" not in layer.kv_cache_dtype):
                 logger.warning_once(
-                    "Using KV cache scaling factor 1.0 for fp8_e4m3. This "
-                    "may cause accuracy issues. Please make sure k/v_scale "
-                    "scaling factors are available in the fp8 checkpoint.")
+                    "Using KV cache scaling factor 1.0 for fp8_e4m3. "
+                    "If this is unintended, verify that k/v_scale "
+                    "scaling factors are properly set in the checkpoint.")
 
         if layer.q_scale > 0.0:
             q_scale = layer.q_scale

vllm/model_executor/layers/quantization/modelopt.py

@@ -1064,7 +1064,7 @@ def __init__(
         self.allow_flashinfer = _nvfp4.allow_flashinfer
         self.use_marlin = _nvfp4.use_marlin
         self.flashinfer_moe_backend = None
-
+        self._cache_permute_indices: dict[torch.Size, torch.Tensor] = {}
         if self.allow_flashinfer:
             self.flashinfer_moe_backend = get_flashinfer_moe_backend()
             logger.info_once(
@@ -1197,19 +1197,23 @@ def create_weights(self, layer: torch.nn.Module, num_experts: int,
                                                  weight_loader=weight_loader)
         layer.register_parameter("w2_input_scale", w2_input_scale)
 
-    def prepare_static_weight_layouts_for_trtllm_moe(
+    def prepare_static_weights_for_trtllm_fp4_moe(
         self,
-        gemm1_weights: torch.Tensor,
-        gemm2_weights: torch.Tensor,
-        gemm1_scales_linear_fp4_bytes: torch.Tensor,
-        gemm2_scales_linear_fp4_bytes: torch.Tensor,
-        hidden_size: int,
-        intermediate_size: int,
-        num_experts: int,
-    ) -> tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]:
+        # args_dequant,
+        # args,
+        gemm1_weights,
+        gemm2_weights,
+        gemm1_scales_linear_fp4_bytes,
+        gemm2_scales_linear_fp4_bytes,
+        hidden_size,
+        intermediate_size,
+        num_experts,
+    ):
+        from flashinfer import nvfp4_block_scale_interleave
+        from flashinfer.fused_moe.core import (
+            _maybe_get_cached_w2_permute_indices,
+            _maybe_get_cached_w3_w1_permute_indices)
         """Prepare quantized weights for kernel (done offline with weights)."""
-        from flashinfer import (reorder_rows_for_gated_act_gemm,
-                                shuffle_matrix_a, shuffle_matrix_sf_a)
         epilogue_tile_m = 128  # FIXME: this depends on the kernel internals
 
         # Convert quantized weights to proper formats
@@ -1227,48 +1231,54 @@ def prepare_static_weight_layouts_for_trtllm_moe(
                                          intermediate_size //
                                          16)  # fp8 scaling factors
 
-        # Reorder rows of W1 and scales for fused gated activation
-        gemm1_weights_fp4_interleaved = []
-        gemm1_scales_fp4_interleaved = []
-        for i in range(num_experts):
-            gemm1_weights_fp4_interleaved.append(
-                reorder_rows_for_gated_act_gemm(gemm1_weights_fp4[i].clone()))
-            gemm1_scales_fp4_interleaved.append(
-                reorder_rows_for_gated_act_gemm(
-                    gemm1_scales_linear_fp4[i].clone()))
-
-        # Stack weights and scales for all experts
-        gemm1_weights_fp4_interleaved = torch.stack(
-            gemm1_weights_fp4_interleaved).reshape(num_experts,
-                                                   2 * intermediate_size,
-                                                   hidden_size // 2)
-        gemm1_scales_fp4_interleaved = torch.stack(
-            gemm1_scales_fp4_interleaved).reshape(num_experts,
-                                                  2 * intermediate_size,
-                                                  hidden_size // 16)
-
-        # Shuffle weights and scaling factors for transposed mma output
         gemm1_weights_fp4_shuffled = []
         gemm1_scales_fp4_shuffled = []
         gemm2_weights_fp4_shuffled = []
         gemm2_scales_fp4_shuffled = []
         for i in range(num_experts):
-            gemm1_weights_fp4_shuffled.append(
-                shuffle_matrix_a(
-                    gemm1_weights_fp4_interleaved[i].view(torch.uint8),
-                    epilogue_tile_m))
+            # Calculate the permute indices for the following:
+            # 1. Reorder rows of W1 and scales for fused gated activation
+            # 2. Shuffle weights and scaling factors for transposed mma output
+            # for both w3_w1 and w2 weights and scale factors
+            permute_indices = _maybe_get_cached_w3_w1_permute_indices(
+                self._cache_permute_indices,
+                gemm1_weights_fp4[i].view(torch.uint8),
+                epilogue_tile_m,
+            )
+            gemm1_weights_fp4_shuffled.append(gemm1_weights_fp4[i].view(
+                torch.uint8)[permute_indices.to(
+                    gemm1_weights_fp4.device)].contiguous())
+
+            permute_sf_indices = _maybe_get_cached_w3_w1_permute_indices(
+                self._cache_permute_indices,
+                gemm1_scales_linear_fp4[i].view(torch.uint8),
+                epilogue_tile_m,
+                num_elts_per_sf=16,
+            )
             gemm1_scales_fp4_shuffled.append(
-                shuffle_matrix_sf_a(
-                    gemm1_scales_fp4_interleaved[i].view(torch.uint8),
-                    epilogue_tile_m))
-
-            gemm2_weights_fp4_shuffled.append(
-                shuffle_matrix_a(gemm2_weights_fp4[i].view(torch.uint8),
-                                 epilogue_tile_m))
+                nvfp4_block_scale_interleave(gemm1_scales_linear_fp4[i].view(
+                    torch.uint8)[permute_sf_indices.to(
+                        gemm1_scales_linear_fp4.device)].contiguous()))
+
+            permute_indices = _maybe_get_cached_w2_permute_indices(
+                self._cache_permute_indices,
+                gemm2_weights_fp4[i].view(torch.uint8),
+                epilogue_tile_m,
+            )
+            gemm2_weights_fp4_shuffled.append(gemm2_weights_fp4[i].view(
+                torch.uint8)[permute_indices.to(
+                    gemm2_weights_fp4.device)].contiguous())
+
+            permute_sf_indices = _maybe_get_cached_w2_permute_indices(
+                self._cache_permute_indices,
+                gemm2_scales_linear_fp4[i].view(torch.uint8),
+                epilogue_tile_m,
+                num_elts_per_sf=16,
+            )
             gemm2_scales_fp4_shuffled.append(
-                shuffle_matrix_sf_a(
-                    gemm2_scales_linear_fp4[i].view(torch.uint8),
-                    epilogue_tile_m))
+                nvfp4_block_scale_interleave(gemm2_scales_linear_fp4[i].view(
+                    torch.uint8)[permute_sf_indices.to(
+                        gemm2_scales_linear_fp4.device)].contiguous()))
 
         # Stack weights for all experts
         gemm1_weights_fp4_shuffled = torch.stack(gemm1_weights_fp4_shuffled)
@@ -1283,8 +1293,12 @@ def prepare_static_weight_layouts_for_trtllm_moe(
             torch.stack(gemm2_scales_fp4_shuffled).view(
                 torch.float8_e4m3fn).reshape(num_experts, hidden_size,
                                              intermediate_size // 16))
-        return (gemm1_weights_fp4_shuffled, gemm1_scales_fp4_shuffled,
-                gemm2_weights_fp4_shuffled, gemm2_scales_fp4_shuffled)
+        return (
+            gemm1_weights_fp4_shuffled,
+            gemm1_scales_fp4_shuffled,
+            gemm2_weights_fp4_shuffled,
+            gemm2_scales_fp4_shuffled,
+        )
 
     def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
         # GEMM 1 processing
@@ -1334,9 +1348,10 @@ def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
         if self.allow_flashinfer and \
             self.flashinfer_moe_backend == FlashinferMoeBackend.TENSORRT_LLM:
             # Prepare static weights for TRT-LLM kernel
+            # alternate: prepare_static_weight_layouts_for_trtllm_moe
             (gemm1_weights_fp4_shuffled, gemm1_scales_fp4_shuffled,
              gemm2_weights_fp4_shuffled, gemm2_scales_fp4_shuffled
-             ) = self.prepare_static_weight_layouts_for_trtllm_moe(
+             ) = self.prepare_static_weights_for_trtllm_fp4_moe(
                  layer.w13_weight,
                  layer.w2_weight,
                  layer.w13_weight_scale,
@@ -1345,6 +1360,7 @@ def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
                  layer.w13_weight.size(-2) // 2,  # intermediate_size
                  layer.w13_weight.size(0),  # num_experts
              )
+            logger.debug_once("Finished shuffling weights for TRT-LLM MOE")
 
             layer.gemm1_weights_fp4_shuffled = Parameter(
                 gemm1_weights_fp4_shuffled, requires_grad=False)

vllm/model_executor/model_loader/weight_utils.py

@@ -1003,12 +1003,18 @@ def maybe_remap_kv_scale_name(name: str, params_dict: dict) -> Optional[str]:
             return None
         return remapped_name
 
+    if any("mla_attn" in key for key in params_dict):
+        attn_str = "mla_attn.mla_attn"
+        logger.debug_once(f"Found mla_attn with k_scale and v_scale in "
+                          f"the checkpoint, using {attn_str} as attn_str")
+    else:
+        attn_str = "attn"
     # Define scale name mapping patterns in order of precedence
     scale_mapping_patterns = [
         # ModelOpt format: .self_attn.{k,v}_proj.{k,v}_scale ->
         # .self_attn.attn.{k,v}_scale
         (r"\.self_attn\.([kv])_proj\.([kv])_scale$",
-         r".self_attn.attn.\2_scale"),
+         rf".self_attn.{attn_str}.\2_scale"),
         # QKV proj format: .self_attn.qkv_proj.{k,v}_scale ->
         # .self_attn.attn.{k,v}_scale
         (r"\.self_attn\.qkv_proj\.([kv])_scale$", r".self_attn.attn.\1_scale"),