[V1] [Hybrid] Enable compile and piecewise CUDA graph for MiniMax-Text models

vllm/config/compilation.py

@@ -425,4 +425,5 @@ def set_splitting_ops_for_v1(self):
                 "vllm.unified_attention",
                 "vllm.unified_attention_with_output",
                 "vllm.mamba_mixer2",
+                "vllm.linear_attention",
             ]

vllm/model_executor/models/minimax_text_01.py

@@ -1,7 +1,6 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 """Inference-only MiniMaxText01 model."""
-import copy
 import math
 from collections.abc import Iterable
 from typing import Optional, Union
@@ -16,12 +15,13 @@
 
 from vllm import envs
 from vllm.attention import Attention, AttentionMetadata
+from vllm.compilation.decorators import support_torch_compile
 from vllm.config import CacheConfig, VllmConfig, get_current_vllm_config
 from vllm.distributed.communication_op import tensor_model_parallel_all_reduce
 from vllm.distributed.parallel_state import (
     get_pp_group, get_tensor_model_parallel_rank,
     get_tensor_model_parallel_world_size)
-from vllm.forward_context import get_forward_context
+from vllm.forward_context import ForwardContext, get_forward_context
 from vllm.model_executor.custom_op import CustomOp
 from vllm.model_executor.layers.activation import SiluAndMul
 from vllm.model_executor.layers.fused_moe import FusedMoE
@@ -44,7 +44,9 @@
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
 from vllm.model_executor.models.utils import maybe_prefix
 from vllm.model_executor.sampling_metadata import SamplingMetadata
+from vllm.platforms import current_platform
 from vllm.sequence import IntermediateTensors
+from vllm.utils import direct_register_custom_op
 from vllm.v1.attention.backends.linear_attn import LinearAttentionMetadata
 
 from .interfaces import HasInnerState, IsHybrid
@@ -507,20 +509,40 @@ def _decode_infer(self, q, k, v, kv_cache, state_indices_tensor,
                                               slot_id, 32)
         return hidden
 
-    def forward(self, hidden_states: torch.Tensor, positions: torch.Tensor,
-                kv_caches: MinimaxCacheParams, **kwargs) -> torch.Tensor:
-        qkv, _ = self.qkv_proj(hidden_states)
+    def forward(self, hidden_states: torch.Tensor, output: torch.Tensor,
+                positions: torch.Tensor,
+                kv_caches: MinimaxCacheParams) -> torch.Tensor:
+        if not envs.VLLM_USE_V1:
+            self._forward(hidden_states, output, positions, kv_caches)
+        else:
+            torch.ops.vllm.linear_attention(
+                hidden_states,
+                output,
+                positions,
+                self.prefix,
+            )
+
+    def _forward(self, hidden_states: torch.Tensor, output: torch.Tensor,
+                 positions: torch.Tensor,
+                 kv_caches: MinimaxCacheParams) -> torch.Tensor:
+        forward_context = get_forward_context()
+        attn_metadata: AttentionMetadata = forward_context.attn_metadata
+        if envs.VLLM_USE_V1 and attn_metadata is not None:
+            assert isinstance(attn_metadata, dict)
+            attn_metadata = attn_metadata[self.prefix]
+            assert isinstance(attn_metadata, LinearAttentionMetadata)
+            num_actual_tokens = attn_metadata.num_prefill_tokens + \
+                attn_metadata.num_decode_tokens
+        else:
+            num_actual_tokens = hidden_states.shape[0]
+
+        qkv, _ = self.qkv_proj(hidden_states[:num_actual_tokens])
         qkv32 = qkv.to(torch.float32)
         qkvact = torch.nn.functional.silu(qkv32)
         qkvact = qkvact.view((qkv.shape[0], self.tp_heads, -1))
         q, k, v = torch.split(qkvact, [self.head_dim] * 3, dim=-1)
-        forward_context = get_forward_context()
-        attn_metadata = forward_context.attn_metadata
         if envs.VLLM_USE_V1:
             if attn_metadata is not None:
-                assert isinstance(attn_metadata, dict)
-                attn_metadata = attn_metadata[self.prefix]
-                assert isinstance(attn_metadata, LinearAttentionMetadata)
                 kv_cache = self.kv_cache[forward_context.virtual_engine][0]
                 state_indices_tensor = attn_metadata.state_indices_tensor
 
@@ -559,13 +581,11 @@ def forward(self, hidden_states: torch.Tensor, positions: torch.Tensor,
                 hidden = self._decode_infer(q, k, v, kv_cache,
                                             state_indices_tensor,
                                             attn_metadata)
-
         hidden = self.norm._forward(hidden)
-        gate, _ = self.output_gate(hidden_states)
+        gate, _ = self.output_gate(hidden_states[:num_actual_tokens])
         hidden = F.sigmoid(gate) * hidden
         hidden = hidden.to(hidden_states.dtype)
-        hidden, _ = self.out_proj(hidden)
-        return hidden
+        output[:num_actual_tokens], _ = self.out_proj(hidden)
 
 
 class MiniMaxText01Attention(nn.Module):
@@ -635,8 +655,8 @@ def __init__(
         )
         return
 
-    def forward(self, hidden_states: torch.Tensor, positions: torch.Tensor,
-                **kwargs) -> torch.Tensor:
+    def forward(self, hidden_states: torch.Tensor, output: torch.Tensor,
+                positions: torch.Tensor, **kwargs) -> None:
         forward_context = get_forward_context()
         attn_metadata = forward_context.attn_metadata
         qkv, _ = self.qkv_proj(hidden_states)
@@ -648,8 +668,7 @@ def forward(self, hidden_states: torch.Tensor, positions: torch.Tensor,
         else:
             q, k = attn_metadata.rotary_emb(positions, q, k)
         attn_output = self.attn(q, k, v)
-        output, _ = self.o_proj(attn_output)
-        return output
+        output[:], _ = self.o_proj(attn_output)
 
 
 class MiniMaxText01DecoderLayer(nn.Module):
@@ -794,16 +813,15 @@ def forward(self,
                 is_warmup: bool = False,
                 **kwargs) -> tuple[torch.Tensor, torch.Tensor]:
 
-        forward_context = get_forward_context()
-        attn_metadata = forward_context.attn_metadata
         layernorm_input = hidden_states
         layernorm_output = self.input_layernorm(layernorm_input)
         residual = layernorm_output if self.postnorm else layernorm_input
-        self_attention_output = self.self_attn(
+        self_attention_output = torch.empty_like(layernorm_output)
+        self.self_attn(
             hidden_states=layernorm_output,
+            output=self_attention_output,
             positions=positions,
             kv_caches=kv_caches,
-            attn_metadata=attn_metadata,
         )
 
         residual = residual * self.layernorm_attention_alpha
@@ -817,8 +835,8 @@ def forward(self,
         if self.expert_num == 1:
             hidden_states = self.mlp(layernorm_output)
         else:
-            moe_hidden_states = self.block_sparse_moe(
-                copy.deepcopy(layernorm_output))
+            moe_layernorm_output = layernorm_output.clone()
+            moe_hidden_states = self.block_sparse_moe(moe_layernorm_output)
             if self.shared_moe:
                 before_moe_dtype = layernorm_output.dtype
                 moe_hidden_fp32 = moe_hidden_states.to(torch.float32)
@@ -856,17 +874,15 @@ def shared_moe_coefficient_loader(param: torch.Tensor,
         return
 
 
+@support_torch_compile
 class MiniMaxText01Model(nn.Module):
 
-    def __init__(
-        self,
-        config: MiniMaxConfig,
-        quant_config: Optional[QuantizationConfig] = None,
-        cache_config: Optional[CacheConfig] = None,
-        scheduler_config=None,
-        prefix: str = "",
-    ) -> None:
+    def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         super().__init__()
+        config: MiniMaxConfig = vllm_config.model_config.hf_config
+        cache_config = vllm_config.cache_config
+        quant_config = vllm_config.quant_config
+        scheduler_config = vllm_config.scheduler_config
 
         self.padding_idx = config.pad_token_id
         self.vocab_size = config.vocab_size
@@ -1019,12 +1035,11 @@ def forward(self,
         attn_metadata = forward_context.attn_metadata
         if not envs.VLLM_USE_V1 and attn_metadata is None:
             return None
-        if "request_ids_to_seq_ids" not in kwargs:
-            kwargs["request_ids_to_seq_ids"] = {}
-        if "finished_requests_ids" not in kwargs:
-            kwargs["finished_requests_ids"] = []
-
         if not envs.VLLM_USE_V1:
+            if "request_ids_to_seq_ids" not in kwargs:
+                kwargs["request_ids_to_seq_ids"] = {}
+            if "finished_requests_ids" not in kwargs:
+                kwargs["finished_requests_ids"] = []
             (
                 minimax_cache_tensors,
                 state_indices_tensor,
@@ -1096,7 +1111,6 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
 
         super().__init__()
         config = vllm_config.model_config.hf_config
-        quant_config = vllm_config.quant_config
         lora_config = vllm_config.lora_config
         self.config = config
         self.lora_config = lora_config
@@ -1109,12 +1123,8 @@ def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
         self.unpadded_vocab_size = self.config.vocab_size
         if hasattr(vllm_config.model_config, "max_model_len"):
             self.config.max_model_len = vllm_config.model_config.max_model_len
-        self.model = MiniMaxText01Model(
-            self.config,
-            quant_config,
-            cache_config=vllm_config.cache_config,
-            scheduler_config=vllm_config.scheduler_config,
-            prefix=maybe_prefix(prefix, "model"))
+        self.model = MiniMaxText01Model(vllm_config=vllm_config,
+                                        prefix=maybe_prefix(prefix, "model"))
         if get_pp_group().is_last_rank:
             self.lm_head = ParallelLMHead(
                 self.unpadded_vocab_size,
@@ -1433,3 +1443,36 @@ def get_mamba_state_shape_from_config(
             tp_size=parallel_config.tensor_parallel_size,
             head_dim=hf_config.head_dim,
         )
+
+
+def linear_attention(
+    hidden_states: torch.Tensor,
+    output: torch.Tensor,
+    positions: torch.Tensor,
+    layer_name: str,
+) -> None:
+    forward_context: ForwardContext = get_forward_context()
+    print("layer_name: ", layer_name)
+    self = forward_context.no_compile_layers[layer_name]
+    self._forward(hidden_states=hidden_states,
+                  output=output,
+                  positions=positions,
+                  kv_caches=None)
+
+
+def linear_attention_fake(
+    hidden_states: torch.Tensor,
+    output: torch.Tensor,
+    positions: torch.Tensor,
+    layer_name: str,
+) -> None:
+    return
+
+
+direct_register_custom_op(
+    op_name="linear_attention",
+    op_func=linear_attention,
+    mutates_args=["output"],
+    fake_impl=linear_attention_fake,
+    dispatch_key=current_platform.dispatch_key,
+)