[Core] Refactor `QKVCrossParallelLinear` implementation to support BNB 4-bit quantization (#14545)

Signed-off-by: Isotr0py <2037008807@qq.com>

tests/models/encoder_decoder/vision_language/test_mllama.py

@@ -17,6 +17,7 @@ from vllm.sequence import SampleLogprobs
 
 from ....conftest import (IMAGE_ASSETS, HfRunner, PromptImageInput, VllmRunner,
                           _ImageAssets)
+from ....quantization.utils import is_quant_method_supported
 from ....utils import large_gpu_test
 from ...utils import check_logprobs_close
 
@@ -397,6 +398,50 @@ def test_models_interleaved_images(hf_runner, vllm_runner, image_assets, model,
         )
 
 
+@large_gpu_test(min_gb=48)
+@pytest.mark.core_model
+@pytest.mark.parametrize("model", models)
+@pytest.mark.parametrize("dtype", ["float16"])
+@pytest.mark.parametrize("max_tokens", [32])
+@pytest.mark.skipif(not is_quant_method_supported("bitsandbytes"),
+                    reason='bitsandbytes is not supported on this GPU type.')
+def test_bnb_regression(
+    image_assets: _ImageAssets,
+    model: str,
+    dtype: str,
+    max_tokens: int,
+):
+    stop_sign = image_assets[0].pil_image
+    prompts = [
+        {
+            "prompt": "<|begin_of_text|>The content of the image <|image|> is",
+            "multi_modal_data": {
+                "image": stop_sign
+            },
+        },
+        {
+            "prompt":
+            "The color of the sky is blue but sometimes it can also be",
+        },
+    ]
+    # Test regression about QKVCrossParallelLinear
+    llm = LLM(
+        model=model,
+        dtype=dtype,
+        max_model_len=4096,
+        max_num_seqs=2,
+        enforce_eager=True,
+        quantization="bitsandbytes",
+        load_format="bitsandbytes",
+    )
+    sampling_params = SamplingParams(
+        temperature=0,
+        max_tokens=max_tokens,
+    )
+    outputs = llm.generate(prompts, sampling_params)
+    assert outputs
+
+
 @large_gpu_test(min_gb=48)
 @pytest.mark.core_model
 @pytest.mark.parametrize("model", models)

vllm/model_executor/layers/linear.py

@@ -2,9 +2,10 @@
 
 import itertools
 from abc import abstractmethod
-from typing import Optional, Union
+from typing import Any, Literal, Optional, Union
 
 import torch
+import torch.nn as nn
 import torch.nn.functional as F
 from torch.nn.parameter import Parameter, UninitializedParameter
 
@@ -84,6 +85,43 @@ def adjust_scalar_to_fused_array(param, loaded_weight, shard_id):
     return param[shard_id], loaded_weight
 
 
+# TODO(Isotr0py): We might need a more flexible structure to handle
+# bitsandbytes shard offsets.
+def left_shift_bitsandbytes_4bit_shard(bnb_weight_attrs: dict[str, Any]):
+    """
+    Separate the BitsAndBytes 4-bit shard.
+
+    For example, given bnb weight attributes as below:
+    {
+        'bnb_shard_offsets': array([0, 4, 8, 16]), 
+        'bnb_quant_state': {0: ..., 1: ..., 2: ...},
+    }
+
+    The function will return:
+    {
+        'bnb_shard_offsets': array([0, 4]), 
+        'bnb_quant_state': {0: ...},
+    }
+    and
+    {
+        'bnb_shard_offsets': array([0, 4, 12]),
+        'bnb_quant_state': {0: ..., 1: ...},
+    }
+    """
+    shard_offsets = bnb_weight_attrs["bnb_shard_offsets"]
+    offset_l = shard_offsets[:2]
+    offset_r = shard_offsets[1:] - shard_offsets[1]
+    quant_state_l = {0: bnb_weight_attrs["bnb_quant_state"][0]}
+    quant_state_r = {
+        i - 1: bnb_weight_attrs["bnb_quant_state"][i]
+        for i in range(1,
+                       len(shard_offsets) - 1)
+    }
+    left = dict(bnb_shard_offsets=offset_l, bnb_quant_state=quant_state_l)
+    right = dict(bnb_shard_offsets=offset_r, bnb_quant_state=quant_state_r)
+    return left, right
+
+
 class LinearMethodBase(QuantizeMethodBase):
     """Base class for different (maybe quantized) linear methods."""
 
@@ -1229,7 +1267,24 @@ class RowParallelLinear(LinearBase):
         return s
 
 
-class QKVCrossParallelLinear(torch.nn.Module):
+class QKVCrossParallelLinear(LinearBase):
+    """Linear layers for efficient cross-attention's QKV transformation.
+
+    Args:
+        hidden_size: input hidden state size of the transformer.
+        head_size: size of each attention head.
+        total_num_heads: total number of attention query heads.
+        total_num_kv_heads: total number of attention key/value heads. If
+                            None, assume total_num_kv_heads = total_num_heads.
+        bias: If true, add bias.
+        skip_bias_add: This was added to enable performance optimizations where
+                       bias can be fused with other element-wise operations. we
+                       skip adding bias but instead return it.
+        params_dtype: Data type for the parameters.
+        quant_config: Quantization configure.
+        prefix: The name of the layer in the state dict, including all parents
+                        (e.g. model.layers.0.qkv_proj)
+    """
 
     def __init__(self,
                  hidden_size: int,
@@ -1241,12 +1296,28 @@ class QKVCrossParallelLinear(torch.nn.Module):
                  params_dtype: Optional[torch.dtype] = None,
                  quant_config: Optional[QuantizationConfig] = None,
                  prefix: str = ""):
-        super().__init__()
+        # input_size and output_size are not used, just for alignment
+        input_size = hidden_size
+        output_size = (total_num_heads + (total_num_kv_heads or 0)) * head_size
+        super().__init__(input_size=input_size,
+                         output_size=output_size,
+                         skip_bias_add=skip_bias_add,
+                         params_dtype=params_dtype,
+                         quant_config=quant_config,
+                         prefix=prefix)
+
+        self.quant_config = quant_config
+
         # Empty placeholders for loading as a single module.
-        self.weight = torch.nn.Parameter()
-        set_weight_attrs(self.weight, {
-            "weight_loader": self.weight_loader_weight,
-        })
+        placeholder_size = 0
+        assert self.quant_method is not None
+        self.quant_method.create_weights(self,
+                                         placeholder_size, [placeholder_size],
+                                         placeholder_size,
+                                         placeholder_size,
+                                         self.params_dtype,
+                                         weight_loader=self.weight_loader)
+
         # Use a dictionary to avoid submodules parameters auto-registration:
         # drop-in replacement for a `QKVParallelLinear` module.
         self.proj = dict()
@@ -1276,18 +1347,94 @@ class QKVCrossParallelLinear(torch.nn.Module):
         if bias:
             self.bias = torch.nn.Parameter()
             set_weight_attrs(self.bias, {
-                "weight_loader": self.weight_loader_bias,
+                "output_dim": 0,
+                "weight_loader": self.weight_loader,
             })
+        else:
+            self.bias = None
 
     @property
-    def q_proj_decoder(self):
-        return self.proj["q_proj_decoder"]
+    def q_proj_decoder(self) -> ColumnParallelLinear:
+        layer = self.proj["q_proj_decoder"]
+        for name, param in self.named_parameters():
+            target_param = getattr(layer, name)
+            self.sync_weight_attrs(param, target_param, mode="q_proj_decoder")
+        return layer
 
     @property
-    def kv_proj_encoder(self):
-        return self.proj["kv_proj_encoder"]
+    def kv_proj_encoder(self) -> QKVParallelLinear:
+        layer = self.proj["kv_proj_encoder"]
+        for name, param in self.named_parameters():
+            target_param = getattr(layer, name)
+            self.sync_weight_attrs(param, target_param, mode="kv_proj_encoder")
+        return layer
+
+    def sync_weight_attrs(
+        self,
+        src_param: nn.Parameter,
+        tgt_param: nn.Parameter,
+        mode: Literal["q_proj_decoder", "kv_proj_encoder"],
+    ):
+        missing_attrs_dict = {
+            k: getattr(src_param, k)
+            for k in (set(src_param.__dict__.keys()) -
+                      set(tgt_param.__dict__.keys()))
+        }
+        # TODO(Isotr0py): handle bitsandbytes 8bit
+        use_bitsandbytes_4bit = getattr(src_param, "use_bitsandbytes_4bit",
+                                        False)
+        if (missing_attrs_dict and use_bitsandbytes_4bit):
+            q_proj_attrs, kv_proj_attrs = left_shift_bitsandbytes_4bit_shard(
+                missing_attrs_dict)
+            if mode == "q_proj_decoder":
+                set_weight_attrs(tgt_param, q_proj_attrs)
+            elif mode == "kv_proj_encoder":
+                set_weight_attrs(tgt_param, kv_proj_attrs)
+        else:
+            set_weight_attrs(tgt_param, missing_attrs_dict)
 
-    def forward(self, decoder_hidden_states, encoder_hidden_states):
+    def _is_same_param(
+        self,
+        src_param: torch.nn.Parameter,
+        map_param: torch.nn.Parameter,
+    ) -> bool:
+        """Check if two parameters are exactly pointing to same things."""
+        # ignore weight_loader because it's always different
+        key_to_ignore = ["weight_loader", "_weight_loader"]
+        has_same_type_name = type(src_param) is type(map_param)
+        src_param_attrs = {
+            k: v
+            for k, v in src_param.__dict__.items() if k not in key_to_ignore
+        }
+        map_param_attrs = {
+            k: v
+            for k, v in map_param.__dict__.items() if k not in key_to_ignore
+        }
+        has_same_attrs = src_param_attrs == map_param_attrs
+        return has_same_type_name and has_same_attrs
+
+    def select_proj_params(
+        self,
+        layer: nn.Module,
+        param: nn.Parameter,
+    ) -> nn.Parameter:
+        """
+        Given the placeholder param, 
+        return the corresponding param in the proj layers.
+        """
+        target_param_list = [
+            v for _, v in layer.named_parameters()
+            if self._is_same_param(param, v)
+        ]
+        assert len(target_param_list) == 1
+        target_param = target_param_list[0]
+        return target_param
+
+    def forward(  # type: ignore[override]
+        self,
+        decoder_hidden_states: torch.Tensor,
+        encoder_hidden_states: torch.Tensor,
+    ) -> tuple[torch.Tensor, ...]:
         q, _ = self.q_proj_decoder(decoder_hidden_states)
         if encoder_hidden_states is None:
             # Encoder KV already cached.
@@ -1300,25 +1447,21 @@ class QKVCrossParallelLinear(torch.nn.Module):
             k, v = kv_enc.split(self.kv_size, dim=-1)
         return q, k, v
 
-    def weight_loader_weight(self,
-                             param: torch.nn.Parameter,
-                             loaded_weight: torch.Tensor,
-                             loaded_shard_id: Optional[str] = None):
-        # NOTE Use QKV/ColumnParallel weight_loader, ignore placeholder param.
-        param = self.q_proj_decoder.weight if loaded_shard_id == "q" \
-            else self.kv_proj_encoder.weight
-        param.weight_loader(
-            param,
-            loaded_weight) if loaded_shard_id == "q" else param.weight_loader(
-                param, loaded_weight, loaded_shard_id)
-
-    def weight_loader_bias(self,
-                           param: torch.nn.Parameter,
-                           loaded_weight: torch.Tensor,
-                           loaded_shard_id: Optional[str] = None):
-        param = self.q_proj_decoder.bias if loaded_shard_id == "q" \
-            else self.kv_proj_encoder.bias
-        param.weight_loader(
-            param,
-            loaded_weight) if loaded_shard_id == "q" else param.weight_loader(
-                param, loaded_weight, loaded_shard_id)
+    def weight_loader(self,
+                      param: torch.nn.Parameter,
+                      loaded_weight: torch.Tensor,
+                      loaded_shard_id: Optional[str] = None):
+        layer = (self.q_proj_decoder
+                 if loaded_shard_id == "q" else self.kv_proj_encoder)
+        target_param = self.select_proj_params(layer, param)
+        shard_id_args = (loaded_shard_id, ) if loaded_shard_id != "q" else ()
+        layer.weight_loader(target_param, loaded_weight, *shard_id_args)
+
+    def extra_repr(self) -> str:
+        s = f"in_features={self.input_size}"
+        s += f", q_size={self.q_proj_decoder.output_size_per_partition}"
+        s += f", kv_size={self.kv_size}"
+        s += f", bias={self.bias is not None}"
+        s += f", tp_size={get_tensor_model_parallel_world_size()}"
+        s += ", gather_output=False"
+        return s
\ No newline at end of file

vllm/model_executor/models/mllama.py

@@ -43,6 +43,7 @@ from vllm.forward_context import get_forward_context
 from vllm.logger import init_logger
 from vllm.model_executor.layers.layernorm import RMSNorm
 from vllm.model_executor.layers.linear import (ColumnParallelLinear,
+                                               QKVCrossParallelLinear,
                                                QKVParallelLinear,
                                                RowParallelLinear)
 from vllm.model_executor.layers.logits_processor import LogitsProcessor
@@ -813,20 +814,11 @@ class MllamaTextCrossAttention(nn.Module):
         self.q_local_size = self.num_local_heads * self.head_dim
         self.kv_local_size = self.num_local_key_value_heads * self.head_dim
 
-        # TODO(Isotr0py): Use QKVCrossParallelLinear when it supports
-        # quantization
-        self.q_proj = ColumnParallelLinear(
-            input_size=self.hidden_size,
-            output_size=self.num_heads * self.head_dim,
-            bias=False,
-            quant_config=quant_config,
-            prefix=f"{prefix}.q_proj",
-        )
-        self.kv_proj = QKVParallelLinear(
+        self.qkv_proj = QKVCrossParallelLinear(
             self.hidden_size,
             self.head_dim,
-            total_num_heads=0,
-            total_num_kv_heads=self.num_key_value_heads,
+            self.num_heads,
+            self.num_key_value_heads,
             bias=False,
             quant_config=quant_config,
             prefix=f"{prefix}.qkv_proj",
@@ -862,15 +854,11 @@ class MllamaTextCrossAttention(nn.Module):
         kv_range_for_decode: Optional[List[Tuple[int, int]]],
         cross_attention_states: Optional[torch.Tensor],
     ) -> torch.Tensor:
-        q, _ = self.q_proj(hidden_states)
+        q, k, v = self.qkv_proj(hidden_states, cross_attention_states)
         if cross_attention_states is not None:
-            kv, _ = self.kv_proj(cross_attention_states)
-            k, v = kv.split([self.kv_local_size, self.kv_local_size], dim=-1)
             k = k.view(-1, self.num_local_key_value_heads, self.head_dim)
             v = v.view(-1, self.num_local_key_value_heads, self.head_dim)
             k = self.k_norm(k)
-        else:
-            k = v = None
 
         q = q.view(-1, self.num_local_heads, self.head_dim)
         q = self.q_norm(q)
@@ -1161,13 +1149,8 @@ class MllamaForCausalLM(nn.Module):
 class MllamaForConditionalGeneration(nn.Module, SupportsMultiModal,
                                      SupportsV0Only):
     packed_modules_mapping = {
-        "self_attn.qkv_proj": [
-            "self_attn.q_proj",
-            "self_attn.k_proj",
-            "self_attn.v_proj",
-        ],
-        "cross_attn.kv_proj": ["cross_attn.k_proj", "cross_attn.v_proj"],
-        "gate_up_proj": ["gate_proj", "up_proj"],
+        "qkv_proj": ["q_proj", "k_proj", "v_proj"],
+        "gate_up_proj": ["gate_proj", "up_proj"]
     }
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
@@ -1437,11 +1420,9 @@ class MllamaForConditionalGeneration(nn.Module, SupportsMultiModal,
                                                    torch.Tensor]]) -> Set[str]:
         stacked_params_mapping = [
             # (param_name, shard_name, shard_id)
-            (".self_attn.qkv_proj", ".self_attn.q_proj", "q"),
-            (".self_attn.qkv_proj", ".self_attn.k_proj", "k"),
-            (".self_attn.qkv_proj", ".self_attn.v_proj", "v"),
-            (".cross_attn.kv_proj", ".cross_attn.k_proj", "k"),
-            (".cross_attn.kv_proj", ".cross_attn.v_proj", "v"),
+            (".qkv_proj", ".q_proj", "q"),
+            (".qkv_proj", ".k_proj", "k"),
+            (".qkv_proj", ".v_proj", "v"),
             (".gate_up_proj", ".gate_proj", 0),
             (".gate_up_proj", ".up_proj", 1),
         ]
@@ -1570,4 +1551,4 @@ def convert_dense_cross_attention_mask_to_tensor(
     full_text_mask = ((mask != ninf).any(dim=-1).type_as(mask)[..., None])
     mask *= full_text_mask
     # (num_prompt_tokens, num_encoder_tokens)
-    return mask
+    return mask
\ No newline at end of file