[Model] Support quantization of PixtralHFTransformer for PixtralHF (#9921)

Signed-off-by: mgoin <michael@neuralmagic.com>

vllm/model_executor/layers/activation.py

@@ -299,3 +299,33 @@ def get_act_fn(
         return ScaledActivation(act_fn, intermediate_size, input_is_parallel,
                                 params_dtype)
     return act_fn
+
+
+_ACTIVATION_AND_MUL_REGISTRY = LazyDict({
+    "gelu": lambda: GeluAndMul(),
+    "silu": lambda: SiluAndMul(),
+})
+
+
+def get_act_and_mul_fn(
+    act_fn_name: str,
+    quant_config: Optional[QuantizationConfig] = None,
+    intermediate_size: Optional[int] = None,
+    input_is_parallel: bool = True,
+    params_dtype: Optional[torch.dtype] = None,
+) -> nn.Module:
+    """Get an activation-and-mul (i.e. SiluAndMul) function by name."""
+    act_fn_name = act_fn_name.lower()
+    if act_fn_name not in _ACTIVATION_AND_MUL_REGISTRY:
+        raise ValueError(
+            f"Activation function {act_fn_name!r} is not supported.")
+
+    act_fn = _ACTIVATION_AND_MUL_REGISTRY[act_fn_name]
+    if (quant_config is not None
+            and act_fn_name in quant_config.get_scaled_act_names()):
+        if intermediate_size is None:
+            raise ValueError("intermediate_size must be specified for scaled "
+                             "activation functions.")
+        return ScaledActivation(act_fn, intermediate_size, input_is_parallel,
+                                params_dtype)
+    return act_fn

vllm/model_executor/models/pixtral.py

@@ -19,8 +19,11 @@ from vllm.attention import AttentionMetadata
 from vllm.config import CacheConfig, ModelConfig, MultiModalConfig
 from vllm.inputs import (INPUT_REGISTRY, DecoderOnlyInputs, DummyData,
                          InputContext, token_inputs)
-from vllm.model_executor.layers.activation import get_act_fn
+from vllm.model_executor.layers.activation import get_act_and_mul_fn
 from vllm.model_executor.layers.layernorm import RMSNorm
+from vllm.model_executor.layers.linear import (MergedColumnParallelLinear,
+                                               QKVParallelLinear,
+                                               RowParallelLinear)
 from vllm.model_executor.layers.quantization import QuantizationConfig
 from vllm.model_executor.layers.sampler import Sampler, SamplerOutput
 from vllm.model_executor.model_loader.weight_utils import default_weight_loader
@@ -798,20 +801,24 @@ class PixtralHFMLP(nn.Module):
         super().__init__()
 
         assert config.intermediate_size is not None
-        # TODO: Use quant_config and prefix after optimizing this
-        self.gate_proj = nn.Linear(config.hidden_size,
-                                   config.intermediate_size,
-                                   bias=False)
-        self.up_proj = nn.Linear(config.hidden_size,
-                                 config.intermediate_size,
-                                 bias=False)
-        self.down_proj = nn.Linear(config.intermediate_size,
-                                   config.hidden_size,
-                                   bias=False)
-        self.act = get_act_fn(config.hidden_act)
+        self.gate_up_proj = MergedColumnParallelLinear(
+            input_size=config.hidden_size,
+            output_sizes=[config.intermediate_size] * 2,
+            bias=False,
+            quant_config=quant_config,
+            prefix=f"{prefix}.gate_up_proj")
+        self.down_proj = RowParallelLinear(input_size=config.intermediate_size,
+                                           output_size=config.hidden_size,
+                                           bias=False,
+                                           quant_config=quant_config,
+                                           prefix=f"{prefix}.down_proj")
+        self.act_and_mul = get_act_and_mul_fn(config.hidden_act)
 
     def forward(self, x: torch.Tensor) -> torch.Tensor:
-        return self.down_proj(self.act(self.gate_proj(x)) * self.up_proj(x))
+        gate_up, _ = self.gate_up_proj(x)
+        x = self.act_and_mul(gate_up)
+        x, _ = self.down_proj(x)
+        return x
 
 
 class PixtralHFAttention(nn.Module):
@@ -830,21 +837,21 @@ class PixtralHFAttention(nn.Module):
         self.n_heads = config.num_attention_heads
         self.head_dim = config.hidden_size // config.num_attention_heads
 
-        self.scale = self.head_dim**-0.5
-
-        # TODO: Use quant_config and prefix after optimizing this
-        self.q_proj = nn.Linear(config.hidden_size,
-                                config.hidden_size,
-                                bias=False)
-        self.k_proj = nn.Linear(config.hidden_size,
-                                config.hidden_size,
-                                bias=False)
-        self.v_proj = nn.Linear(config.hidden_size,
-                                config.hidden_size,
-                                bias=False)
-        self.o_proj = nn.Linear(config.hidden_size,
-                                config.hidden_size,
-                                bias=False)
+        self.qkv_proj = QKVParallelLinear(
+            hidden_size=config.hidden_size,
+            head_size=self.head_dim,
+            total_num_heads=self.n_heads,
+            bias=False,
+            quant_config=quant_config,
+            prefix=f"{prefix}.qkv_proj",
+        )
+        self.o_proj = RowParallelLinear(
+            input_size=config.hidden_size,
+            output_size=config.hidden_size,
+            bias=False,
+            quant_config=quant_config,
+            prefix=f"{prefix}.o_proj",
+        )
 
     def forward(
         self,
@@ -854,13 +861,13 @@ class PixtralHFAttention(nn.Module):
     ) -> Tuple[torch.Tensor, Optional[torch.Tensor]]:
         batch, patches, _ = hidden_states.size()
 
-        q = self.q_proj(hidden_states)
-        k = self.k_proj(hidden_states)
-        v = self.v_proj(hidden_states)
+        qkv_states, _ = self.qkv_proj(hidden_states)
+        q, k, v = qkv_states.chunk(3, dim=-1)
 
         # Transpose q and k to apply HF's Rotary Position Embedding
         q = q.view(batch, patches, self.n_heads, self.head_dim).transpose(1, 2)
         k = k.view(batch, patches, self.n_heads, self.head_dim).transpose(1, 2)
+        v = v.view(batch, patches, self.n_heads, self.head_dim)
         cos, sin = position_embeddings
         q, k = apply_rotary_pos_emb(q, k, cos, sin, unsqueeze_dim=0)
 
@@ -868,22 +875,21 @@ class PixtralHFAttention(nn.Module):
             # Transpose q and k back for attention
             q = q.transpose(1, 2).contiguous()
             k = k.transpose(1, 2).contiguous()
-            v = v.reshape(batch, patches, self.n_heads, self.head_dim)
 
             out = xops.memory_efficient_attention(q,
                                                   k,
                                                   v,
                                                   attn_bias=attention_mask)
         else:
-            v = v.reshape(batch, patches, self.n_heads,
-                          self.head_dim).transpose(1, 2)
+            v = v.transpose(1, 2)
             out = nn.functional.scaled_dot_product_attention(
                 q, k, v, attn_mask=attention_mask)
             out = out.transpose(1, 2)
 
-        out = out.reshape(batch, patches, self.n_heads * self.head_dim)
+        out = out.view(batch, patches, self.n_heads * self.head_dim)
+        attn_output, _ = self.o_proj(out)
 
-        return self.o_proj(out)
+        return attn_output, None
 
 
 class PixtralHFTransformerBlock(nn.Module):
@@ -912,7 +918,7 @@ class PixtralHFTransformerBlock(nn.Module):
         attention_mask: torch.Tensor,
         position_embeddings: torch.Tensor,
     ) -> torch.Tensor:
-        r = self.attention.forward(self.attention_norm(hidden_states),
+        r, _ = self.attention.forward(self.attention_norm(hidden_states),
                                       attention_mask=attention_mask,
                                       position_embeddings=position_embeddings)
         h = hidden_states + r
@@ -1053,10 +1059,24 @@ class PixtralHFVisionModel(nn.Module):
     # (TODO) Add prefix argument for filtering out weights to be loaded
     #        ref: https://github.com/vllm-project/vllm/pull/7186#discussion_r1734163986
     def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
-        stacked_params_mapping = []
+        stacked_params_mapping = [
+            # (param_name, shard_name, shard_id)
+            (".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),
+        ]
         params_dict = dict(self.named_parameters())
+        layer_count = len(self.transformer.layers)
 
         for name, loaded_weight in weights:
+            # omit layers when num_hidden_layers_override is set
+            if name.startswith("transformer.layers"):
+                layer_idx = int(name.split(".")[2])
+                if layer_idx >= layer_count:
+                    continue
+
             for (param_name, weight_name, shard_id) in stacked_params_mapping:
                 if weight_name not in name:
                     continue