[WIP] Add Qwen-VL

python/sglang/srt/models/qwenvl.py

@@ -0,0 +1,261 @@
+"""Inference-only Qwen-VL model compatible with HuggingFace weights."""
+
+from typing import List, Optional
+
+import numpy as np
+import torch
+from sglang.srt.managers.router.infer_batch import ForwardMode
+from sglang.srt.managers.router.model_runner import InputMetadata
+from sglang.srt.mm_utils import (
+    get_anyres_image_grid_shape,
+    unpad_image,
+    unpad_image_shape,
+)
+from sglang.srt.models.llava import (
+    clip_vision_embed_forward,
+    monkey_path_clip_vision_embed_forward,
+)
+from sglang.srt.models.qwen import QWenLMHeadModel
+from torch import nn
+from vllm.transformers_utils.configs.qwen import QWenConfig
+from transformers import CLIPVisionModel
+from vllm.model_executor.layers.linear import LinearMethodBase
+from vllm.model_executor.weight_utils import (
+    default_weight_loader,
+    hf_model_weights_iterator,
+)
+from vllm.transformers_utils.configs.qwen import QWenConfig
+
+def get_abs_pos(abs_pos, tgt_size):
+    # abs_pos: L, C
+    # tgt_size: M
+    # return: M, C
+    src_size = int(math.sqrt(abs_pos.size(0)))
+    tgt_size = int(math.sqrt(tgt_size))
+    dtype = abs_pos.dtype
+
+    if src_size != tgt_size:
+        return F.interpolate(
+            abs_pos.float().reshape(1, src_size, src_size, -1).permute(0, 3, 1, 2),
+            size=(tgt_size, tgt_size),
+            mode="bicubic",
+            align_corners=False,
+        ).permute(0, 2, 3, 1).flatten(0, 2).to(dtype=dtype)
+    else:
+        return abs_pos
+
+# https://github.com/facebookresearch/mae/blob/efb2a8062c206524e35e47d04501ed4f544c0ae8/util/pos_embed.py#L20
+def get_2d_sincos_pos_embed(embed_dim, grid_size, cls_token=False):
+    """
+    grid_size: int of the grid height and width
+    return:
+    pos_embed: [grid_size*grid_size, embed_dim] or [1+grid_size*grid_size, embed_dim] (w/ or w/o cls_token)
+    """
+    grid_h = np.arange(grid_size, dtype=np.float32)
+    grid_w = np.arange(grid_size, dtype=np.float32)
+    grid = np.meshgrid(grid_w, grid_h)  # here w goes first
+    grid = np.stack(grid, axis=0)
+
+    grid = grid.reshape([2, 1, grid_size, grid_size])
+    pos_embed = get_2d_sincos_pos_embed_from_grid(embed_dim, grid)
+    if cls_token:
+        pos_embed = np.concatenate([np.zeros([1, embed_dim]), pos_embed], axis=0)
+    return pos_embed
+
+def get_2d_sincos_pos_embed_from_grid(embed_dim, grid):
+    assert embed_dim % 2 == 0
+
+    # use half of dimensions to encode grid_h
+    emb_h = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[0])  # (H*W, D/2)
+    emb_w = get_1d_sincos_pos_embed_from_grid(embed_dim // 2, grid[1])  # (H*W, D/2)
+
+    emb = np.concatenate([emb_h, emb_w], axis=1) # (H*W, D)
+    return emb
+
+
+def get_1d_sincos_pos_embed_from_grid(embed_dim, pos):
+    """
+    embed_dim: output dimension for each position
+    pos: a list of positions to be encoded: size (M,)
+    out: (M, D)
+    """
+    assert embed_dim % 2 == 0
+    omega = np.arange(embed_dim // 2, dtype=np.float32)
+    omega /= embed_dim / 2.
+    omega = 1. / 10000**omega  # (D/2,)
+
+    pos = pos.reshape(-1)  # (M,)
+    out = np.einsum('m,d->md', pos, omega)  # (M, D/2), outer product
+
+    emb_sin = np.sin(out) # (M, D/2)
+    emb_cos = np.cos(out) # (M, D/2)
+
+    emb = np.concatenate([emb_sin, emb_cos], axis=1)  # (M, D)
+    return emb
+
+class Resampler(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.grid_size = 16
+        self.num_queries = self.grid_size * self.grid_size # Default is 256 since grid_size ** 2
+        self.embed_dim = config.visual.output_dim
+        self.num_heads = config.visual.num_heads
+
+        self.pos_embed = nn.Parameter(
+            torch.from_numpy(get_2d_sincos_pos_embed(self.embed_dim, self.grid_size)).float()
+        ).requires_grad_(False)
+
+        self.query = nn.Parameter(torch.zeros(self.num_queries, self.embed_dim))
+        self.kv_proj = nn.Linear(kv_dim, self.embed_dim, bias=False)
+        self.attn = nn.MultiHeadAttention(self.embed_dim, self.num_heads)
+        self.ln_q = nn.LayerNorm(self.embed_dim)
+        self.ln_kv = nn.LayerNorm(self.embed_dim)
+
+    def forward(self, x, attn_mask=None):
+        pos_embed = get_abs_pos(self.pos_embed, x.size(1))
+
+        x = self.kv_proj(x)
+        x = self.ln_kv(x).permute(1, 0, 2)
+
+        N = x.shape[1]
+        q = self.ln_q(self.query)
+        repeated_query = q.unsqueeze(1).repeat(1, N, 1)
+        out = self.attn(
+            repeated_query + self.pos_embed.unsqueeze(1),
+            x + pos_embed.unsqueeze(1),
+            x,
+            attn_mask=attn_mask)[0]
+        return out.permute(1, 0, 2)
+
+class QwenVLForCausalLM(nn.Module):
+    def __init__(
+        self,
+        config: QwenConfig,
+        linear_method: Optional[LinearMethodBase] = None,
+    ) -> None:
+        super().__init__()
+        self.config = config
+        self.vision_tower = None
+        self.config.vision_config.hidden_size = config.visual.hidden_size
+        self.config.text_config.hidden_size = config.hidden_size
+        self.resampler = QwenResampler(self.config)
+        self.language_model = QWenLMHeadModel(config, linear_method)
+
+    def pad_input_ids(self, input_ids, pad_value, pt_shape=None, image_size=None):
+        new_image_feature_len = self.image_feature_len
+        # now only support spatial_unpad + anyres
+        if self.mm_patch_merge_type.startswith("spatial"):
+            height = width = self.num_patches_per_side
+            if pt_shape[0] > 1:
+                if self.image_aspect_ratio == "anyres":
+                    num_patch_width, num_patch_height = get_anyres_image_grid_shape(
+                        image_size,
+                        self.image_grid_pinpoints,
+                        self.vision_tower.config.image_size,
+                    )
+                if "unpad" in self.mm_patch_merge_type:
+                    h = num_patch_height * height
+                    w = num_patch_width * width
+                    new_h, new_w = unpad_image_shape(h, w, image_size)
+                    new_image_feature_len += new_h * (new_w + 1)
+
+        pad_ids = pad_value * (
+            (new_image_feature_len + len(pad_value)) // len(pad_value)
+        )
+        offset = input_ids.index(self.config.image_token_index)
+        # old_len + pad_len - 1, because we need to remove image_token_id
+        new_input_ids = (
+            input_ids[:offset]
+            + pad_ids[:new_image_feature_len]
+            + input_ids[offset + 1 :]
+        )
+        return new_input_ids, offset
+
+    def encode_images(self, pixel_values: torch.Tensor) -> torch.Tensor:
+        image_outputs = self.vision_tower(pixel_values, output_hidden_states=True)
+        # NOTE: This is not memory efficient. (output_hidden_states=True) will save all the hidden stated.
+
+        selected_image_feature = image_outputs.hidden_states[self.vision_feature_layer]
+        if self.vision_feature_select_strategy in ["default", "patch"]:
+            selected_image_feature = selected_image_feature[:, 1:]
+        elif self.vision_feature_select_strategy == "full":
+            selected_image_feature = selected_image_feature
+        else:
+            raise ValueError(
+                f"Unexpected select feature strategy: {self.config.vision_feature_select_strategy}"
+            )
+        image_features = self.multi_modal_projector(selected_image_feature)
+
+        return image_features
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor,
+        positions: torch.Tensor,
+        input_metadata: InputMetadata,
+        pixel_values: Optional[List[Optional[np.array]]] = None,
+        image_sizes: Optional[List[List[int]]] = None,
+        image_offsets: Optional[List[int]] = None,
+    ) -> torch.Tensor:
+        pass
+
+    def load_weights(
+        self,
+        model_name_or_path: str,
+        cache_dir: Optional[str] = None,
+        load_format: str = "auto",
+        revision: Optional[str] = None,
+    ):
+        # load clip vision model by cfg['mm_vision_tower']:
+        #   huggingface_name or path_of_clip_relative_to_llava_model_dir
+        vision_path = self.config.mm_vision_tower
+        self.vision_tower = CLIPVisionModel.from_pretrained(
+            vision_path, torch_dtype=torch.float16
+        ).cuda()
+        self.vision_tower.eval()
+
+        self.vision_feature_layer = self.config.mm_vision_select_layer
+        self.vision_feature_select_strategy = self.config.mm_vision_select_feature
+        self.image_size = self.vision_tower.config.image_size
+        self.patch_size = self.vision_tower.config.patch_size
+
+        # NOTE(chris): Not relevant?
+        # self.mm_patch_merge_type = getattr(self.config, "mm_patch_merge_type", "flat")
+        # self.image_aspect_ratio = getattr(self.config, "image_aspect_ratio", "square")
+        # self.image_grid_pinpoints = getattr(self.config, "image_grid_pinpoints", None)
+
+        self.image_feature_len = int((self.image_size / self.patch_size) ** 2)
+        if self.vision_feature_select_strategy == "patch":
+            pass
+        elif self.vision_feature_select_strategy == "cls_patch":
+            self.image_feature_len += 1
+        else:
+            raise ValueError(f"Unexpected select feature: {self.select_feature}")
+
+        # load mm_projector
+        projector_weights = {
+            "model.mm_projector.0": "multi_modal_projector.linear_1",
+            "model.mm_projector.2": "multi_modal_projector.linear_2",
+            "model.vision_tower.vision_tower": "vision_tower",  # Update the vision tower weights if we find them in the checkpoint (it may be finetuned).
+        }
+        params_dict = dict(self.named_parameters())
+        for name, loaded_weight in hf_model_weights_iterator(
+            model_name_or_path, cache_dir, load_format, revision
+        ):
+            # FIXME: why projector weights read two times?
+            if "projector" in name or "vision_tower" in name:
+                for weight_name, param_name in projector_weights.items():
+                    if weight_name in name:
+                        name = name.replace(weight_name, param_name)
+                param = params_dict[name]
+                weight_loader = getattr(param, "weight_loader", default_weight_loader)
+                weight_loader(param, loaded_weight)
+
+        # load language model
+        self.language_model.load_weights(
+            model_name_or_path, cache_dir, load_format, revision
+        )
+
+        monkey_path_clip_vision_embed_forward()
+
+EntryClass = QwenVLForCausalLM