Merge pull request #2121 from huggingface/cleanup_vit_convert

Improve vit conversions. OpenAI convert pass through main convert

timm/models/vision_transformer.py

@@ -771,28 +771,20 @@ def resize_pos_embed(
         antialias: bool = False,
 ) -> torch.Tensor:
     """ Rescale the grid of position embeddings when loading from state_dict.
-
+    *DEPRECATED* This function is being deprecated in favour of using resample_abs_pos_embed
-    *DEPRECATED* This function is being deprecated in favour of resample_abs_pos_embed
-
-    Adapted from:
-        https://github.com/google-research/vision_transformer/blob/00883dd691c63a6830751563748663526e811cee/vit_jax/checkpoint.py#L224
     """
-    ntok_new = posemb_new.shape[1]
+    ntok_new = posemb_new.shape[1] - num_prefix_tokens
-    if num_prefix_tokens:
+    ntok_old = posemb.shape[1] - num_prefix_tokens
-        posemb_prefix, posemb_grid = posemb[:, :num_prefix_tokens], posemb[0, num_prefix_tokens:]
+    gs_old = [int(math.sqrt(ntok_old))] * 2
-        ntok_new -= num_prefix_tokens
-    else:
-        posemb_prefix, posemb_grid = posemb[:, :0], posemb[0]
-    gs_old = int(math.sqrt(len(posemb_grid)))
     if not len(gs_new):  # backwards compatibility
         gs_new = [int(math.sqrt(ntok_new))] * 2
-    assert len(gs_new) >= 2
+    return resample_abs_pos_embed(
-    _logger.info(f'Resized position embedding: {posemb.shape} ({[gs_old, gs_old]}) to {posemb_new.shape} ({gs_new}).')
+        posemb, gs_new, gs_old,
-    posemb_grid = posemb_grid.reshape(1, gs_old, gs_old, -1).permute(0, 3, 1, 2)
+        num_prefix_tokens=num_prefix_tokens,
-    posemb_grid = F.interpolate(posemb_grid, size=gs_new, mode=interpolation, antialias=antialias, align_corners=False)
+        interpolation=interpolation,
-    posemb_grid = posemb_grid.permute(0, 2, 3, 1).reshape(1, gs_new[0] * gs_new[1], -1)
+        antialias=antialias,
-    posemb = torch.cat([posemb_prefix, posemb_grid], dim=1)
+        verbose=True,
-    return posemb
+    )
 
 
 @torch.no_grad()
@@ -962,16 +954,6 @@ def _convert_openai_clip(
             v = v.unsqueeze(0).unsqueeze(1)
         elif k == 'pos_embed':
             v = v.unsqueeze(0)
-            if v.shape[1] != model.pos_embed.shape[1]:
-                # To resize pos embedding when using model at different size from pretrained weights
-                num_prefix_tokens = 0 if getattr(model, 'no_embed_class', False) \
-                    else getattr(model, 'num_prefix_tokens', 1)
-                v = resample_abs_pos_embed(
-                    v,
-                    new_size=model.patch_embed.grid_size,
-                    num_prefix_tokens=num_prefix_tokens,
-                    verbose=True,
-                )
         out_dict[k] = v
     return out_dict
 
@@ -1014,19 +996,17 @@ def checkpoint_filter_fn(
     prefix = ''
 
     if 'visual.class_embedding' in state_dict:
-        return _convert_openai_clip(state_dict, model)
+        state_dict = _convert_openai_clip(state_dict, model)
     elif 'module.visual.class_embedding' in state_dict:
-        return _convert_openai_clip(state_dict, model, prefix='module.visual.')
+        state_dict = _convert_openai_clip(state_dict, model, prefix='module.visual.')
-
+    elif "mask_token" in state_dict:
-    if "mask_token" in state_dict:
         state_dict = _convert_dinov2(state_dict, model)
-
+    elif "encoder" in state_dict:
-    if "encoder" in state_dict:
+        # IJEPA, vit in an 'encoder' submodule
         state_dict = state_dict['encoder']
         prefix = 'module.'
-
+    elif 'visual.trunk.pos_embed' in state_dict:
-    if 'visual.trunk.pos_embed' in state_dict:
+        # OpenCLIP model with timm vision encoder
-        # convert an OpenCLIP model with timm vision encoder
         # FIXME remap final nn.Linear if it exists outside of the timm .trunk (ie in visual.head.proj)
         prefix = 'visual.trunk.'