Chameleon: not supported with fast load (#32091)

fixes

docs/source/en/_toctree.yml

@@ -761,7 +761,7 @@
       - local: model_doc/bros
         title: BROS
       - local: model_doc/chameleon
-        title: chameleon
+        title: Chameleon
       - local: model_doc/chinese_clip
         title: Chinese-CLIP
       - local: model_doc/clip

docs/source/en/model_doc/chameleon.md

@@ -55,14 +55,14 @@ The original code can be found [here](https://github.com/facebookresearch/chamel
 - Chameleon generates in chat format which means that the generated text will always be the "assistant's turn". You can enable a text completion generation by passing `return_for_text_completion=True` when calling the processor.
 
 > [!NOTE]
-> Chameleon implementation in Transformers uses a special image token to indicate where to merge image embeddings. For special image token we didn't add a new one but used one of the reserved tokens: `<reserved08707>`.
+> Chameleon implementation in Transformers uses a special image token to indicate where to merge image embeddings. For special image token we didn't add a new one but used one of the reserved tokens: `<reserved08707>`. You have to add `<image>` to your prompt in the place where the image should be embedded for correct generation.
 
 ## Usage example
 
 ### Single image inference
 
 Chameleon is a gated model so make sure to have access and login to Hugging Face Hub using a token. 
-Here's how to load the model and perform inference in half-precision (`torch.float16`):
+Here's how to load the model and perform inference in half-precision (`torch.bfloat16`):
 
 ```python
 from transformers import ChameleonProcessor, ChameleonForConditionalGeneration
@@ -71,7 +71,7 @@ from PIL import Image
 import requests
 
 processor = ChameleonProcessor.from_pretrained("facebook/chameleon-7b")
-model = ChameleonForConditionalGeneration.from_pretrained("facebook/chameleon-7b", torch_dtype=torch.float16, device_map="cuda")
+model = ChameleonForConditionalGeneration.from_pretrained("facebook/chameleon-7b", torch_dtype=torch.bfloat16, device_map="cuda")
 
 # prepare image and text prompt
 url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
@@ -97,7 +97,7 @@ import requests
 
 processor = ChameleonProcessor.from_pretrained("facebook/chameleon-7b")
 
-model = ChameleonForConditionalGeneration.from_pretrained("facebook/chameleon-7b", torch_dtype=torch.float16, device_map="cuda")
+model = ChameleonForConditionalGeneration.from_pretrained("facebook/chameleon-7b", torch_dtype=torch.bfloat16, device_map="cuda")
 
 # Get three different images
 url = "https://www.ilankelman.org/stopsigns/australia.jpg"
@@ -117,7 +117,7 @@ prompts = [
 
 # We can simply feed images in the order they have to be used in the text prompt
 # Each "<image>" token uses one image leaving the next for the subsequent "<image>" tokens
-inputs = processor(text=prompts, images=[image_stop, image_cats, image_snowman], padding=True, return_tensors="pt").to(device="cuda", dtype=torch.float16)
+inputs = processor(text=prompts, images=[image_stop, image_cats, image_snowman], padding=True, return_tensors="pt").to(device="cuda", dtype=torch.bfloat16)
 
 # Generate
 generate_ids = model.generate(**inputs, max_new_tokens=50)
@@ -153,7 +153,7 @@ from transformers import ChameleonForConditionalGeneration
 model_id = "facebook/chameleon-7b"
 model = ChameleonForConditionalGeneration.from_pretrained(
     model_id, 
-    torch_dtype=torch.float16, 
+    torch_dtype=torch.bfloat16, 
     low_cpu_mem_usage=True,
     attn_implementation="flash_attention_2"
 ).to(0)

src/transformers/models/chameleon/modeling_chameleon.py

@@ -1096,6 +1096,7 @@ class ChameleonPreTrainedModel(PreTrainedModel):
     _supports_quantized_cache = True
     _supports_cache_class = True
     _supports_static_cache = True
+    _supports_param_buffer_assignment = False
 
     def _init_weights(self, module):
         std = self.config.initializer_range