Update README.md

README.md

@@ -1,3 +1,63 @@
----
-license: apache-2.0
----
+---
+license: apache-2.0
+library_name: transformers
+---
+
+
+#### Quickstart
+
+```python
+from PIL import Image
+from transformers import AutoTokenizer, AutoModel, AutoImageProcessor, AutoModelForCausalLM
+from transformers.generation.configuration_utils import GenerationConfig
+import torch
+
+import sys
+sys.path.append(PATH_TO_BAAI_Emu3-Chat_MODEL)
+from processing_emu3 import Emu3Processor
+
+
+# model path
+EMU_HUB = "BAAI/Emu3-Chat"
+VQ_HUB = "BAAI/Emu3-VisionTokenizer"
+
+# prepare model and processor
+model = AutoModelForCausalLM.from_pretrained(
+    EMU_HUB,
+    device_map="cuda:0",
+    torch_dtype=torch.bfloat16,
+    attn_implementation="flash_attention_2",
+    trust_remote_code=True,
+)
+
+tokenizer = AutoTokenizer.from_pretrained(EMU_HUB, trust_remote_code=True)
+image_processor = AutoImageProcessor.from_pretrained(VQ_HUB, trust_remote_code=True)
+image_tokenizer = AutoModel.from_pretrained(VQ_HUB, device_map="cuda:0", trust_remote_code=True).eval()
+processor = Emu3Processor(image_processor, image_tokenizer, tokenizer)
+
+# prepare input
+text = "Please describe the image"
+image = Image.open("assets/demo.png")
+
+inputs = processor(
+    text=text,
+    image=image,
+    mode='U',
+    padding_side="left",
+    padding="longest",
+    return_tensors="pt",
+)
+
+# prepare hyper parameters
+GENERATION_CONFIG = GenerationConfig(pad_token_id=tokenizer.pad_token_id, bos_token_id=tokenizer.bos_token_id, eos_token_id=tokenizer.eos_token_id)
+
+# generate
+outputs = model.generate(
+    inputs.input_ids.to("cuda:0"),
+    GENERATION_CONFIG,
+    max_new_tokens=320,
+)
+
+outputs = outputs[:, inputs.input_ids.shape[-1]:]
+print(processor.batch_decode(outputs, skip_special_tokens=True)[0])
+```