Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
|
@@ -17,24 +17,22 @@ from transformers.generation import (
|
|
| 17 |
)
|
| 18 |
import torch
|
| 19 |
from emu3.mllm.processing_emu3 import Emu3Processor
|
| 20 |
-
import spaces
|
| 21 |
|
| 22 |
import io
|
| 23 |
import base64
|
| 24 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 25 |
def image2str(image):
|
| 26 |
buf = io.BytesIO()
|
| 27 |
image.save(buf, format="PNG")
|
| 28 |
i_str = base64.b64encode(buf.getvalue()).decode()
|
| 29 |
return f'<div style="float:left"><img src="data:image/png;base64, {i_str}"></div>'
|
| 30 |
|
| 31 |
-
# Install flash attention, skipping CUDA build if necessary
|
| 32 |
-
subprocess.run(
|
| 33 |
-
"pip install flash-attn --no-build-isolation",
|
| 34 |
-
env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
|
| 35 |
-
shell=True,
|
| 36 |
-
)
|
| 37 |
-
|
| 38 |
print(gr.__version__)
|
| 39 |
|
| 40 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
@@ -46,7 +44,6 @@ VQ_HUB = "BAAI/Emu3-VisionTokenizer"
|
|
| 46 |
|
| 47 |
|
| 48 |
# uncomment to use gen model
|
| 49 |
-
"""
|
| 50 |
# Prepare models and processors
|
| 51 |
# Emu3-Gen model and processor
|
| 52 |
gen_model = AutoModelForCausalLM.from_pretrained(
|
|
@@ -55,7 +52,15 @@ gen_model = AutoModelForCausalLM.from_pretrained(
|
|
| 55 |
torch_dtype=torch.bfloat16,
|
| 56 |
attn_implementation="flash_attention_2",
|
| 57 |
trust_remote_code=True,
|
| 58 |
-
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 59 |
|
| 60 |
tokenizer = AutoTokenizer.from_pretrained(EMU_CHAT_HUB, trust_remote_code=True)
|
| 61 |
image_processor = AutoImageProcessor.from_pretrained(
|
|
@@ -66,14 +71,12 @@ image_tokenizer = AutoModel.from_pretrained(
|
|
| 66 |
).eval()
|
| 67 |
|
| 68 |
print(device)
|
| 69 |
-
gen_model.to(device)
|
| 70 |
image_tokenizer.to(device)
|
| 71 |
|
| 72 |
processor = Emu3Processor(
|
| 73 |
image_processor, image_tokenizer, tokenizer
|
| 74 |
)
|
| 75 |
|
| 76 |
-
@spaces.GPU(duration=300)
|
| 77 |
def generate_image(prompt):
|
| 78 |
POSITIVE_PROMPT = " masterpiece, film grained, best quality."
|
| 79 |
NEGATIVE_PROMPT = (
|
|
@@ -104,6 +107,9 @@ def generate_image(prompt):
|
|
| 104 |
top_k=2048,
|
| 105 |
)
|
| 106 |
|
|
|
|
|
|
|
|
|
|
| 107 |
h, w = pos_inputs.image_size[0]
|
| 108 |
constrained_fn = processor.build_prefix_constrained_fn(h, w)
|
| 109 |
logits_processor = LogitsProcessorList(
|
|
@@ -128,54 +134,17 @@ def generate_image(prompt):
|
|
| 128 |
)
|
| 129 |
|
| 130 |
mm_list = processor.decode(outputs[0])
|
|
|
|
| 131 |
for idx, im in enumerate(mm_list):
|
| 132 |
if isinstance(im, Image.Image):
|
| 133 |
-
|
| 134 |
-
|
| 135 |
|
| 136 |
-
|
| 137 |
-
|
| 138 |
-
|
| 139 |
-
|
| 140 |
-
# Use Emu3-Gen for image generation
|
| 141 |
-
generated_image = generate_image(user_input)
|
| 142 |
-
if generated_image is not None:
|
| 143 |
-
# Append the user input and generated image to the history
|
| 144 |
-
history = history + [(user_input, image2str(generated_image))]
|
| 145 |
-
else:
|
| 146 |
-
# If image generation failed, respond with an error message
|
| 147 |
-
history = history + [
|
| 148 |
-
(user_input, "Sorry, I could not generate an image.")
|
| 149 |
-
]
|
| 150 |
-
return history, history, gr.update(value=None)
|
| 151 |
-
"""
|
| 152 |
|
| 153 |
-
# Emu3-Chat model and processor
|
| 154 |
-
chat_model = AutoModelForCausalLM.from_pretrained(
|
| 155 |
-
EMU_CHAT_HUB,
|
| 156 |
-
device_map="cpu",
|
| 157 |
-
torch_dtype=torch.bfloat16,
|
| 158 |
-
attn_implementation="flash_attention_2",
|
| 159 |
-
trust_remote_code=True,
|
| 160 |
-
)
|
| 161 |
-
|
| 162 |
-
tokenizer = AutoTokenizer.from_pretrained(EMU_CHAT_HUB, trust_remote_code=True)
|
| 163 |
-
image_processor = AutoImageProcessor.from_pretrained(
|
| 164 |
-
VQ_HUB, trust_remote_code=True
|
| 165 |
-
)
|
| 166 |
-
image_tokenizer = AutoModel.from_pretrained(
|
| 167 |
-
VQ_HUB, device_map="cpu", trust_remote_code=True
|
| 168 |
-
).eval()
|
| 169 |
-
|
| 170 |
-
print(device)
|
| 171 |
-
chat_model.to(device)
|
| 172 |
-
image_tokenizer.to(device)
|
| 173 |
-
|
| 174 |
-
processor = Emu3Processor(
|
| 175 |
-
image_processor, image_tokenizer, tokenizer
|
| 176 |
-
)
|
| 177 |
-
|
| 178 |
-
@spaces.GPU
|
| 179 |
def vision_language_understanding(image, text):
|
| 180 |
inputs = processor(
|
| 181 |
text=text,
|
|
@@ -194,6 +163,9 @@ def vision_language_understanding(image, text):
|
|
| 194 |
max_new_tokens=320,
|
| 195 |
)
|
| 196 |
|
|
|
|
|
|
|
|
|
|
| 197 |
# Generate
|
| 198 |
outputs = chat_model.generate(
|
| 199 |
inputs.input_ids.to(device),
|
|
@@ -203,8 +175,13 @@ def vision_language_understanding(image, text):
|
|
| 203 |
|
| 204 |
outputs = outputs[:, inputs.input_ids.shape[-1] :]
|
| 205 |
response = processor.batch_decode(outputs, skip_special_tokens=True)[0]
|
|
|
|
|
|
|
|
|
|
|
|
|
| 206 |
return response
|
| 207 |
|
|
|
|
| 208 |
def chat(history, user_input, user_image):
|
| 209 |
if user_image is not None:
|
| 210 |
# Use Emu3-Chat for vision-language understanding
|
|
@@ -212,21 +189,32 @@ def chat(history, user_input, user_image):
|
|
| 212 |
# Append the user input and response to the history
|
| 213 |
history = history + [(image2str(user_image) + "<br>" + user_input, response)]
|
| 214 |
else:
|
| 215 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 216 |
|
| 217 |
return history, history, gr.update(value=None)
|
| 218 |
|
| 219 |
-
|
| 220 |
-
# """
|
| 221 |
-
|
| 222 |
def clear_input():
|
| 223 |
return gr.update(value="")
|
| 224 |
|
|
|
|
| 225 |
with gr.Blocks() as demo:
|
| 226 |
gr.Markdown("# Emu3 Chatbot Demo")
|
| 227 |
gr.Markdown(
|
| 228 |
"This is a chatbot demo for image generation and vision-language understanding using Emu3 models."
|
| 229 |
)
|
|
|
|
|
|
|
|
|
|
| 230 |
|
| 231 |
chatbot = gr.Chatbot()
|
| 232 |
state = gr.State([])
|
|
|
|
| 17 |
)
|
| 18 |
import torch
|
| 19 |
from emu3.mllm.processing_emu3 import Emu3Processor
|
|
|
|
| 20 |
|
| 21 |
import io
|
| 22 |
import base64
|
| 23 |
|
| 24 |
+
subprocess.run(
|
| 25 |
+
"pip3 install flash-attn --no-build-isolation",
|
| 26 |
+
env={"FLASH_ATTENTION_SKIP_CUDA_BUILD": "TRUE"},
|
| 27 |
+
shell=True,
|
| 28 |
+
)
|
| 29 |
+
|
| 30 |
def image2str(image):
|
| 31 |
buf = io.BytesIO()
|
| 32 |
image.save(buf, format="PNG")
|
| 33 |
i_str = base64.b64encode(buf.getvalue()).decode()
|
| 34 |
return f'<div style="float:left"><img src="data:image/png;base64, {i_str}"></div>'
|
| 35 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
print(gr.__version__)
|
| 37 |
|
| 38 |
device = "cuda" if torch.cuda.is_available() else "cpu"
|
|
|
|
| 44 |
|
| 45 |
|
| 46 |
# uncomment to use gen model
|
|
|
|
| 47 |
# Prepare models and processors
|
| 48 |
# Emu3-Gen model and processor
|
| 49 |
gen_model = AutoModelForCausalLM.from_pretrained(
|
|
|
|
| 52 |
torch_dtype=torch.bfloat16,
|
| 53 |
attn_implementation="flash_attention_2",
|
| 54 |
trust_remote_code=True,
|
| 55 |
+
).eval()
|
| 56 |
+
|
| 57 |
+
chat_model = AutoModelForCausalLM.from_pretrained(
|
| 58 |
+
EMU_CHAT_HUB,
|
| 59 |
+
device_map="cpu",
|
| 60 |
+
torch_dtype=torch.bfloat16,
|
| 61 |
+
attn_implementation="flash_attention_2",
|
| 62 |
+
trust_remote_code=True,
|
| 63 |
+
).eval()
|
| 64 |
|
| 65 |
tokenizer = AutoTokenizer.from_pretrained(EMU_CHAT_HUB, trust_remote_code=True)
|
| 66 |
image_processor = AutoImageProcessor.from_pretrained(
|
|
|
|
| 71 |
).eval()
|
| 72 |
|
| 73 |
print(device)
|
|
|
|
| 74 |
image_tokenizer.to(device)
|
| 75 |
|
| 76 |
processor = Emu3Processor(
|
| 77 |
image_processor, image_tokenizer, tokenizer
|
| 78 |
)
|
| 79 |
|
|
|
|
| 80 |
def generate_image(prompt):
|
| 81 |
POSITIVE_PROMPT = " masterpiece, film grained, best quality."
|
| 82 |
NEGATIVE_PROMPT = (
|
|
|
|
| 107 |
top_k=2048,
|
| 108 |
)
|
| 109 |
|
| 110 |
+
torch.cuda.empty_cache()
|
| 111 |
+
gen_model.to(device)
|
| 112 |
+
|
| 113 |
h, w = pos_inputs.image_size[0]
|
| 114 |
constrained_fn = processor.build_prefix_constrained_fn(h, w)
|
| 115 |
logits_processor = LogitsProcessorList(
|
|
|
|
| 134 |
)
|
| 135 |
|
| 136 |
mm_list = processor.decode(outputs[0])
|
| 137 |
+
result = None
|
| 138 |
for idx, im in enumerate(mm_list):
|
| 139 |
if isinstance(im, Image.Image):
|
| 140 |
+
result = im
|
| 141 |
+
break
|
| 142 |
|
| 143 |
+
gen_model.cpu()
|
| 144 |
+
torch.cuda.empty_cache()
|
| 145 |
+
|
| 146 |
+
return result
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 147 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 148 |
def vision_language_understanding(image, text):
|
| 149 |
inputs = processor(
|
| 150 |
text=text,
|
|
|
|
| 163 |
max_new_tokens=320,
|
| 164 |
)
|
| 165 |
|
| 166 |
+
torch.cuda.empty_cache()
|
| 167 |
+
chat_model.to(device)
|
| 168 |
+
|
| 169 |
# Generate
|
| 170 |
outputs = chat_model.generate(
|
| 171 |
inputs.input_ids.to(device),
|
|
|
|
| 175 |
|
| 176 |
outputs = outputs[:, inputs.input_ids.shape[-1] :]
|
| 177 |
response = processor.batch_decode(outputs, skip_special_tokens=True)[0]
|
| 178 |
+
|
| 179 |
+
chat_model.cpu()
|
| 180 |
+
torch.cuda.empty_cache()
|
| 181 |
+
|
| 182 |
return response
|
| 183 |
|
| 184 |
+
|
| 185 |
def chat(history, user_input, user_image):
|
| 186 |
if user_image is not None:
|
| 187 |
# Use Emu3-Chat for vision-language understanding
|
|
|
|
| 189 |
# Append the user input and response to the history
|
| 190 |
history = history + [(image2str(user_image) + "<br>" + user_input, response)]
|
| 191 |
else:
|
| 192 |
+
# Use Emu3-Gen for image generation
|
| 193 |
+
generated_image = generate_image(user_input)
|
| 194 |
+
if generated_image is not None:
|
| 195 |
+
# Append the user input and generated image to the history
|
| 196 |
+
history = history + [(user_input, image2str(generated_image))]
|
| 197 |
+
else:
|
| 198 |
+
# If image generation failed, respond with an error message
|
| 199 |
+
history = history + [
|
| 200 |
+
(user_input, "Sorry, I could not generate an image.")
|
| 201 |
+
]
|
| 202 |
|
| 203 |
return history, history, gr.update(value=None)
|
| 204 |
|
| 205 |
+
|
|
|
|
|
|
|
| 206 |
def clear_input():
|
| 207 |
return gr.update(value="")
|
| 208 |
|
| 209 |
+
|
| 210 |
with gr.Blocks() as demo:
|
| 211 |
gr.Markdown("# Emu3 Chatbot Demo")
|
| 212 |
gr.Markdown(
|
| 213 |
"This is a chatbot demo for image generation and vision-language understanding using Emu3 models."
|
| 214 |
)
|
| 215 |
+
gr.Markdown(
|
| 216 |
+
"Please pass only text input for image generation and both image and text for vision-language understanding"
|
| 217 |
+
)
|
| 218 |
|
| 219 |
chatbot = gr.Chatbot()
|
| 220 |
state = gr.State([])
|