ryanzhangfan
commited on
Commit
•
40b3bb8
1
Parent(s):
c059b33
Update README.md
Browse files
README.md
CHANGED
@@ -63,7 +63,7 @@ model = AutoModelForCausalLM.from_pretrained(
|
|
63 |
trust_remote_code=True,
|
64 |
)
|
65 |
|
66 |
-
tokenizer = AutoTokenizer.from_pretrained(EMU_HUB, trust_remote_code=True)
|
67 |
image_processor = AutoImageProcessor.from_pretrained(VQ_HUB, trust_remote_code=True)
|
68 |
image_tokenizer = AutoModel.from_pretrained(VQ_HUB, device_map="cuda:0", trust_remote_code=True).eval()
|
69 |
processor = Emu3Processor(image_processor, image_tokenizer, tokenizer)
|
@@ -81,6 +81,7 @@ kwargs = dict(
|
|
81 |
ratio="1:1",
|
82 |
image_area=model.config.image_area,
|
83 |
return_tensors="pt",
|
|
|
84 |
)
|
85 |
pos_inputs = processor(text=prompt, **kwargs)
|
86 |
neg_inputs = processor(text=NEGATIVE_PROMPT, **kwargs)
|
@@ -95,7 +96,8 @@ GENERATION_CONFIG = GenerationConfig(
|
|
95 |
top_k=2048,
|
96 |
)
|
97 |
|
98 |
-
h
|
|
|
99 |
constrained_fn = processor.build_prefix_constrained_fn(h, w)
|
100 |
logits_processor = LogitsProcessorList([
|
101 |
UnbatchedClassifierFreeGuidanceLogitsProcessor(
|
@@ -113,7 +115,8 @@ logits_processor = LogitsProcessorList([
|
|
113 |
outputs = model.generate(
|
114 |
pos_inputs.input_ids.to("cuda:0"),
|
115 |
GENERATION_CONFIG,
|
116 |
-
logits_processor=logits_processor
|
|
|
117 |
)
|
118 |
|
119 |
mm_list = processor.decode(outputs[0])
|
@@ -121,5 +124,4 @@ for idx, im in enumerate(mm_list):
|
|
121 |
if not isinstance(im, Image.Image):
|
122 |
continue
|
123 |
im.save(f"result_{idx}.png")
|
124 |
-
|
125 |
```
|
|
|
63 |
trust_remote_code=True,
|
64 |
)
|
65 |
|
66 |
+
tokenizer = AutoTokenizer.from_pretrained(EMU_HUB, trust_remote_code=True, padding_side="left")
|
67 |
image_processor = AutoImageProcessor.from_pretrained(VQ_HUB, trust_remote_code=True)
|
68 |
image_tokenizer = AutoModel.from_pretrained(VQ_HUB, device_map="cuda:0", trust_remote_code=True).eval()
|
69 |
processor = Emu3Processor(image_processor, image_tokenizer, tokenizer)
|
|
|
81 |
ratio="1:1",
|
82 |
image_area=model.config.image_area,
|
83 |
return_tensors="pt",
|
84 |
+
padding="longest",
|
85 |
)
|
86 |
pos_inputs = processor(text=prompt, **kwargs)
|
87 |
neg_inputs = processor(text=NEGATIVE_PROMPT, **kwargs)
|
|
|
96 |
top_k=2048,
|
97 |
)
|
98 |
|
99 |
+
h = pos_inputs.image_size[:, 0]
|
100 |
+
w = pos_inputs.image_size[:, 1]
|
101 |
constrained_fn = processor.build_prefix_constrained_fn(h, w)
|
102 |
logits_processor = LogitsProcessorList([
|
103 |
UnbatchedClassifierFreeGuidanceLogitsProcessor(
|
|
|
115 |
outputs = model.generate(
|
116 |
pos_inputs.input_ids.to("cuda:0"),
|
117 |
GENERATION_CONFIG,
|
118 |
+
logits_processor=logits_processor,
|
119 |
+
attention_mask=pos_inputs.attention_mask.to("cuda:0"),
|
120 |
)
|
121 |
|
122 |
mm_list = processor.decode(outputs[0])
|
|
|
124 |
if not isinstance(im, Image.Image):
|
125 |
continue
|
126 |
im.save(f"result_{idx}.png")
|
|
|
127 |
```
|