Update app.py
Browse files
app.py
CHANGED
@@ -9,6 +9,13 @@ import torch
|
|
9 |
from gtts import gTTS
|
10 |
import tempfile
|
11 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
# —––––––– Page Config —–––––––
|
13 |
st.set_page_config(page_title="Magic Story Generator (Qwen2.5)", layout="centered")
|
14 |
st.title("📖✨ Turn Images into Children's Stories (Qwen2.5-Omni-7B)")
|
@@ -27,7 +34,7 @@ def load_clients():
|
|
27 |
token=hf_token
|
28 |
)
|
29 |
|
30 |
-
# 2) Load Qwen2.5-Omni model & tokenizer
|
31 |
t0 = time.time()
|
32 |
tokenizer = AutoTokenizer.from_pretrained(
|
33 |
"Qwen/Qwen2.5-Omni-7B",
|
@@ -40,7 +47,6 @@ def load_clients():
|
|
40 |
torch_dtype=torch.bfloat16,
|
41 |
attn_implementation="flash_attention_2"
|
42 |
)
|
43 |
-
# 3) Build text2text pipeline
|
44 |
storyteller = pipeline(
|
45 |
task="text2text-generation",
|
46 |
model=model,
|
@@ -85,12 +91,9 @@ def generate_story(caption: str) -> str:
|
|
85 |
st.text(f"⏱ Generated in {gen_time:.1f}s on GPU/CPU")
|
86 |
|
87 |
story = result[0]["generated_text"].strip()
|
88 |
-
# Enforce ≤100 words
|
89 |
words = story.split()
|
90 |
if len(words) > 100:
|
91 |
-
story = " ".join(words[:100])
|
92 |
-
if not story.endswith('.'):
|
93 |
-
story += '.'
|
94 |
return story
|
95 |
|
96 |
# —––––––– Main App —–––––––
|
|
|
9 |
from gtts import gTTS
|
10 |
import tempfile
|
11 |
|
12 |
+
# —––––––– Requirements —–––––––
|
13 |
+
# This app uses a Hugging Face Transformers version that supports
|
14 |
+
# the Qwen2.5-Omni architecture via `trust_remote_code`.
|
15 |
+
# Install using:
|
16 |
+
# pip install git+https://github.com/huggingface/transformers.git
|
17 |
+
# and the rest of the requirements listed at the end.
|
18 |
+
|
19 |
# —––––––– Page Config —–––––––
|
20 |
st.set_page_config(page_title="Magic Story Generator (Qwen2.5)", layout="centered")
|
21 |
st.title("📖✨ Turn Images into Children's Stories (Qwen2.5-Omni-7B)")
|
|
|
34 |
token=hf_token
|
35 |
)
|
36 |
|
37 |
+
# 2) Load Qwen2.5-Omni model & tokenizer via remote code
|
38 |
t0 = time.time()
|
39 |
tokenizer = AutoTokenizer.from_pretrained(
|
40 |
"Qwen/Qwen2.5-Omni-7B",
|
|
|
47 |
torch_dtype=torch.bfloat16,
|
48 |
attn_implementation="flash_attention_2"
|
49 |
)
|
|
|
50 |
storyteller = pipeline(
|
51 |
task="text2text-generation",
|
52 |
model=model,
|
|
|
91 |
st.text(f"⏱ Generated in {gen_time:.1f}s on GPU/CPU")
|
92 |
|
93 |
story = result[0]["generated_text"].strip()
|
|
|
94 |
words = story.split()
|
95 |
if len(words) > 100:
|
96 |
+
story = " ".join(words[:100]) + ('.' if not story.endswith('.') else '')
|
|
|
|
|
97 |
return story
|
98 |
|
99 |
# —––––––– Main App —–––––––
|