Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
@@ -47,79 +47,54 @@ filetypes = {
|
|
47 |
"Audio": [".wav", ".mp3", ".flac", ".aac"],
|
48 |
}
|
49 |
|
|
|
50 |
def infer_filetype(ext):
|
51 |
return next((k for k, v in filetypes.items() if ext in v), None)
|
52 |
|
53 |
-
|
54 |
def uniform_sample(seq, n):
|
55 |
step = max(len(seq) // n, 1)
|
56 |
return seq[::step][:n]
|
57 |
|
58 |
-
|
59 |
def frames_from_video(path):
|
60 |
vr = VideoReader(path, ctx = cpu(0))
|
61 |
idx = uniform_sample(range(len(vr)), MAX_FRAMES)
|
62 |
batch = vr.get_batch(idx).asnumpy()
|
63 |
return [Image.fromarray(frame.astype("uint8")) for frame in batch]
|
64 |
|
65 |
-
|
66 |
def audio_from_video(path):
|
67 |
clip = VideoFileClip(path)
|
68 |
-
|
69 |
clip.close()
|
70 |
-
return librosa.to_mono(
|
71 |
-
|
72 |
|
73 |
def load_audio(path):
|
74 |
audio_np, _ = librosa.load(path, sr = AUDIO_SR, mono = True)
|
75 |
return audio_np
|
76 |
|
77 |
-
|
78 |
def build_video_omni(path, prefix, instruction):
|
79 |
frames = frames_from_video(path)
|
80 |
audio = audio_from_video(path)
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
|
86 |
-
|
87 |
-
|
88 |
-
)
|
89 |
-
|
90 |
|
91 |
def build_image_omni(path, prefix, instruction):
|
92 |
image = Image.open(path).convert("RGB")
|
93 |
-
return
|
94 |
-
frames = [image],
|
95 |
-
audio = None,
|
96 |
-
prefix = prefix,
|
97 |
-
instruction = instruction
|
98 |
-
)
|
99 |
-
|
100 |
|
101 |
def build_gif_omni(path, prefix, instruction):
|
102 |
-
img
|
103 |
-
frames = [
|
104 |
frames = uniform_sample(frames, MAX_FRAMES)
|
105 |
-
return
|
106 |
-
frames = frames,
|
107 |
-
audio = None,
|
108 |
-
prefix = prefix,
|
109 |
-
instruction = instruction
|
110 |
-
)
|
111 |
-
|
112 |
|
113 |
def build_audio_omni(path, prefix, instruction):
|
114 |
audio = load_audio(path)
|
115 |
-
return
|
116 |
-
frames = None,
|
117 |
-
audio = audio,
|
118 |
-
prefix = prefix,
|
119 |
-
instruction = instruction,
|
120 |
-
sr = AUDIO_SR
|
121 |
-
)
|
122 |
-
|
123 |
|
124 |
@spaces.GPU(duration = 60)
|
125 |
def generate(input,
|
@@ -136,30 +111,32 @@ def generate(input,
|
|
136 |
filetype = infer_filetype(extension)
|
137 |
if not filetype:
|
138 |
return "unsupported file type."
|
139 |
-
filename
|
140 |
-
prefix
|
141 |
-
builder_map
|
142 |
"Image": build_image_omni,
|
143 |
"GIF" : build_gif_omni,
|
144 |
"Video": build_video_omni,
|
145 |
"Audio": build_audio_omni
|
146 |
}
|
147 |
-
omni_content
|
148 |
-
sys_msg
|
149 |
-
msgs
|
150 |
-
output
|
151 |
-
msgs
|
152 |
-
tokenizer
|
153 |
-
sampling
|
154 |
-
temperature
|
155 |
-
top_p
|
156 |
-
top_k
|
157 |
-
repetition_penalty
|
158 |
-
max_new_tokens
|
159 |
-
omni_input
|
160 |
-
use_image_id
|
161 |
-
max_slice_nums
|
162 |
)
|
|
|
|
|
163 |
return output
|
164 |
|
165 |
def cloud():
|
|
|
47 |
"Audio": [".wav", ".mp3", ".flac", ".aac"],
|
48 |
}
|
49 |
|
50 |
+
# Functions
|
51 |
def infer_filetype(ext):
|
52 |
return next((k for k, v in filetypes.items() if ext in v), None)
|
53 |
|
|
|
54 |
def uniform_sample(seq, n):
|
55 |
step = max(len(seq) // n, 1)
|
56 |
return seq[::step][:n]
|
57 |
|
|
|
58 |
def frames_from_video(path):
|
59 |
vr = VideoReader(path, ctx = cpu(0))
|
60 |
idx = uniform_sample(range(len(vr)), MAX_FRAMES)
|
61 |
batch = vr.get_batch(idx).asnumpy()
|
62 |
return [Image.fromarray(frame.astype("uint8")) for frame in batch]
|
63 |
|
|
|
64 |
def audio_from_video(path):
|
65 |
clip = VideoFileClip(path)
|
66 |
+
wav = clip.audio.to_soundarray(fps = AUDIO_SR)
|
67 |
clip.close()
|
68 |
+
return librosa.to_mono(wav.T)
|
|
|
69 |
|
70 |
def load_audio(path):
|
71 |
audio_np, _ = librosa.load(path, sr = AUDIO_SR, mono = True)
|
72 |
return audio_np
|
73 |
|
|
|
74 |
def build_video_omni(path, prefix, instruction):
|
75 |
frames = frames_from_video(path)
|
76 |
audio = audio_from_video(path)
|
77 |
+
contents = [prefix + instruction]
|
78 |
+
total = max(len(frames), math.ceil(len(audio) / AUDIO_SR))
|
79 |
+
for i in range(total):
|
80 |
+
frame = frames[i] if i < len(frames) else frames[-1]
|
81 |
+
chunk = audio[AUDIO_SR * i : AUDIO_SR * (i + 1)]
|
82 |
+
contents.extend(["<unit>", frame, chunk])
|
83 |
+
return contents
|
|
|
|
|
84 |
|
85 |
def build_image_omni(path, prefix, instruction):
|
86 |
image = Image.open(path).convert("RGB")
|
87 |
+
return [prefix + instruction, image]
|
|
|
|
|
|
|
|
|
|
|
|
|
88 |
|
89 |
def build_gif_omni(path, prefix, instruction):
|
90 |
+
img = Image.open(path)
|
91 |
+
frames = [f.copy().convert("RGB") for f in ImageSequence.Iterator(img)]
|
92 |
frames = uniform_sample(frames, MAX_FRAMES)
|
93 |
+
return [prefix + instruction, *frames]
|
|
|
|
|
|
|
|
|
|
|
|
|
94 |
|
95 |
def build_audio_omni(path, prefix, instruction):
|
96 |
audio = load_audio(path)
|
97 |
+
return [prefix + instruction, audio]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
98 |
|
99 |
@spaces.GPU(duration = 60)
|
100 |
def generate(input,
|
|
|
111 |
filetype = infer_filetype(extension)
|
112 |
if not filetype:
|
113 |
return "unsupported file type."
|
114 |
+
filename = os.path.basename(input)
|
115 |
+
prefix = input_prefixes[filetype].replace("█", filename)
|
116 |
+
builder_map = {
|
117 |
"Image": build_image_omni,
|
118 |
"GIF" : build_gif_omni,
|
119 |
"Video": build_video_omni,
|
120 |
"Audio": build_audio_omni
|
121 |
}
|
122 |
+
omni_content = builder_map[filetype](input, prefix, instruction)
|
123 |
+
sys_msg = repo.get_sys_prompt(mode = "omni", language = "en")
|
124 |
+
msgs = [sys_msg, { "role": "user", "content": omni_content }]
|
125 |
+
output = repo.chat(
|
126 |
+
msgs = msgs,
|
127 |
+
tokenizer = tokenizer,
|
128 |
+
sampling = sampling,
|
129 |
+
temperature = temperature,
|
130 |
+
top_p = top_p,
|
131 |
+
top_k = top_k,
|
132 |
+
repetition_penalty = repetition_penalty,
|
133 |
+
max_new_tokens = max_tokens,
|
134 |
+
omni_input = True,
|
135 |
+
use_image_id = False,
|
136 |
+
max_slice_nums = 2
|
137 |
)
|
138 |
+
torch.cuda.empty_cache()
|
139 |
+
gc.collect()
|
140 |
return output
|
141 |
|
142 |
def cloud():
|