Spaces:
Running
on
Zero
Running
on
Zero
Update app.py
Browse files
app.py
CHANGED
@@ -36,11 +36,13 @@ footer {
|
|
36 |
}
|
37 |
'''
|
38 |
|
|
|
|
|
39 |
input_prefixes = {
|
40 |
-
"Image": "
|
41 |
-
"GIF": "
|
42 |
-
"Video": "
|
43 |
-
"Audio": "
|
44 |
}
|
45 |
|
46 |
filetypes = {
|
@@ -67,8 +69,7 @@ def frames_from_video(path):
|
|
67 |
def audio_from_video(path):
|
68 |
clip = VideoFileClip(path)
|
69 |
with tempfile.NamedTemporaryFile(suffix = ".wav", delete = True) as tmp:
|
70 |
-
clip.audio.write_audiofile(tmp.name, codec = "pcm_s16le",
|
71 |
-
fps = AUDIO_SR, verbose = False, logger = None)
|
72 |
audio_np, _ = librosa.load(tmp.name, sr = AUDIO_SR, mono = True)
|
73 |
clip.close()
|
74 |
return audio_np
|
@@ -77,10 +78,10 @@ def load_audio(path):
|
|
77 |
audio_np, _ = librosa.load(path, sr = AUDIO_SR, mono = True)
|
78 |
return audio_np
|
79 |
|
80 |
-
def build_video_omni(path,
|
81 |
frames = frames_from_video(path)
|
82 |
audio = audio_from_video(path)
|
83 |
-
contents = [
|
84 |
total = max(len(frames), math.ceil(len(audio) / AUDIO_SR))
|
85 |
for i in range(total):
|
86 |
frame = frames[i] if i < len(frames) else frames[-1]
|
@@ -88,21 +89,21 @@ def build_video_omni(path, prefix, instruction):
|
|
88 |
contents.extend(["<unit>", frame, chunk])
|
89 |
return contents
|
90 |
|
91 |
-
def build_image_omni(path,
|
92 |
image = Image.open(path).convert("RGB")
|
93 |
-
return [
|
94 |
|
95 |
-
def build_gif_omni(path,
|
96 |
img = Image.open(path)
|
97 |
frames = [f.copy().convert("RGB") for f in ImageSequence.Iterator(img)]
|
98 |
frames = uniform_sample(frames, MAX_FRAMES)
|
99 |
-
return [
|
100 |
|
101 |
-
def build_audio_omni(path,
|
102 |
audio = load_audio(path)
|
103 |
-
return [
|
104 |
|
105 |
-
@spaces.GPU(duration
|
106 |
def generate(input,
|
107 |
instruction = DEFAULT_INPUT,
|
108 |
sampling = False,
|
@@ -111,12 +112,12 @@ def generate(input,
|
|
111 |
top_k = 100,
|
112 |
repetition_penalty = 1.05,
|
113 |
max_tokens = 512):
|
114 |
-
if not input:
|
115 |
-
|
116 |
extension = os.path.splitext(input)[1].lower()
|
117 |
filetype = infer_filetype(extension)
|
118 |
-
if not filetype:
|
119 |
-
|
120 |
filename = os.path.basename(input)
|
121 |
prefix = input_prefixes[filetype].replace("β", filename)
|
122 |
builder_map = {
|
@@ -125,9 +126,14 @@ def generate(input,
|
|
125 |
"Video": build_video_omni,
|
126 |
"Audio": build_audio_omni
|
127 |
}
|
128 |
-
|
|
|
|
|
129 |
sys_msg = repo.get_sys_prompt(mode = "omni", language = "en")
|
130 |
msgs = [sys_msg, { "role": "user", "content": omni_content }]
|
|
|
|
|
|
|
131 |
output = repo.chat(
|
132 |
msgs = msgs,
|
133 |
tokenizer = tokenizer,
|
@@ -141,8 +147,10 @@ def generate(input,
|
|
141 |
use_image_id = False,
|
142 |
max_slice_nums = 2
|
143 |
)
|
|
|
144 |
torch.cuda.empty_cache()
|
145 |
gc.collect()
|
|
|
146 |
return output
|
147 |
|
148 |
def cloud():
|
|
|
36 |
}
|
37 |
'''
|
38 |
|
39 |
+
global_instruction = "Describe the given content with as much keywords and always take a guess."
|
40 |
+
|
41 |
input_prefixes = {
|
42 |
+
"Image": "A image file called β has been attached, describe the image content.",
|
43 |
+
"GIF": "A GIF file called β has been attached, describe the GIF content.",
|
44 |
+
"Video": "A audio video file called β has been attached, describe the video content and the audio content.",
|
45 |
+
"Audio": "A audio file called β has been attached, describe the audio content.",
|
46 |
}
|
47 |
|
48 |
filetypes = {
|
|
|
69 |
def audio_from_video(path):
|
70 |
clip = VideoFileClip(path)
|
71 |
with tempfile.NamedTemporaryFile(suffix = ".wav", delete = True) as tmp:
|
72 |
+
clip.audio.write_audiofile(tmp.name, codec = "pcm_s16le", fps = AUDIO_SR, verbose = False, logger = None)
|
|
|
73 |
audio_np, _ = librosa.load(tmp.name, sr = AUDIO_SR, mono = True)
|
74 |
clip.close()
|
75 |
return audio_np
|
|
|
78 |
audio_np, _ = librosa.load(path, sr = AUDIO_SR, mono = True)
|
79 |
return audio_np
|
80 |
|
81 |
+
def build_video_omni(path, instruction):
|
82 |
frames = frames_from_video(path)
|
83 |
audio = audio_from_video(path)
|
84 |
+
contents = [instruction]
|
85 |
total = max(len(frames), math.ceil(len(audio) / AUDIO_SR))
|
86 |
for i in range(total):
|
87 |
frame = frames[i] if i < len(frames) else frames[-1]
|
|
|
89 |
contents.extend(["<unit>", frame, chunk])
|
90 |
return contents
|
91 |
|
92 |
+
def build_image_omni(path, instruction):
|
93 |
image = Image.open(path).convert("RGB")
|
94 |
+
return [instruction, image]
|
95 |
|
96 |
+
def build_gif_omni(path, instruction):
|
97 |
img = Image.open(path)
|
98 |
frames = [f.copy().convert("RGB") for f in ImageSequence.Iterator(img)]
|
99 |
frames = uniform_sample(frames, MAX_FRAMES)
|
100 |
+
return [instruction, *frames]
|
101 |
|
102 |
+
def build_audio_omni(path, instruction):
|
103 |
audio = load_audio(path)
|
104 |
+
return [instruction, audio]
|
105 |
|
106 |
+
@spaces.GPU(duration=30)
|
107 |
def generate(input,
|
108 |
instruction = DEFAULT_INPUT,
|
109 |
sampling = False,
|
|
|
112 |
top_k = 100,
|
113 |
repetition_penalty = 1.05,
|
114 |
max_tokens = 512):
|
115 |
+
if not input: return "no input provided."
|
116 |
+
|
117 |
extension = os.path.splitext(input)[1].lower()
|
118 |
filetype = infer_filetype(extension)
|
119 |
+
if not filetype: return "unsupported file type."
|
120 |
+
|
121 |
filename = os.path.basename(input)
|
122 |
prefix = input_prefixes[filetype].replace("β", filename)
|
123 |
builder_map = {
|
|
|
126 |
"Video": build_video_omni,
|
127 |
"Audio": build_audio_omni
|
128 |
}
|
129 |
+
|
130 |
+
instruction = f"{global_instruction}\n{prefix}\n{instruction}"
|
131 |
+
omni_content = builder_map[filetype](input, instruction)
|
132 |
sys_msg = repo.get_sys_prompt(mode = "omni", language = "en")
|
133 |
msgs = [sys_msg, { "role": "user", "content": omni_content }]
|
134 |
+
|
135 |
+
print(msgs)
|
136 |
+
|
137 |
output = repo.chat(
|
138 |
msgs = msgs,
|
139 |
tokenizer = tokenizer,
|
|
|
147 |
use_image_id = False,
|
148 |
max_slice_nums = 2
|
149 |
)
|
150 |
+
|
151 |
torch.cuda.empty_cache()
|
152 |
gc.collect()
|
153 |
+
|
154 |
return output
|
155 |
|
156 |
def cloud():
|