Spaces:
Sleeping
Sleeping
j
commited on
Commit
·
c1f878d
1
Parent(s):
4bbc7e3
Changed the chatbot message format to use strings instead of dictionaries
Browse files- demo/web_demo_audio.py +54 -5
demo/web_demo_audio.py
CHANGED
@@ -135,20 +135,69 @@ def _launch_demo(args):
|
|
135 |
|
136 |
task_history = gr.State([])
|
137 |
|
138 |
-
# Update event handlers for new input components
|
139 |
def process_input(text, audio, chatbot, history):
|
|
|
140 |
content = []
|
|
|
|
|
141 |
if audio is not None:
|
142 |
content.append({'type': 'audio', 'audio_url': audio})
|
|
|
|
|
143 |
if text:
|
144 |
content.append({'type': 'text', 'text': text})
|
|
|
145 |
|
146 |
history.append({"role": "user", "content": content})
|
147 |
-
chatbot
|
148 |
-
|
149 |
-
None
|
150 |
-
])
|
151 |
return "", None, chatbot, history
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
152 |
|
153 |
submit_btn.click(
|
154 |
fn=process_input,
|
|
|
135 |
|
136 |
task_history = gr.State([])
|
137 |
|
|
|
138 |
def process_input(text, audio, chatbot, history):
|
139 |
+
"""Process input with correct message formatting for Chatbot."""
|
140 |
content = []
|
141 |
+
message_text = []
|
142 |
+
|
143 |
if audio is not None:
|
144 |
content.append({'type': 'audio', 'audio_url': audio})
|
145 |
+
message_text.append(f"[Audio file uploaded]")
|
146 |
+
|
147 |
if text:
|
148 |
content.append({'type': 'text', 'text': text})
|
149 |
+
message_text.append(text)
|
150 |
|
151 |
history.append({"role": "user", "content": content})
|
152 |
+
# Format message for chatbot as a string instead of dict
|
153 |
+
chatbot.append([" ".join(message_text), None])
|
|
|
|
|
154 |
return "", None, chatbot, history
|
155 |
+
|
156 |
+
def predict(chatbot, task_history):
|
157 |
+
"""Generate a response from the model."""
|
158 |
+
print(f"{task_history=}")
|
159 |
+
print(f"{chatbot=}")
|
160 |
+
text = processor.apply_chat_template(task_history, add_generation_prompt=True, tokenize=False)
|
161 |
+
audios = []
|
162 |
+
for message in task_history:
|
163 |
+
if isinstance(message["content"], list):
|
164 |
+
for ele in message["content"]:
|
165 |
+
if ele["type"] == "audio":
|
166 |
+
audios.append(
|
167 |
+
librosa.load(ele['audio_url'], sr=processor.feature_extractor.sampling_rate)[0]
|
168 |
+
)
|
169 |
+
|
170 |
+
if len(audios)==0:
|
171 |
+
audios=None
|
172 |
+
print(f"{text=}")
|
173 |
+
print(f"{audios=}")
|
174 |
+
inputs = processor(text=text, audios=audios, return_tensors="pt", padding=True)
|
175 |
+
inputs["input_ids"] = inputs.input_ids.cuda()
|
176 |
+
|
177 |
+
generate_ids = model.generate(**inputs, max_length=256)
|
178 |
+
generate_ids = generate_ids[:, inputs.input_ids.size(1):]
|
179 |
+
|
180 |
+
response = processor.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
|
181 |
+
print(f"{response=}")
|
182 |
+
task_history.append({'role': 'assistant',
|
183 |
+
'content': response})
|
184 |
+
chatbot.append((None, response))
|
185 |
+
return chatbot, task_history
|
186 |
+
|
187 |
+
# Update event handlers for new input components
|
188 |
+
# def process_input(text, audio, chatbot, history):
|
189 |
+
# content = []
|
190 |
+
# if audio is not None:
|
191 |
+
# content.append({'type': 'audio', 'audio_url': audio})
|
192 |
+
# if text:
|
193 |
+
# content.append({'type': 'text', 'text': text})
|
194 |
+
#
|
195 |
+
# history.append({"role": "user", "content": content})
|
196 |
+
# chatbot.append([
|
197 |
+
# {"text": text, "audio": audio},
|
198 |
+
# None
|
199 |
+
# ])
|
200 |
+
# return "", None, chatbot, history
|
201 |
|
202 |
submit_btn.click(
|
203 |
fn=process_input,
|