Update app.py
Browse files
app.py
CHANGED
|
@@ -46,21 +46,20 @@ def predict(message, history):
|
|
| 46 |
stop = StopOnTokens()
|
| 47 |
|
| 48 |
# Formatting the input for the model.
|
| 49 |
-
messages = system_prompt + sft_end_token.join([sft_end_token.join([f"\n{sft_start_token}{user_role}\n" + item[0], f"\n{sft_start_token}{assistant_role}\n" + item[1]])
|
| 50 |
-
|
| 51 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 52 |
streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
|
| 53 |
generate_kwargs = dict(
|
| 54 |
model_inputs,
|
| 55 |
streamer=streamer,
|
| 56 |
-
max_new_tokens=
|
| 57 |
-
do_sample=
|
| 58 |
-
|
| 59 |
-
top_k= 60,
|
| 60 |
-
temperature=0.2,
|
| 61 |
-
num_beams=1,
|
| 62 |
-
stopping_criteria=StoppingCriteriaList([stop]),
|
| 63 |
-
repetition_penalty=1.1,
|
| 64 |
)
|
| 65 |
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
| 66 |
t.start() # Starting the generation in a separate thread.
|
|
@@ -93,7 +92,7 @@ placeholder = """
|
|
| 93 |
</div>
|
| 94 |
"""
|
| 95 |
|
| 96 |
-
chatbot = gr.Chatbot(label='Sailor', placeholder=
|
| 97 |
with gr.Blocks(theme=gr.themes.Soft(), fill_height=True) as demo:
|
| 98 |
# gr.Markdown("""<center><font size=8>Sailor-Chat Bot⚓</center>""")
|
| 99 |
gr.Markdown("""<p align="center"><img src="https://github.com/sail-sg/sailor-llm/raw/main/misc/wide_sailor_banner.jpg" style="height: 110px"/><p>""")
|
|
|
|
| 46 |
stop = StopOnTokens()
|
| 47 |
|
| 48 |
# Formatting the input for the model.
|
| 49 |
+
# messages = system_prompt + sft_end_token.join([sft_end_token.join([f"\n{sft_start_token}{user_role}\n" + item[0], f"\n{sft_start_token}{assistant_role}\n" + item[1]])
|
| 50 |
+
# for item in history_transformer_format])
|
| 51 |
+
|
| 52 |
+
messages = [{user_role: item[0], assistant_role: item[1]} for item in history_transformer_format]
|
| 53 |
+
model_inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt").to(device)
|
| 54 |
+
# model_inputs = tokenizer([messages], return_tensors="pt").to(device)
|
| 55 |
+
|
| 56 |
streamer = TextIteratorStreamer(tokenizer, timeout=10., skip_prompt=True, skip_special_tokens=True)
|
| 57 |
generate_kwargs = dict(
|
| 58 |
model_inputs,
|
| 59 |
streamer=streamer,
|
| 60 |
+
max_new_tokens=1024,
|
| 61 |
+
do_sample=False,
|
| 62 |
+
# stopping_criteria=StoppingCriteriaList([stop])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
)
|
| 64 |
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
| 65 |
t.start() # Starting the generation in a separate thread.
|
|
|
|
| 92 |
</div>
|
| 93 |
"""
|
| 94 |
|
| 95 |
+
chatbot = gr.Chatbot(label='Sailor', placeholder=None)
|
| 96 |
with gr.Blocks(theme=gr.themes.Soft(), fill_height=True) as demo:
|
| 97 |
# gr.Markdown("""<center><font size=8>Sailor-Chat Bot⚓</center>""")
|
| 98 |
gr.Markdown("""<p align="center"><img src="https://github.com/sail-sg/sailor-llm/raw/main/misc/wide_sailor_banner.jpg" style="height: 110px"/><p>""")
|