Spaces:
Runtime error
Runtime error
File size: 5,285 Bytes
ef366f8 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 |
import os
import gradio as gr
from text_generation import Client, InferenceAPIClient
def get_client(model: str):
if model == "Rallio67/joi_20B_instruct_alpha":
return Client(os.getenv("API_URL"))
return InferenceAPIClient(model, token=os.getenv("HF_TOKEN", None))
def get_usernames(model: str):
if model == "Rallio67/joi_20B_instruct_alpha":
return "User: ", "Joi: "
return "User: ", "Assistant: "
def predict(
model: str,
inputs: str,
top_p: float,
temperature: float,
top_k: int,
repetition_penalty: float,
watermark: bool,
chatbot,
history,
):
client = get_client(model)
user_name, assistant_name = get_usernames(model)
history.append(inputs)
past = []
for data in chatbot:
user_data, model_data = data
if not user_data.startswith(user_name):
user_data = user_name + user_data
if not model_data.startswith("\n\n" + assistant_name):
model_data = "\n\n" + assistant_name + model_data
past.append(user_data + model_data + "\n\n")
if not inputs.startswith(user_name):
inputs = user_name + inputs
total_inputs = "".join(past) + inputs + "\n\n" + assistant_name
print(total_inputs)
partial_words = ""
for i, response in enumerate(client.generate_stream(
total_inputs,
top_p=top_p,
top_k=top_k,
repetition_penalty=repetition_penalty,
watermark=watermark,
temperature=temperature,
max_new_tokens=1000,
stop_sequences=["User:"],
)):
if response.token.special:
continue
partial_words = partial_words + response.token.text
if partial_words.endswith(user_name.rstrip()):
partial_words = partial_words.rstrip(user_name.rstrip())
if i == 0:
history.append(" " + partial_words)
else:
history[-1] = partial_words
chat = [
(history[i], history[i + 1]) for i in range(0, len(history) - 1, 2)
]
yield chat, history
def reset_textbox():
return gr.update(value="")
title = """<h1 align="center">🔥Large Language Model API 🚀Streaming🚀</h1>"""
description = """Language models can be conditioned to act like dialogue agents through a conversational prompt that typically takes the form:
```
User: <utterance>
Assistant: <utterance>
User: <utterance>
Assistant: <utterance>
...
```
In this app, you can explore the outputs of multiple LLMs when prompted in this way.
"""
with gr.Blocks(
css="""#col_container {width: 1000px; margin-left: auto; margin-right: auto;}
#chatbot {height: 520px; overflow: auto;}"""
) as demo:
gr.HTML(title)
with gr.Column(elem_id="col_container"):
model = gr.Radio(
value="Rallio67/joi_20B_instruct_alpha",
choices=[
"Rallio67/joi_20B_instruct_alpha",
"google/flan-t5-xxl",
"google/flan-ul2",
"bigscience/bloom",
"bigscience/bloomz",
"EleutherAI/gpt-neox-20b",
],
label="Model",
interactive=True,
)
chatbot = gr.Chatbot(elem_id="chatbot")
inputs = gr.Textbox(
placeholder="Hi there!", label="Type an input and press Enter"
)
state = gr.State([])
b1 = gr.Button()
with gr.Accordion("Parameters", open=False):
top_p = gr.Slider(
minimum=-0,
maximum=1.0,
value=0.95,
step=0.05,
interactive=True,
label="Top-p (nucleus sampling)",
)
temperature = gr.Slider(
minimum=-0,
maximum=5.0,
value=0.5,
step=0.1,
interactive=True,
label="Temperature",
)
top_k = gr.Slider(
minimum=1,
maximum=50,
value=4,
step=1,
interactive=True,
label="Top-k",
)
repetition_penalty = gr.Slider(
minimum=0.1,
maximum=3.0,
value=1.03,
step=0.01,
interactive=True,
label="Repetition Penalty",
)
watermark = gr.Checkbox(value=True, label="Text watermarking")
inputs.submit(
predict,
[
model,
inputs,
top_p,
temperature,
top_k,
repetition_penalty,
watermark,
chatbot,
state,
],
[chatbot, state],
)
b1.click(
predict,
[
model,
inputs,
top_p,
temperature,
top_k,
repetition_penalty,
watermark,
chatbot,
state,
],
[chatbot, state],
)
b1.click(reset_textbox, [], [inputs])
inputs.submit(reset_textbox, [], [inputs])
gr.Markdown(description)
demo.queue().launch(debug=True)
|