Spaces:
Runtime error
Runtime error
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,266 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import os
|
3 |
+
import shutil
|
4 |
+
import requests
|
5 |
+
|
6 |
+
import gradio as gr
|
7 |
+
from huggingface_hub import Repository
|
8 |
+
from text_generation import Client
|
9 |
+
|
10 |
+
HF_TOKEN = "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"
|
11 |
+
|
12 |
+
API_URL = "https://api-inference.huggingface.co/models/bigcode/starcoder"
|
13 |
+
API_URL_BASE ="https://api-inference.huggingface.co/models/bigcode/starcoderbase"
|
14 |
+
API_URL_PLUS = "https://api-inference.huggingface.co/models/bigcode/starcoderplus"
|
15 |
+
|
16 |
+
FIM_PREFIX = "<fim_prefix>"
|
17 |
+
FIM_MIDDLE = "<fim_middle>"
|
18 |
+
FIM_SUFFIX = "<fim_suffix>"
|
19 |
+
|
20 |
+
FIM_INDICATOR = "<FILL_HERE>"
|
21 |
+
|
22 |
+
FORMATS = """## Model Formats
|
23 |
+
|
24 |
+
The model is pretrained on code and is formatted with special tokens in addition to the pure code data,\
|
25 |
+
such as prefixes specifying the source of the file or tokens separating code from a commit message.\
|
26 |
+
Use these templates to explore the model's capacities:
|
27 |
+
|
28 |
+
### 1. Prefixes 🏷️
|
29 |
+
For pure code files, use any combination of the following prefixes:
|
30 |
+
|
31 |
+
```
|
32 |
+
<reponame>REPONAME<filename>FILENAME<gh_stars>STARS\ncode<|endoftext|>
|
33 |
+
```
|
34 |
+
STARS can be one of: 0, 1-10, 10-100, 100-1000, 1000+
|
35 |
+
|
36 |
+
### 2. Commits 💾
|
37 |
+
The commits data is formatted as follows:
|
38 |
+
|
39 |
+
```
|
40 |
+
<commit_before>code<commit_msg>text<commit_after>code<|endoftext|>
|
41 |
+
```
|
42 |
+
|
43 |
+
### 3. Jupyter Notebooks 📓
|
44 |
+
The model is trained on Jupyter notebooks as Python scripts and structured formats like:
|
45 |
+
|
46 |
+
```
|
47 |
+
<start_jupyter><jupyter_text>text<jupyter_code>code<jupyter_output>output<jupyter_text>
|
48 |
+
```
|
49 |
+
|
50 |
+
### 4. Issues 🐛
|
51 |
+
We also trained on GitHub issues using the following formatting:
|
52 |
+
```
|
53 |
+
<issue_start><issue_comment>text<issue_comment>...<issue_closed>
|
54 |
+
```
|
55 |
+
|
56 |
+
### 5. Fill-in-the-middle 🧩
|
57 |
+
Fill in the middle requires rearranging the model inputs. The playground handles this for you - all you need is to specify where to fill:
|
58 |
+
```
|
59 |
+
code before<FILL_HERE>code after
|
60 |
+
```
|
61 |
+
"""
|
62 |
+
|
63 |
+
theme = gr.themes.Monochrome(
|
64 |
+
primary_hue="indigo",
|
65 |
+
secondary_hue="blue",
|
66 |
+
neutral_hue="slate",
|
67 |
+
radius_size=gr.themes.sizes.radius_sm,
|
68 |
+
font=[
|
69 |
+
gr.themes.GoogleFont("Open Sans"),
|
70 |
+
"ui-sans-serif",
|
71 |
+
"system-ui",
|
72 |
+
"sans-serif",
|
73 |
+
],
|
74 |
+
)
|
75 |
+
|
76 |
+
client = Client(
|
77 |
+
API_URL,
|
78 |
+
headers={"Authorization": f"Bearer {HF_TOKEN}"},
|
79 |
+
)
|
80 |
+
client_base = Client(
|
81 |
+
API_URL_BASE, headers={"Authorization": f"Bearer {HF_TOKEN}"},
|
82 |
+
)
|
83 |
+
client_plus = Client(
|
84 |
+
API_URL_PLUS, headers={"Authorization": f"Bearer {HF_TOKEN}"},
|
85 |
+
)
|
86 |
+
|
87 |
+
def generate(
|
88 |
+
prompt, temperature=0.9, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0, version="StarCoder",
|
89 |
+
):
|
90 |
+
|
91 |
+
temperature = float(temperature)
|
92 |
+
if temperature < 1e-2:
|
93 |
+
temperature = 1e-2
|
94 |
+
top_p = float(top_p)
|
95 |
+
fim_mode = False
|
96 |
+
|
97 |
+
generate_kwargs = dict(
|
98 |
+
temperature=temperature,
|
99 |
+
max_new_tokens=max_new_tokens,
|
100 |
+
top_p=top_p,
|
101 |
+
repetition_penalty=repetition_penalty,
|
102 |
+
do_sample=True,
|
103 |
+
seed=42,
|
104 |
+
)
|
105 |
+
|
106 |
+
if FIM_INDICATOR in prompt:
|
107 |
+
fim_mode = True
|
108 |
+
try:
|
109 |
+
prefix, suffix = prompt.split(FIM_INDICATOR)
|
110 |
+
except:
|
111 |
+
raise ValueError(f"Only one {FIM_INDICATOR} allowed in prompt!")
|
112 |
+
prompt = f"{FIM_PREFIX}{prefix}{FIM_SUFFIX}{suffix}{FIM_MIDDLE}"
|
113 |
+
|
114 |
+
if version == "StarCoder":
|
115 |
+
stream = client.generate_stream(prompt, **generate_kwargs)
|
116 |
+
elif version == "StarCoderPlus":
|
117 |
+
stream = client_plus.generate_stream(prompt, **generate_kwargs)
|
118 |
+
else:
|
119 |
+
stream = client_base.generate_stream(prompt, **generate_kwargs)
|
120 |
+
|
121 |
+
if fim_mode:
|
122 |
+
output = prefix
|
123 |
+
else:
|
124 |
+
output = prompt
|
125 |
+
|
126 |
+
previous_token = ""
|
127 |
+
for response in stream:
|
128 |
+
if response.token.text == "<|endoftext|>":
|
129 |
+
if fim_mode:
|
130 |
+
output += suffix
|
131 |
+
else:
|
132 |
+
return output
|
133 |
+
else:
|
134 |
+
output += response.token.text
|
135 |
+
previous_token = response.token.text
|
136 |
+
yield output
|
137 |
+
return output
|
138 |
+
|
139 |
+
|
140 |
+
examples = [
|
141 |
+
"X_train, y_train, X_test, y_test = train_test_split(X, y, test_size=0.1)\n\n# Train a logistic regression model, predict the labels on the test set and compute the accuracy score",
|
142 |
+
"// Returns every other value in the array as a new array.\nfunction everyOther(arr) {",
|
143 |
+
"Poor English: She no went to the market. Corrected English:",
|
144 |
+
"def alternating(list1, list2):\n results = []\n for i in range(min(len(list1), len(list2))):\n results.append(list1[i])\n results.append(list2[i])\n if len(list1) > len(list2):\n <FILL_HERE>\n else:\n results.extend(list2[i+1:])\n return results",
|
145 |
+
]
|
146 |
+
|
147 |
+
|
148 |
+
def process_example(args):
|
149 |
+
for x in generate(args):
|
150 |
+
pass
|
151 |
+
return x
|
152 |
+
|
153 |
+
|
154 |
+
css = ".generating {visibility: hidden}"
|
155 |
+
|
156 |
+
monospace_css = """
|
157 |
+
#q-input textarea {
|
158 |
+
font-family: monospace, 'Consolas', Courier, monospace;
|
159 |
+
}
|
160 |
+
"""
|
161 |
+
|
162 |
+
|
163 |
+
css += monospace_css + ".gradio-container {color: black}"
|
164 |
+
|
165 |
+
|
166 |
+
description = """
|
167 |
+
<div style="text-align: center;">
|
168 |
+
<h1> ⭐ StarCoder <span style='color: #e6b800;'>Models</span> Playground</h1>
|
169 |
+
</div>
|
170 |
+
<div style="text-align: left;">
|
171 |
+
<p>This is a demo to generate text and code with the following StarCoder models:</p>
|
172 |
+
<ul>
|
173 |
+
<li><a href="https://huggingface.co/bigcode/starcoderplus" style='color: #e6b800;'>StarCoderPlus</a>: A finetuned version of StarCoderBase on English web data, making it strong in both English text and code generation.</li>
|
174 |
+
<li><a href="https://huggingface.co/bigcode/starcoderbase" style='color: #e6b800;'>StarCoderBase</a>: A code generation model trained on 80+ programming languages, providing broad language coverage for code generation tasks.</li>
|
175 |
+
<li><a href="https://huggingface.co/bigcode/starcoder" style='color: #e6b800;'>StarCoder</a>: A finetuned version of StarCoderBase specifically focused on Python, while also maintaining strong performance on other programming languages.</li>
|
176 |
+
</ul>
|
177 |
+
<p><b>Please note:</b> These models are not designed for instruction purposes. If you're looking for instruction or want to chat with a fine-tuned model, you can visit the <a href="https://huggingface.co/spaces/HuggingFaceH4/starchat-playground">StarChat Playground</a>.</p>
|
178 |
+
</div>
|
179 |
+
"""
|
180 |
+
disclaimer = """⚠️<b>Any use or sharing of this demo constitues your acceptance of the BigCode [OpenRAIL-M](https://huggingface.co/spaces/bigcode/bigcode-model-license-agreement) License Agreement and the use restrictions included within.</b>\
|
181 |
+
<br>**Intended Use**: this app and its [supporting model](https://huggingface.co/bigcode) are provided for demonstration purposes; not to serve as replacement for human expertise. For more details on the model's limitations in terms of factuality and biases, see the [model card.](hf.co/bigcode)"""
|
182 |
+
|
183 |
+
with gr.Blocks(theme=theme, analytics_enabled=False, css=css) as demo:
|
184 |
+
with gr.Column():
|
185 |
+
gr.Markdown(description)
|
186 |
+
with gr.Row():
|
187 |
+
version = gr.Dropdown(
|
188 |
+
["StarCoderPlus", "StarCoderBase", "StarCoder"],
|
189 |
+
value="StarCoder",
|
190 |
+
label="Model",
|
191 |
+
info="Choose a model from the list",
|
192 |
+
)
|
193 |
+
with gr.Row():
|
194 |
+
with gr.Column():
|
195 |
+
instruction = gr.Textbox(
|
196 |
+
placeholder="Enter your code here",
|
197 |
+
lines=5,
|
198 |
+
label="Input",
|
199 |
+
elem_id="q-input",
|
200 |
+
)
|
201 |
+
submit = gr.Button("Generate", variant="primary")
|
202 |
+
output = gr.Code(elem_id="q-output", lines=30, label="Output")
|
203 |
+
with gr.Row():
|
204 |
+
with gr.Column():
|
205 |
+
with gr.Accordion("Advanced settings", open=False):
|
206 |
+
with gr.Row():
|
207 |
+
column_1, column_2 = gr.Column(), gr.Column()
|
208 |
+
with column_1:
|
209 |
+
temperature = gr.Slider(
|
210 |
+
label="Temperature",
|
211 |
+
value=0.2,
|
212 |
+
minimum=0.0,
|
213 |
+
maximum=1.0,
|
214 |
+
step=0.05,
|
215 |
+
interactive=True,
|
216 |
+
info="Higher values produce more diverse outputs",
|
217 |
+
)
|
218 |
+
max_new_tokens = gr.Slider(
|
219 |
+
label="Max new tokens",
|
220 |
+
value=256,
|
221 |
+
minimum=0,
|
222 |
+
maximum=8192,
|
223 |
+
step=64,
|
224 |
+
interactive=True,
|
225 |
+
info="The maximum numbers of new tokens",
|
226 |
+
)
|
227 |
+
with column_2:
|
228 |
+
top_p = gr.Slider(
|
229 |
+
label="Top-p (nucleus sampling)",
|
230 |
+
value=0.90,
|
231 |
+
minimum=0.0,
|
232 |
+
maximum=1,
|
233 |
+
step=0.05,
|
234 |
+
interactive=True,
|
235 |
+
info="Higher values sample more low-probability tokens",
|
236 |
+
)
|
237 |
+
repetition_penalty = gr.Slider(
|
238 |
+
label="Repetition penalty",
|
239 |
+
value=1.2,
|
240 |
+
minimum=1.0,
|
241 |
+
maximum=2.0,
|
242 |
+
step=0.05,
|
243 |
+
interactive=True,
|
244 |
+
info="Penalize repeated tokens",
|
245 |
+
)
|
246 |
+
|
247 |
+
gr.Markdown(disclaimer)
|
248 |
+
with gr.Group(elem_id="share-btn-container"):
|
249 |
+
share_button = gr.Button(
|
250 |
+
"Share to community", elem_id="share-btn", visible=True
|
251 |
+
)
|
252 |
+
gr.Examples(
|
253 |
+
examples=examples,
|
254 |
+
inputs=[instruction],
|
255 |
+
cache_examples=False,
|
256 |
+
fn=process_example,
|
257 |
+
outputs=[output],
|
258 |
+
)
|
259 |
+
gr.Markdown(FORMATS)
|
260 |
+
|
261 |
+
submit.click(
|
262 |
+
generate,
|
263 |
+
inputs=[instruction, temperature, max_new_tokens, top_p, repetition_penalty, version],
|
264 |
+
outputs=[output],
|
265 |
+
)
|
266 |
+
demo.queue(concurrency_count=16).launch(debug=True)
|