Spaces:
Sleeping
Sleeping
File size: 11,753 Bytes
c10c2de b22f922 02fdb50 766be8d 74164b2 73906d6 aef6b86 b22f922 dee4e98 9087846 b22f922 9087846 b22f922 9087846 b22f922 74164b2 b22f922 e82f42a e2121e2 9087846 b22f922 9087846 b22f922 e2121e2 b22f922 c10c2de 9087846 ccc2190 a86d269 c10c2de 9087846 c10c2de 9087846 e82f42a 9087846 e82f42a 9087846 e82f42a 9087846 c10c2de 9087846 02fdb50 b22f922 02fdb50 9087846 02fdb50 9087846 b22f922 32e427d 02fdb50 b22f922 32e427d b22f922 9087846 73906d6 b22f922 73906d6 02fdb50 b22f922 9087846 32e427d 02fdb50 9087846 32e427d 02fdb50 b22f922 9087846 ccc2190 32e427d 74164b2 e82f42a 74164b2 9087846 766be8d 73906d6 9087846 74164b2 b22f922 ccc2190 02fdb50 b22f922 ccc2190 f08712c b22f922 f08712c ccc2190 b22f922 f08712c 53e8d77 f08712c 3e126b2 f08712c ece0a10 f08712c ece0a10 f08712c ccc2190 b22f922 9087846 b22f922 ccc2190 b22f922 9087846 b22f922 02fdb50 b22f922 ccc2190 b22f922 9087846 b22f922 9087846 b22f922 02fdb50 b22f922 02fdb50 b22f922 9087846 32e427d b22f922 02fdb50 a6b89f7 02fdb50 b22f922 1b4d3b3 b22f922 9087846 ccc2190 9087846 f08712c 766be8d 9087846 ccc2190 9087846 e0f6e72 a86d269 9087846 e82f42a a86d269 b22f922 02fdb50 b22f922 9087846 73906d6 9087846 1b4d3b3 9087846 b22f922 74164b2 b22f922 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 |
from config import DEMO_TITLE, IS_SHARE, IS_DEBUG, CV_EXT, EXT_TXT
from config import CHEAP_API_BASE, CHEAP_API_KEY, CHEAP_MODEL
from config import STRONG_API_BASE, STRONG_API_KEY, STRONG_MODEL
from util import is_valid_url
from util import mylogger
from util import stream_together
# from util import checkAPI
from taskNonAI import extract_url, file_to_html, compile_pdf
from taskAI import TaskAI
## load data
from _data_test import mock_jd, mock_cv
## ui
import gradio as gr
## dependency
from pypandoc.pandoc_download import download_pandoc
## std
import os
import json
## debug
from icecream import ic
## platform specific
from _hf import HF, REPO_ID
logger = mylogger(__name__, "%(asctime)s:%(levelname)s:%(message)s")
info = logger.info
def init():
try:
os.system("shot-scraper install -b firefox")
download_pandoc()
except Exception as e:
ic(e)
if HF:
HF.restart_space(REPO_ID)
## Config Functions
def set_same_cheap_strong(
set_same: bool,
cheap_base,
cheap_key,
cheap_model,
strong_base,
strong_key,
strong_model,
):
# setup_zone = gr.Accordion("AI setup (OpenAI-compatible LLM API)", open=True)
if set_same:
return (
gr.Textbox(value=cheap_base, label="API Base", interactive=False),
gr.Textbox(
value=cheap_key, label="API key", type="password", interactive=False
),
gr.Textbox(value=cheap_model, label="Model ID", interactive=False),
# setup_zone,
)
else:
return (
gr.Textbox(value=strong_base, label="API Base", interactive=True),
gr.Textbox(
value=strong_key, label="API key", type="password", interactive=True
),
gr.Textbox(value=strong_model, label="Model ID", interactive=True),
# setup_zone,
)
## Main Functions
def prepare_input(jd_info, cv_file: str, cv_text):
if jd_info:
if is_valid_url(jd_info):
jd = extract_url(jd_info)
else:
jd = jd_info
else:
jd = mock_jd
if cv_text:
cv = cv_text
elif cv_file:
if any([cv_file.endswith(ext) for ext in EXT_TXT]):
with open(cv_file, "r", encoding="utf8") as f:
cv = f.read()
else:
cv = file_to_html(cv_file)
else:
cv = mock_cv
return jd, cv
def run_refine(api_base, api_key, api_model, jd_info, cv_text):
jd, cv = jd_info, cv_text
cheapAPI = {"base": api_base, "key": api_key, "model": api_model}
taskAI = TaskAI(cheapAPI, temperature=0.2, max_tokens=2048) # max_tokens=2048
info("API initialized")
gen = stream_together(
taskAI.jd_preprocess(input=jd),
taskAI.cv_preprocess(input=cv),
)
for result in gen:
yield result
def run_compose(api_base, api_key, api_model, min_jd, min_cv, is_debug):
strongAPI = {"base": api_base, "key": api_key, "model": api_model}
taskAI = TaskAI(strongAPI, is_debug=is_debug, temperature=0.6, max_tokens=4000)
info("Composing letter with CoT ...")
result = ""
for response in taskAI.compose_letter_CoT(jd=min_jd, resume=min_cv):
result += response.delta
yield result
def finalize_letter_txt(api_base, api_key, api_model, debug_CoT):
cheapAPI = {"base": api_base, "key": api_key, "model": api_model}
taskAI = TaskAI(cheapAPI, temperature=0.2, max_tokens=2048)
info("Finalizing letter ...")
result = ""
for response in taskAI.purify_letter(full_text=debug_CoT):
result += response.delta
yield result
def finalize_letter_pdf(
api_base, api_key, api_model, jd, cv, cover_letter_text, is_debug
):
cheapAPI = {"base": api_base, "key": api_key, "model": api_model}
taskAI = TaskAI(cheapAPI, temperature=0.1, max_tokens=100)
meta_data = next(taskAI.get_jobapp_meta(JD=jd, CV=cv))
pdf_context = json.loads(meta_data)
if is_debug:
ic(pdf_context)
pdf_context["letter_body"] = cover_letter_text
return meta_data, compile_pdf(
pdf_context,
tmpl_path="app/typst/template_letter.tmpl",
output_path=f"cover_letter_by_{pdf_context['applicantFullName']}_to_{pdf_context['companyFullName']}.pdf",
is_debug=is_debug,
)
with gr.Blocks(
title=DEMO_TITLE,
theme=gr.themes.Soft(
primary_hue="sky", secondary_hue="emerald", neutral_hue="stone"
),
) as app:
intro = f"""# {DEMO_TITLE}
> You provide job description and résumé, and I write Cover letter for you!
"""
gr.Markdown(intro)
with gr.Accordion("READ ME FIRST", open=False):
guide = gr.Markdown("""## Setup
> Expand "AI Setup" panel and setup API for 2 AI agents, **CheapAI** and **StrongAI**, before you start.
+ `API Key`: If you have no idea, go to https://beta.openai.com/account/api-keys
+ `Model ID`:
- **CheapAI**:
+ `gpt-3.5-turbo` any variant should be fine unless OpenAI make them lazier and dumber.
+ `Mistral-7B-Instruct-v0.1` works well
+ `gemma-7b-it` usually can't understand instructions properly and cause error.
- **StrongAI**:
+ `Mistral-7B-Instruct-v0.1` can do the job.
+ Models with small context window size like won't work, such as
- `gpt-3.5-turbo-0613`
- perhaps `gpt-4-0613`
Model choice can be a interesting topic involing quite a few experiments. Feel free to try any other models you like!
## Troubleshooting
- If your API is on localhost / local network, you may need to host this app on local network too.
- If your API is on some AI service platform, review if you have enough balance / credits / quota on the platform.
- If you are sure that you have set up the API correctly, but encounter an error along the way, try click the "Go!" button again.
- Try change AI provider / model.
- Report to GitHub issue if you believe it's a bug.
""")
with gr.Row():
with gr.Column(scale=1):
with gr.Accordion(
"AI setup (OpenAI-compatible LLM API)", open=False
) as setup_zone:
is_debug = gr.Checkbox(label="Debug Mode", value=IS_DEBUG)
gr.Markdown(
"**CheapAI**, an honest format converter and refiner, extracts essential info from job description and résumé, to reduce subsequent cost on Strong AI."
)
with gr.Group():
cheap_base = gr.Textbox(value=CHEAP_API_BASE, label="API Base")
cheap_key = gr.Textbox(
value=CHEAP_API_KEY, label="API key", type="password"
)
cheap_model = gr.Textbox(value=CHEAP_MODEL, label="Model ID")
gr.Markdown(
"---\n**StrongAI**, a thoughtful wordsmith, generates perfect cover letters to make both you and recruiters happy."
)
is_same_cheap_strong = gr.Checkbox(
label="the same as Cheap AI", value=False, container=False
)
with gr.Group():
strong_base = gr.Textbox(value=STRONG_API_BASE, label="API Base")
strong_key = gr.Textbox(
value=STRONG_API_KEY, label="API key", type="password"
)
strong_model = gr.Textbox(value=STRONG_MODEL, label="Model ID")
with gr.Group():
gr.Markdown("## Employer - Job Description")
jd_info = gr.Textbox(
label="Job Description",
placeholder="Paste as Full Text (recommmend) or URL",
lines=5,
max_lines=10,
)
with gr.Group():
gr.Markdown("## Applicant - CV / Résumé")
# with gr.Row():
cv_file = gr.File(
label="Allowed formats: " + " ".join(CV_EXT),
file_count="single",
file_types=CV_EXT,
type="filepath",
)
cv_text = gr.TextArea(
label="Or enter text",
placeholder="If attempting to both upload a file and enter text, only this text will be used.",
)
with gr.Column(scale=2):
gr.Markdown("## Result")
with gr.Accordion("Reformatting", open=True) as reformat_zone:
with gr.Row():
min_jd = gr.TextArea(label="Clean verion of Job Description")
min_cv = gr.TextArea(label="Clean verion of CV / Résumé")
with gr.Accordion("Expert Zone", open=False) as expert_zone:
debug_CoT = gr.Textbox(label="Chain of Thoughts")
debug_jobapp = gr.Textbox(label="Job application meta data")
cover_letter_text = gr.Textbox(label="Cover Letter")
cover_letter_pdf = gr.File(
label="Cover Letter PDF",
file_count="multiple",
type="filepath",
)
infer_btn = gr.Button("Go!", variant="primary")
is_same_cheap_strong.change(
fn=set_same_cheap_strong,
inputs=[
is_same_cheap_strong,
cheap_base,
cheap_key,
cheap_model,
strong_base,
strong_key,
strong_model,
],
outputs=[strong_base, strong_key, strong_model],
).then(
fn=lambda: gr.Accordion("AI setup (OpenAI-compatible LLM API)", open=True),
inputs=None,
outputs=[setup_zone],
)
infer_btn.click(
fn=set_same_cheap_strong,
inputs=[
is_same_cheap_strong,
cheap_base,
cheap_key,
cheap_model,
strong_base,
strong_key,
strong_model,
],
outputs=[strong_base, strong_key, strong_model],
).then(
fn=lambda: gr.Accordion("AI setup (OpenAI-compatible LLM API)", open=False),
inputs=None,
outputs=[setup_zone],
).success(
fn=prepare_input, inputs=[jd_info, cv_file, cv_text], outputs=[jd_info, cv_text]
).success(
fn=lambda: gr.Accordion("Reformatting", open=True),
inputs=None,
outputs=[reformat_zone],
).success(
fn=run_refine,
inputs=[cheap_base, cheap_key, cheap_model, jd_info, cv_text],
outputs=[min_jd, min_cv],
).success(
fn=lambda: [
gr.Accordion("Expert Zone", open=True),
gr.Accordion("Reformatting", open=False),
],
inputs=None,
outputs=[expert_zone, reformat_zone],
).success(
fn=run_compose,
inputs=[strong_base, strong_key, strong_model, min_jd, min_cv, is_debug],
outputs=[debug_CoT],
).success(
fn=lambda: gr.Accordion("Expert Zone", open=False),
inputs=None,
outputs=[expert_zone],
).success(
fn=finalize_letter_txt,
inputs=[cheap_base, cheap_key, cheap_model, debug_CoT],
outputs=[cover_letter_text],
).success(
fn=finalize_letter_pdf,
inputs=[
cheap_base,
cheap_key,
cheap_model,
jd_info,
cv_text,
cover_letter_text,
is_debug,
],
outputs=[debug_jobapp, cover_letter_pdf],
)
if __name__ == "__main__":
init()
app.queue(max_size=1, default_concurrency_limit=1).launch(
show_error=True, debug=True, share=IS_SHARE
)
|