File size: 21,002 Bytes
eead5d8 48e003d caf1faa 48e003d caf1faa eead5d8 a5686cb ff42e3f f0fc5f8 71ab0a8 5f9881c f842a0e 4b4bf28 48e003d d26538b 887905a 4b4bf28 f0fc5f8 f842a0e f0fc5f8 6d2199d f0fc5f8 91c4196 5f9881c f0fc5f8 5f9881c d14568c 48e003d 5f9881c 48e003d 5f9881c caf1faa 48e003d d14568c f0fc5f8 ff42e3f 6d2199d ff42e3f f0fc5f8 abfa81d ff42e3f 46e3999 6d2199d f0fc5f8 7498c33 99e2b1f 6d2199d 91c4196 6d2199d 91c4196 6d2199d d4c1a74 6d2199d 91c4196 6d2199d a4595fc ff42e3f c974ee5 f0fc5f8 aa37f44 eee8932 48e003d f0fc5f8 d4c1a74 5f9881c c974ee5 48e003d 887905a 5f9881c 91f77da 3d561c7 91f77da 48e003d 5f9881c 48e003d 3d561c7 48e003d d4c1a74 5f9881c caf1faa 5f9881c 48e003d d14568c 3d561c7 48e003d 24f8d00 48e003d d14568c 48e003d d14568c 48e003d 34de7db 48e003d caf1faa d14568c 48e003d 24f8d00 48e003d 435c75a 24f8d00 63d1de4 24f8d00 4b4bf28 24f8d00 ee3f645 24f8d00 435c75a 24f8d00 d14568c 4b4bf28 d14568c 4b4bf28 7fa8087 d14568c 4b4bf28 d14568c 4b4bf28 d14568c 4b4bf28 d14568c 46e3999 a5686cb 91c4196 6d2199d 91c4196 6d2199d 91c4196 12574b1 dc1d7e6 fdf1622 91c4196 6d2199d 91c4196 6d2199d 91c4196 caf1faa f0fc5f8 ff42e3f 787d3cb c974ee5 787d3cb 38ed905 787d3cb f0fc5f8 38ed905 787d3cb 5f9881c 787d3cb f0fc5f8 c974ee5 3c9e1e2 868be0d f0fc5f8 5f9881c f0fc5f8 fa9f031 f0fc5f8 5f9881c 48e003d 38ed905 12f47f8 27a9af5 48e003d a3bf481 c974ee5 3d561c7 a3bf481 887905a 48e003d fa9f031 3c9e1e2 5f9881c 3c9e1e2 5f9881c 8edfef8 5f9881c 3c9e1e2 5f9881c caf1faa 3c9e1e2 d14568c 3c9e1e2 887905a 5f9881c 3c9e1e2 887905a 5f9881c 3c9e1e2 5f9881c 3c9e1e2 f0fc5f8 d14568c f0fc5f8 4b4bf28 48e003d caf1faa 48e003d caf1faa 48e003d caf1faa 48e003d caf1faa 48e003d caf1faa 48e003d caf1faa 887905a f0fc5f8 887905a 4b4bf28 48e003d 4b4bf28 d14568c 4b4bf28 d14568c 4b4bf28 887905a b6bb4d7 d730458 e77b244 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 |
from climateqa.engine.embeddings import get_embeddings_function
embeddings_function = get_embeddings_function()
from climateqa.knowledge.openalex import OpenAlex
from sentence_transformers import CrossEncoder
# reranker = CrossEncoder("mixedbread-ai/mxbai-rerank-xsmall-v1")
oa = OpenAlex()
import gradio as gr
import pandas as pd
import numpy as np
import os
import time
import re
import json
from gradio import ChatMessage
# from gradio_modal import Modal
from io import BytesIO
import base64
from datetime import datetime
from azure.storage.fileshare import ShareServiceClient
from utils import create_user_id
# ClimateQ&A imports
from climateqa.engine.llm import get_llm
from climateqa.engine.vectorstore import get_pinecone_vectorstore
# from climateqa.knowledge.retriever import ClimateQARetriever
from climateqa.engine.reranker import get_reranker
from climateqa.engine.embeddings import get_embeddings_function
from climateqa.engine.chains.prompts import audience_prompts
from climateqa.sample_questions import QUESTIONS
from climateqa.constants import POSSIBLE_REPORTS
from climateqa.utils import get_image_from_azure_blob_storage
from climateqa.engine.keywords import make_keywords_chain
# from climateqa.engine.chains.answer_rag import make_rag_papers_chain
from climateqa.engine.graph import make_graph_agent,display_graph
from front.utils import make_html_source, make_html_figure_sources,parse_output_llm_with_sources,serialize_docs,make_toolbox
# Load environment variables in local mode
try:
from dotenv import load_dotenv
load_dotenv()
except Exception as e:
pass
# Set up Gradio Theme
theme = gr.themes.Base(
primary_hue="blue",
secondary_hue="red",
font=[gr.themes.GoogleFont("Poppins"), "ui-sans-serif", "system-ui", "sans-serif"],
)
init_prompt = ""
system_template = {
"role": "system",
"content": init_prompt,
}
account_key = os.environ["BLOB_ACCOUNT_KEY"]
if len(account_key) == 86:
account_key += "=="
credential = {
"account_key": account_key,
"account_name": os.environ["BLOB_ACCOUNT_NAME"],
}
account_url = os.environ["BLOB_ACCOUNT_URL"]
file_share_name = "climateqa"
service = ShareServiceClient(account_url=account_url, credential=credential)
share_client = service.get_share_client(file_share_name)
user_id = create_user_id()
# Create vectorstore and retriever
vectorstore = get_pinecone_vectorstore(embeddings_function)
llm = get_llm(provider="openai",max_tokens = 1024,temperature = 0.0)
reranker = get_reranker("large")
agent = make_graph_agent(llm,vectorstore,reranker)
async def chat(query,history,audience,sources,reports):
"""taking a query and a message history, use a pipeline (reformulation, retriever, answering) to yield a tuple of:
(messages in gradio format, messages in langchain format, source documents)"""
date_now = datetime.now().strftime("%Y-%m-%d %H:%M:%S")
print(f">> NEW QUESTION ({date_now}) : {query}")
if audience == "Children":
audience_prompt = audience_prompts["children"]
elif audience == "General public":
audience_prompt = audience_prompts["general"]
elif audience == "Experts":
audience_prompt = audience_prompts["experts"]
else:
audience_prompt = audience_prompts["experts"]
# Prepare default values
if len(sources) == 0:
sources = ["IPCC"]
# if len(reports) == 0: # TODO
reports = []
inputs = {"user_input": query,"audience": audience_prompt,"sources":sources}
result = agent.astream_events(inputs,version = "v1")
# path_reformulation = "/logs/reformulation/final_output"
# path_keywords = "/logs/keywords/final_output"
# path_retriever = "/logs/find_documents/final_output"
# path_answer = "/logs/answer/streamed_output_str/-"
docs = []
docs_html = ""
output_query = ""
output_language = ""
output_keywords = ""
gallery = []
start_streaming = False
figures = '<div class="figures-container"> <p> Go to the "Figures" tab at the top of the page to see full size images </p> </div>'
steps_display = {
"categorize_intent":("ποΈ Analyzing user message",True),
"transform_query":("ποΈ Thinking step by step to answer the question",True),
"retrieve_documents":("ποΈ Searching in the knowledge base",False),
}
used_documents = []
answer_message_content = ""
try:
async for event in result:
if "langgraph_node" in event["metadata"]:
node = event["metadata"]["langgraph_node"]
if event["event"] == "on_chain_end" and event["name"] == "retrieve_documents" :# when documents are retrieved
try:
docs = event["data"]["output"]["documents"]
docs_html = []
textual_docs = [d for d in docs if d.metadata["chunk_type"] == "text"]
for i, d in enumerate(textual_docs, 1):
if d.metadata["chunk_type"] == "text":
docs_html.append(make_html_source(d, i))
used_documents = used_documents + [f"{d.metadata['short_name']} - {d.metadata['name']}" for d in docs]
history[-1].content = "Adding sources :\n\n - " + "\n - ".join(np.unique(used_documents))
docs_html = "".join(docs_html)
except Exception as e:
print(f"Error getting documents: {e}")
print(event)
elif event["name"] in steps_display.keys() and event["event"] == "on_chain_start": #display steps
event_description,display_output = steps_display[node]
if not hasattr(history[-1], 'metadata') or history[-1].metadata["title"] != event_description: # if a new step begins
history.append(ChatMessage(role="assistant", content = "", metadata={'title' :event_description}))
elif event["name"] != "transform_query" and event["event"] == "on_chat_model_stream" and node in ["answer_rag", "answer_search","answer_chitchat"]:# if streaming answer
if start_streaming == False:
start_streaming = True
history.append(ChatMessage(role="assistant", content = ""))
answer_message_content += event["data"]["chunk"].content
answer_message_content = parse_output_llm_with_sources(answer_message_content)
history[-1] = ChatMessage(role="assistant", content = answer_message_content)
# history.append(ChatMessage(role="assistant", content = new_message_content))
if event["name"] == "transform_query" and event["event"] =="on_chain_end":
if hasattr(history[-1],"content"):
history[-1].content += "Decompose question into sub-questions: \n\n - " + "\n - ".join([q["question"] for q in event["data"]["output"]["remaining_questions"]])
if event["name"] == "categorize_intent" and event["event"] == "on_chain_start":
print("X")
yield history,docs_html,output_query,output_language,gallery, figures #,output_query,output_keywords
except Exception as e:
print(event, "has failed")
raise gr.Error(f"{e}")
try:
# Log answer on Azure Blob Storage
if os.getenv("GRADIO_ENV") != "local":
timestamp = str(datetime.now().timestamp())
file = timestamp + ".json"
prompt = history[1]["content"]
logs = {
"user_id": str(user_id),
"prompt": prompt,
"query": prompt,
"question":output_query,
"sources":sources,
"docs":serialize_docs(docs),
"answer": history[-1].content,
"time": timestamp,
}
log_on_azure(file, logs, share_client)
except Exception as e:
print(f"Error logging on Azure Blob Storage: {e}")
raise gr.Error(f"ClimateQ&A Error: {str(e)[:100]} - The error has been noted, try another question and if the error remains, you can contact us :)")
# image_dict = {}
# for i,doc in enumerate(docs):
# if doc.metadata["chunk_type"] == "image":
# try:
# key = f"Image {i+1}"
# image_path = doc.metadata["image_path"].split("documents/")[1]
# img = get_image_from_azure_blob_storage(image_path)
# # Convert the image to a byte buffer
# buffered = BytesIO()
# img.save(buffered, format="PNG")
# img_str = base64.b64encode(buffered.getvalue()).decode()
# # Embedding the base64 string in Markdown
# markdown_image = f"![Alt text](data:image/png;base64,{img_str})"
# image_dict[key] = {"img":img,"md":markdown_image,"short_name": doc.metadata["short_name"],"figure_code":doc.metadata["figure_code"],"caption":doc.page_content,"key":key,"figure_code":doc.metadata["figure_code"], "img_str" : img_str}
# except Exception as e:
# print(f"Skipped adding image {i} because of {e}")
# if len(image_dict) > 0:
# gallery = [x["img"] for x in list(image_dict.values())]
# img = list(image_dict.values())[0]
# img_md = img["md"]
# img_caption = img["caption"]
# img_code = img["figure_code"]
# if img_code != "N/A":
# img_name = f"{img['key']} - {img['figure_code']}"
# else:
# img_name = f"{img['key']}"
# history.append(ChatMessage(role="assistant", content = f"\n\n{img_md}\n<p class='chatbot-caption'><b>{img_name}</b> - {img_caption}</p>"))
docs_figures = [d for d in docs if d.metadata["chunk_type"] == "image"]
for i, doc in enumerate(docs_figures):
if doc.metadata["chunk_type"] == "image":
try:
key = f"Image {i+1}"
image_path = doc.metadata["image_path"].split("documents/")[1]
img = get_image_from_azure_blob_storage(image_path)
# Convert the image to a byte buffer
buffered = BytesIO()
img.save(buffered, format="PNG")
img_str = base64.b64encode(buffered.getvalue()).decode()
figures = figures + make_html_figure_sources(doc, i, img_str)
gallery.append(img)
except Exception as e:
print(f"Skipped adding image {i} because of {e}")
yield history,docs_html,output_query,output_language,gallery, figures#,output_query,output_keywords
def save_feedback(feed: str, user_id):
if len(feed) > 1:
timestamp = str(datetime.now().timestamp())
file = user_id + timestamp + ".json"
logs = {
"user_id": user_id,
"feedback": feed,
"time": timestamp,
}
log_on_azure(file, logs, share_client)
return "Feedback submitted, thank you!"
def log_on_azure(file, logs, share_client):
logs = json.dumps(logs)
file_client = share_client.get_file_client(file)
file_client.upload_file(logs)
def generate_keywords(query):
chain = make_keywords_chain(llm)
keywords = chain.invoke(query)
keywords = " AND ".join(keywords["keywords"])
return keywords
papers_cols_widths = {
"doc":50,
"id":100,
"title":300,
"doi":100,
"publication_year":100,
"abstract":500,
"rerank_score":100,
"is_oa":50,
}
papers_cols = list(papers_cols_widths.keys())
papers_cols_widths = list(papers_cols_widths.values())
# --------------------------------------------------------------------
# Gradio
# --------------------------------------------------------------------
init_prompt = """
Hello, I am ClimateQ&A, a conversational assistant designed to help you understand climate change and biodiversity loss. I will answer your questions by **sifting through the IPCC and IPBES scientific reports**.
β How to use
- **Language**: You can ask me your questions in any language.
- **Audience**: You can specify your audience (children, general public, experts) to get a more adapted answer.
- **Sources**: You can choose to search in the IPCC or IPBES reports, or both.
β οΈ Limitations
*Please note that the AI is not perfect and may sometimes give irrelevant answers. If you are not satisfied with the answer, please ask a more specific question or report your feedback to help us improve the system.*
What do you want to learn ?
"""
def vote(data: gr.LikeData):
if data.liked:
print(data.value)
else:
print(data)
with gr.Blocks(title="Climate Q&A", css_paths=os.getcwd()+ "/style.css", theme=theme,elem_id = "main-component") as demo:
with gr.Tab("ClimateQ&A"):
with gr.Row(elem_id="chatbot-row"):
with gr.Column(scale=2):
chatbot = gr.Chatbot(
value = [ChatMessage(role="assistant", content=init_prompt)],
type = "messages",
show_copy_button=True,
show_label = False,
elem_id="chatbot",
layout = "panel",
avatar_images = (None,"https://i.ibb.co/YNyd5W2/logo4.png"),
max_height="80vh",
height="100vh"
)
# bot.like(vote,None,None)
with gr.Row(elem_id = "input-message"):
textbox=gr.Textbox(placeholder="Ask me anything here!",show_label=False,scale=7,lines = 1,interactive = True,elem_id="input-textbox")
with gr.Column(scale=1, variant="panel",elem_id = "right-panel"):
with gr.Tabs() as tabs:
with gr.TabItem("Examples",elem_id = "tab-examples",id = 0):
examples_hidden = gr.Textbox(visible = False)
first_key = list(QUESTIONS.keys())[0]
dropdown_samples = gr.Dropdown(QUESTIONS.keys(),value = first_key,interactive = True,show_label = True,label = "Select a category of sample questions",elem_id = "dropdown-samples")
samples = []
for i,key in enumerate(QUESTIONS.keys()):
examples_visible = True if i == 0 else False
with gr.Row(visible = examples_visible) as group_examples:
examples_questions = gr.Examples(
QUESTIONS[key],
[examples_hidden],
examples_per_page=8,
run_on_click=False,
elem_id=f"examples{i}",
api_name=f"examples{i}",
# label = "Click on the example question or enter your own",
# cache_examples=True,
)
samples.append(group_examples)
with gr.Tab("Sources",elem_id = "tab-citations",id = 1):
sources_textbox = gr.HTML(show_label=False, elem_id="sources-textbox")
docs_textbox = gr.State("")
# with Modal(visible = False) as config_modal:
with gr.Tab("Configuration",elem_id = "tab-config",id = 2):
gr.Markdown("Reminder: You can talk in any language, ClimateQ&A is multi-lingual!")
dropdown_sources = gr.CheckboxGroup(
["IPCC", "IPBES","IPOS"],
label="Select source",
value=["IPCC"],
interactive=True,
)
dropdown_reports = gr.Dropdown(
POSSIBLE_REPORTS,
label="Or select specific reports",
multiselect=True,
value=None,
interactive=True,
)
dropdown_audience = gr.Dropdown(
["Children","General public","Experts"],
label="Select audience",
value="Experts",
interactive=True,
)
output_query = gr.Textbox(label="Query used for retrieval",show_label = True,elem_id = "reformulated-query",lines = 2,interactive = False)
output_language = gr.Textbox(label="Language",show_label = True,elem_id = "language",lines = 1,interactive = False)
with gr.Tab("Figures",elem_id = "tab-figures",id = 3):
figures_cards = gr.HTML(show_label=False, elem_id="sources-figures")
#---------------------------------------------------------------------------------------
# OTHER TABS
#---------------------------------------------------------------------------------------
with gr.Tab("Figures",elem_id = "tab-images",elem_classes = "max-height other-tabs"):
gallery_component = gr.Gallery()
# with gr.Tab("Papers (beta)",elem_id = "tab-papers",elem_classes = "max-height other-tabs"):
# with gr.Row():
# with gr.Column(scale=1):
# query_papers = gr.Textbox(placeholder="Question",show_label=False,lines = 1,interactive = True,elem_id="query-papers")
# keywords_papers = gr.Textbox(placeholder="Keywords",show_label=False,lines = 1,interactive = True,elem_id="keywords-papers")
# after = gr.Slider(minimum=1950,maximum=2023,step=1,value=1960,label="Publication date",show_label=True,interactive=True,elem_id="date-papers")
# search_papers = gr.Button("Search",elem_id="search-papers",interactive=True)
# with gr.Column(scale=7):
# with gr.Tab("Summary",elem_id="papers-summary-tab"):
# papers_summary = gr.Markdown(visible=True,elem_id="papers-summary")
# with gr.Tab("Relevant papers",elem_id="papers-results-tab"):
# papers_dataframe = gr.Dataframe(visible=True,elem_id="papers-table",headers = papers_cols)
# with gr.Tab("Citations network",elem_id="papers-network-tab"):
# citations_network = gr.HTML(visible=True,elem_id="papers-citations-network")
with gr.Tab("About",elem_classes = "max-height other-tabs"):
with gr.Row():
with gr.Column(scale=1):
gr.Markdown("See more info at [https://climateqa.com](https://climateqa.com/docs/intro/)")
def start_chat(query,history):
# history = history + [(query,None)]
# history = [tuple(x) for x in history]
history = history + [ChatMessage(role="user", content=query)]
return (gr.update(interactive = False),gr.update(selected=1),history)
def finish_chat():
return (gr.update(interactive = True,value = ""))
(textbox
.submit(start_chat, [textbox,chatbot], [textbox,tabs,chatbot],queue = False,api_name = "start_chat_textbox")
.then(chat, [textbox,chatbot,dropdown_audience, dropdown_sources,dropdown_reports], [chatbot,sources_textbox,output_query,output_language,gallery_component,figures_cards],concurrency_limit = 8,api_name = "chat_textbox")
.then(finish_chat, None, [textbox],api_name = "finish_chat_textbox")
)
(examples_hidden
.change(start_chat, [examples_hidden,chatbot], [textbox,tabs,chatbot],queue = False,api_name = "start_chat_examples")
.then(chat, [examples_hidden,chatbot,dropdown_audience, dropdown_sources,dropdown_reports], [chatbot,sources_textbox,output_query,output_language,gallery_component, figures_cards],concurrency_limit = 8,api_name = "chat_examples")
.then(finish_chat, None, [textbox],api_name = "finish_chat_examples")
)
def change_sample_questions(key):
index = list(QUESTIONS.keys()).index(key)
visible_bools = [False] * len(samples)
visible_bools[index] = True
return [gr.update(visible=visible_bools[i]) for i in range(len(samples))]
dropdown_samples.change(change_sample_questions,dropdown_samples,samples)
demo.queue()
demo.launch(ssr_mode=False)
|