File size: 14,026 Bytes
92b4e5c | 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 | """Contains useful utility functions."""
import json
import urllib.parse
from pathlib import Path
import requests
import tiktoken
from openai import AsyncOpenAI, OpenAI
from openai.types.beta.threads.run import Run
from virtual_lab.constants import (
DEFAULT_FINETUNING_EPOCHS,
MODEL_TO_INPUT_PRICE_PER_TOKEN,
MODEL_TO_OUTPUT_PRICE_PER_TOKEN,
FINETUNING_MODEL_TO_TRAINING_PRICE_PER_TOKEN,
PUBMED_TOOL_NAME,
)
from virtual_lab.prompts import format_references
def get_pubmed_central_article(
pmcid: str, abstract_only: bool = False
) -> tuple[str | None, list[str] | None]:
"""Gets the title and content (abstract or full text) of a PubMed Central article given a PMC ID.
Note: This only returns main text, ignoring tables, figures, and references.
:param pmcid: The PMC ID of the article.
:param abstract_only: Whether to return only the abstract instead of the full text.
:return: The title and content (abstract or full text of the article as a list of paragraphs)
or None if the article is not found.
"""
# Get article from PMC ID in JSON form
text_url = f"https://www.ncbi.nlm.nih.gov/research/bionlp/RESTful/pmcoa.cgi/BioC_JSON/PMC{pmcid}/unicode"
response = requests.get(text_url)
response.raise_for_status()
# Try to parse JSON
try:
article = response.json()
except json.JSONDecodeError:
return None, None
# Get document
document = article[0]["documents"][0]
# Get title
title = next(
passage["text"]
for passage in document["passages"]
if passage["infons"]["section_type"] == "TITLE"
)
# Get relevant passages
passages = [
passage
for passage in document["passages"]
if passage["infons"]["type"] in {"abstract", "paragraph"}
]
# Get abstract or full text of article (excluding references)
if abstract_only:
passages = [
passage
for passage in passages
if passage["infons"]["section_type"] in ["ABSTRACT"]
]
else:
passages = [
passage
for passage in passages
if passage["infons"]["section_type"]
in ["ABSTRACT", "INTRO", "RESULTS", "DISCUSS", "CONCL", "METHODS"]
]
# Get content
content = [passage["text"] for passage in passages]
return title, content
def run_pubmed_search(
query: str, num_articles: int = 3, abstract_only: bool = False
) -> str:
"""Runs a PubMed search, returning the full text of the top matching article.
:param query: The query to search PubMed with.
:param num_articles: The number of articles to search for.
:param abstract_only: Whether to return only the abstract instead of the full text.
:return: The full text of the top matching article.
"""
# Print search query
print(
f'Searching PubMed Central for {num_articles} articles ({'abstracts' if abstract_only else 'full text'}) with query: "{query}"'
)
# Perform PubMed Central search for query to get PMC ID
search_url = f"https://eutils.ncbi.nlm.nih.gov/entrez/eutils/esearch.fcgi?db=pmc&term={urllib.parse.quote_plus(query)}&retmax={2 * num_articles}&retmode=json&sort=relevance"
response = requests.get(search_url)
response.raise_for_status()
pmcids_found = response.json()["esearchresult"]["idlist"]
# Loop through top articles
texts = []
titles = []
pmcids = []
for pmcid in pmcids_found:
# Break if reached desired number of articles
if len(pmcids) >= num_articles:
break
title, content = get_pubmed_central_article(
pmcid=pmcid,
abstract_only=abstract_only,
)
if title is None:
continue
texts.append(f"PMCID = {pmcid}\n\nTitle = {title}\n\n{'\n\n'.join(content)}")
titles.append(title)
pmcids.append(pmcid)
# Print articles found
article_count = len(texts)
print(f"Found {article_count:,} articles on PubMed Central")
# Combine texts
if article_count == 0:
combined_text = f'No articles found on PubMed Central for the query "{query}".'
else:
combined_text = format_references(
references=tuple(texts),
reference_type="paper",
intro=f'Here are the top {article_count} articles on PubMed Central for the query "{query}":',
)
return combined_text
def run_tools(run: Run) -> list[dict[str, str]]:
"""Runs the tools in a required action.
:param run: The run to run tools for.
:return: A list of tool outputs.
"""
# Define the list to store tool outputs
tool_outputs = []
# Loop through each tool in the required action and run it
for tool in run.required_action.submit_tool_outputs.tool_calls:
if tool.function.name == PUBMED_TOOL_NAME:
# Extract the query from the tool arguments
args_dict = json.loads(tool.function.arguments)
# Run the tool and append the output to the list of tool outputs
tool_outputs.append(
{
"tool_call_id": tool.id,
"output": run_pubmed_search(**args_dict),
}
)
else:
raise ValueError(f"Unknown tool: {tool.function.name}")
return tool_outputs
def get_messages(client: OpenAI, thread_id: str) -> list[dict]:
"""Gets messages from a thread.
:param client: The OpenAI client.
:param thread_id: The ID of the thread to get messages from.
:return: A list of messages.
"""
# Set up
messages = []
last_message = None
params = {
"thread_id": thread_id,
"limit": 100,
"order": "asc",
}
# Get all messages from the thread page by page
while True:
# Set up params
if last_message is not None:
params["after"] = last_message["id"]
elif "after" in params:
del params["after"]
# Get messages
new_messages = [
message.to_dict() for message in client.beta.threads.messages.list(**params)
]
# Append new messages
messages += new_messages
# Break if no more messages
if len(new_messages) < params["limit"]:
break
# Get last message
last_message = messages[-1]
# Verify all message content is length 1
assert all(len(message["content"]) == 1 for message in messages)
return messages
async def async_get_messages(client: AsyncOpenAI, thread_id: str) -> list[dict]:
"""Gets messages from a thread.
:param client: The async OpenAI client.
:param thread_id: The ID of the thread to get messages from.
:return: A list of messages.
"""
# Set up
messages = []
last_message = None
params = {
"thread_id": thread_id,
"limit": 100,
"order": "asc",
}
# Get all messages from the thread page by page
while True:
# Set up params
if last_message is not None:
params["after"] = last_message["id"]
elif "after" in params:
del params["after"]
# Get messages
new_messages = [
message.to_dict()
async for message in client.beta.threads.messages.list(**params)
]
# Append new messages
messages += new_messages
# Break if no more messages
if len(new_messages) < params["limit"]:
break
# Get last message
last_message = messages[-1]
# Verify all message content is length 1
assert all(len(message["content"]) == 1 for message in messages)
return messages
def count_tokens(string: str, encoding_name: str = "cl100k_base") -> int:
"""Returns the number of tokens in a text string.
:param string: The text string to count tokens in.
:param encoding_name: The name of the encoding to use.
:return: The number of tokens in the text string.
"""
encoding = tiktoken.get_encoding(encoding_name)
num_tokens = len(encoding.encode(string))
return num_tokens
def update_token_counts(
token_counts: dict[str, int],
discussion: list[dict[str, str]],
response: str,
) -> None:
"""Updates the token counts (in place) with a discussion and response.
:param token_counts: The token counts to update.
:param discussion: The discussion to update the token counts with.
:param response: The response to update the token counts with.
"""
new_input_token_count = sum(count_tokens(turn["message"]) for turn in discussion)
new_output_token_count = count_tokens(response)
token_counts["input"] += new_input_token_count
token_counts["output"] += new_output_token_count
token_counts["max"] = max(
token_counts["max"], new_input_token_count + new_output_token_count
)
def count_discussion_tokens(
discussion: list[dict[str, str]],
) -> dict[str, int]:
"""Counts the number of tokens in a discussion.
:param discussion: The discussion to count tokens in.
:return: A dictionary of token counts.
"""
token_counts = {
"input": 0,
"output": 0,
"max": 0,
}
for index, turn in enumerate(discussion):
if turn["agent"] != "User":
update_token_counts(
token_counts=token_counts,
discussion=discussion[:index],
response=turn["message"],
)
return token_counts
def compute_token_cost(
model: str, input_token_count: int, output_token_count: int
) -> float:
"""Computes the token cost of a model given input and output token counts.
:param model: The name of the model.
:param input_token_count: The number of tokens in the input.
:param output_token_count: The number of tokens in the output.
:return: The token cost of the model.
"""
if (
model not in MODEL_TO_INPUT_PRICE_PER_TOKEN
or model not in MODEL_TO_OUTPUT_PRICE_PER_TOKEN
):
raise ValueError(f'Cost of model "{model}" not known')
return (
input_token_count * MODEL_TO_INPUT_PRICE_PER_TOKEN[model]
+ output_token_count * MODEL_TO_OUTPUT_PRICE_PER_TOKEN[model]
)
def print_cost_and_time(
token_counts: dict[str, int],
model: str,
elapsed_time: float,
) -> None:
# Print token counts
print(f"Input token count: {token_counts['input']:,}")
print(f"Output token count: {token_counts['output']:,}")
print(f"Tool token count: {token_counts['tool']:,}")
print(f"Max token length: {token_counts['max']:,}")
# Compute and print cost
try:
cost = compute_token_cost(
model=model,
input_token_count=token_counts["input"] + token_counts["tool"],
output_token_count=token_counts["output"],
)
print(f"Cost: ${cost:.2f}")
except ValueError as e:
print(f"Warning: {e}")
# Print time
print(f"Time: {int(elapsed_time // 60)}:{int(elapsed_time % 60):02d}")
def compute_finetuning_cost(
model: str, token_count: int, num_epochs: int = DEFAULT_FINETUNING_EPOCHS
) -> float:
"""Computes the cost of fine-tuning a model.
:param model: The model that will be finetuned.
:param token_count: The number of training tokens for finetuning.
:param num_epochs: Number of finetuning epochs.
:return: The cost of finetuning.
"""
if model not in FINETUNING_MODEL_TO_TRAINING_PRICE_PER_TOKEN:
raise ValueError(f'Cost of model "{model}" not known')
return (
token_count * FINETUNING_MODEL_TO_TRAINING_PRICE_PER_TOKEN[model] * num_epochs
)
def convert_messages_to_discussion(
messages: list[dict], assistant_id_to_title: dict[str, str]
) -> list[dict[str, str]]:
"""Converts OpenAI messages into discussion format (list of message dictionaries).
:param messages: The messages to convert.
:param assistant_id_to_title: A dictionary mapping assistant IDs to titles.
:return: The discussion format (list of message dictionaries).
"""
return [
{
"agent": (
assistant_id_to_title[message["assistant_id"]]
if message["assistant_id"] is not None
else "User"
),
"message": message["content"][0]["text"]["value"],
}
for message in messages
]
def get_summary(discussion: list[dict[str, str]]) -> str:
"""Get the summary from a discussion.
:param discussion: The discussion to extract the summary from.
:return: The summary.
"""
return discussion[-1]["message"]
def load_summaries(discussion_paths: list[Path]) -> tuple[str, ...]:
"""Load summaries from a list of discussion paths.
:param discussion_paths: The paths to the discussion JSON files. The summary is the last entry in the discussion.
:return: A tuple of summaries.
"""
summaries = []
for discussion_path in discussion_paths:
with open(discussion_path, "r") as file:
discussion = json.load(file)
summaries.append(get_summary(discussion))
return tuple(summaries)
def save_meeting(
save_dir: Path, save_name: str, discussion: list[dict[str, str]]
) -> None:
"""Save a meeting discussion to JSON and Markdown files.
:param save_dir: The directory to save the discussion.
:param save_name: The name of the discussion file that will be saved.
:param discussion: The discussion to save.
"""
# Create the save directory if it does not exist
save_dir.mkdir(parents=True, exist_ok=True)
# Save the discussion as JSON
with open(save_dir / f"{save_name}.json", "w") as f:
json.dump(discussion, f, indent=4)
# Save the discussion as Markdown
with open(save_dir / f"{save_name}.md", "w") as file:
for turn in discussion:
file.write(f"## {turn['agent']}\n\n{turn['message']}\n\n")
|