File size: 38,260 Bytes
b5ac9e4 42bc715 b5ac9e4 33b8666 b5ac9e4 2edae76 b5ac9e4 5928341 ba89be7 b5ac9e4 ba89be7 b5ac9e4 ba89be7 b5ac9e4 42bc715 b5ac9e4 dd34ece b5ac9e4 dd34ece b5ac9e4 2edae76 b5ac9e4 42bc715 b5ac9e4 2edae76 b5ac9e4 fa8f9de b5ac9e4 42bc715 b5ac9e4 fa8f9de b5ac9e4 42bc715 b5ac9e4 42bc715 b5ac9e4 42bc715 b5ac9e4 42bc715 d748a9d b5ac9e4 1a243e9 42bc715 2edae76 42bc715 2edae76 42bc715 d748a9d 42bc715 2edae76 1a243e9 b5ac9e4 fa8f9de b5ac9e4 1a243e9 b5ac9e4 b5ae53a b5ac9e4 2edae76 fa8f9de 1a243e9 b5ac9e4 42bc715 b5ac9e4 42bc715 d748a9d b5ac9e4 2edae76 b5ac9e4 2edae76 b5ac9e4 fa8f9de b5ac9e4 42bc715 fa8f9de b5ac9e4 d748a9d 5f059ed d748a9d b5ac9e4 d748a9d b5ac9e4 42bc715 d748a9d b5ac9e4 d748a9d b5ac9e4 42bc715 d748a9d 42bc715 b5ac9e4 d748a9d b5ac9e4 d748a9d 2edae76 b57b01a d748a9d fa8f9de d748a9d b5ac9e4 2edae76 fa8f9de d748a9d b5ac9e4 b57b01a fa8f9de b5ac9e4 d748a9d b5ac9e4 d748a9d 42bc715 d748a9d 42bc715 b5ac9e4 b57b01a b5ac9e4 fa8f9de b5ac9e4 5f059ed b5ac9e4 fa8f9de b5ac9e4 ba89be7 b5ac9e4 ba89be7 b5ac9e4 2edae76 5f059ed 2edae76 5928341 42bc715 b5ac9e4 42bc715 b5ac9e4 5f059ed b5ac9e4 5f059ed b5ac9e4 5f059ed b5ac9e4 5f059ed b5ac9e4 5f059ed b5ac9e4 5f059ed b5ac9e4 5f059ed b5ac9e4 5f059ed b5ac9e4 5f059ed b5ac9e4 5f059ed b5ac9e4 5f059ed b5ac9e4 5f059ed b5ac9e4 d748a9d b5ac9e4 5f059ed b5ac9e4 d748a9d b5ac9e4 5f059ed b5ac9e4 5f059ed b5ac9e4 e8e31b3 b5ac9e4 42bc715 5f059ed b5ac9e4 d748a9d b5ac9e4 d748a9d b5ac9e4 d748a9d b5ac9e4 d748a9d b5ac9e4 d748a9d b5ac9e4 d748a9d b5ac9e4 d748a9d b5ac9e4 d748a9d b5ac9e4 d748a9d b5ac9e4 d748a9d b5ac9e4 d748a9d 42bc715 5928341 42bc715 5928341 42bc715 d748a9d b5ac9e4 d748a9d b5ac9e4 d748a9d 5f059ed 713bacc b5ac9e4 d748a9d 2edae76 d748a9d b5ac9e4 5f059ed |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 |
import copy
import math
import os
import time
from threading import Thread
import uuid
import gradio as gr
import spaces
import torch
from docling.backend.pypdfium2_backend import PyPdfiumDocumentBackend
from docling.datamodel.pipeline_options import PdfPipelineOptions
from docling.document_converter import DocumentConverter, InputFormat, PdfFormatOption
from langchain.schema.document import Document
from langchain_chroma import Chroma
from langchain_community.embeddings import HuggingFaceBgeEmbeddings
from langchain_docling import DoclingLoader
from langchain_docling.loader import ExportType
from langchain_text_splitters import RecursiveCharacterTextSplitter
from transformers import AutoModelForCausalLM, AutoTokenizer, DynamicCache, TextIteratorStreamer, BitsAndBytesConfig #, Gemma3ForCausalLM
from transformers.models.llama.modeling_llama import rotate_half
import threading
import shutil
import time
from utils import (
calculate_tokens_suggest_compression_ratio,
repeat_kv,
update_retrieval_context,
)
# Initialize the model and tokenizer.
api_token = os.getenv("HUGGING_FACE_HUB_TOKEN")
model_name = "meta-llama/Llama-3.1-8B-Instruct"
# model_name = "google/gemma-3-27b-it"
tokenizer = AutoTokenizer.from_pretrained(model_name, token=api_token)
# quantization_config = BitsAndBytesConfig(load_in_8bit=True)
model = AutoModelForCausalLM.from_pretrained(model_name, token=api_token, torch_dtype=torch.float16)
# model = Gemma3ForCausalLM.from_pretrained(model_name, token=api_token, quantization_config=quantization_config, torch_dtype="auto")
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = model.eval()
model.to(device)
embedding_model = HuggingFaceBgeEmbeddings(
model_name="BAAI/bge-large-en-v1.5",
model_kwargs={"device": str(device)},
encode_kwargs={"normalize_embeddings": True},
query_instruction=""
)
# Create a chat template and split into prefix and suffix.
content_system = ""
content_user = "######"
user_template = [
{"role": "system", "content": content_system},
{"role": "user", "content": content_user}
]
user = tokenizer.apply_chat_template(user_template, add_generation_prompt=True, tokenize=False)
prefix, suffix = user.split(content_user)
sink_tokens = max(4, len(tokenizer.encode(prefix)))
# Default prompt content.
default_task_description = (
"Answer the question based on the provided context."
"Provide only the answer and no additional explanations."
)
default_few_shot = """Examples
context: Climate change is primarily driven by human activities such as burning fossil fuels, deforestation, and industrial processes, which release large amounts of greenhouse gases into the atmosphere, causing global temperatures to rise.
question: What are the main human activities contributing to climate change?
answer: The main human activities contributing to climate change include burning fossil fuels, deforestation, and various industrial processes that emit greenhouse gases.
context: The Renaissance was a cultural movement spanning roughly from the 14th to the 17th century, marked by renewed interest in classical learning and values, advancements in art, literature, and scientific inquiry, and significant cultural developments.
question: What characterized the Renaissance period?
answer: The Renaissance period was characterized by a revival of classical learning, significant advancements in art and literature, and notable developments in scientific inquiry and cultural values.
context: The theory of evolution by natural selection, proposed by Charles Darwin, explains how species adapt and evolve over generations based on the survival and reproduction of individuals best suited to their environment.
question: How does Darwin's theory of evolution explain the adaptation of species?
answer: Darwin's theory explains that species adapt and evolve through natural selection, where individuals best suited to their environment are more likely to survive and reproduce.
"""
CHROMA_DB_DIR = "./chroma_db"
CACHE_DIR = "./cache_dir"
EXPIRATION_SECONDS = 3600
def background_cleanup():
while True:
current_time = int(time.time())
# Clean Chroma collections
if os.path.exists(CHROMA_DB_DIR):
for dirname in os.listdir(CHROMA_DB_DIR):
parts = dirname.split("_")
if len(parts) >= 3 and parts[1].isdigit():
timestamp = int(parts[1])
if current_time - timestamp > EXPIRATION_SECONDS:
path = os.path.join(CHROMA_DB_DIR, dirname)
shutil.rmtree(path, ignore_errors=True)
print(f"[Cleanup] Deleted Chroma collection: {path}")
# Clean cache files
if os.path.exists(CACHE_DIR):
for filename in os.listdir(CACHE_DIR):
parts = filename.split("_")
if len(parts) >= 3 and parts[1].isdigit():
timestamp = int(parts[1])
if current_time - timestamp > EXPIRATION_SECONDS:
path = os.path.join(CACHE_DIR, filename)
os.remove(path)
print(f"[Cleanup] Deleted cache file: {path}")
time.sleep(600)
cleanup_thread = threading.Thread(target=background_cleanup, daemon=True)
cleanup_thread.start()
class FinchCache(DynamicCache):
def __init__(self) -> None:
super().__init__()
self.key_cache = []
self.value_cache = []
@staticmethod
def _rotate_half(x):
x1 = x[..., : x.shape[-1] // 2]
x2 = x[..., x.shape[-1] // 2 :]
return torch.cat((-x2, x1), dim=-1)
def _apply_key_rotary_pos_emb(self, key_states: torch.Tensor, cos: torch.Tensor, sin: torch.Tensor) -> torch.Tensor:
return (key_states * cos) + (self._rotate_half(key_states) * sin)
@staticmethod
def _rerotate_cos_sin(x, inv_freq, important_pos_batch):
B, H, L = important_pos_batch.shape
device = important_pos_batch.device
device_type = x.device.type
dtype = x.dtype
idx = torch.arange(0, L, device=device)
idx = idx.unsqueeze(0)
inv_freq = inv_freq[None, None, :, None].float().expand(B, H, -1, 1) # (B, H, M, 1)
idx = idx[:, None, :].float().expand(B, H, L) # (B, H, L)
delta_pos = idx - important_pos_batch
delta_pos = delta_pos.unsqueeze(2) # (B, H, 1, L)
device_type = device_type if isinstance(device_type, str) and device_type != "mps" else "cpu"
with torch.autocast(device_type=device_type, enabled=False):
freqs = delta_pos.float() * inv_freq.float()
freqs = freqs.transpose(2, 3)
emb = torch.cat((freqs, freqs), dim=-1)
cos = emb.cos().contiguous()
sin = emb.sin().contiguous()
return cos.to(dtype=dtype), sin.to(dtype=dtype)
@staticmethod
def gather_important_tokens(states, indices):
return torch.gather(states, 2, indices.unsqueeze(-1).expand(-1, -1, -1, states.size(3))).contiguous()
def compress_cache(self, layer_index, important_pos, inv_freq):
new_length = important_pos.size(2)
new_cos, new_sin = self._rerotate_cos_sin(self.key_cache[layer_index], inv_freq, important_pos)
gathered_keys = self.gather_important_tokens(self.key_cache[layer_index], important_pos).clone()
self.key_cache[layer_index] = self._apply_key_rotary_pos_emb(gathered_keys, new_cos, new_sin)
gathered_values = self.gather_important_tokens(self.value_cache[layer_index], important_pos).clone()
self.value_cache[layer_index] = gathered_values
self._seen_tokens = new_length
def save(self, path: str):
try:
os.makedirs(os.path.dirname(path), exist_ok=True)
torch.save(
{"key_cache": [k.cpu() for k in self.key_cache], "value_cache": [v.cpu() for v in self.value_cache]},
path,
)
except Exception as e:
print(f"Error occurred while saving: {e}")
@classmethod
def load(cls, path: str, device: str = "cpu") -> "FinchCache":
data = torch.load(path, map_location=device)
cache = cls()
cache.key_cache = [k.to(device) for k in data["key_cache"]]
cache.value_cache = [v.to(device) for v in data["value_cache"]]
cache._seen_tokens = cache.value_cache[0].size(2) if cache.value_cache else 0
return cache
def convert_to_markdown(file_objs, url, do_ocr, do_table_structure):
file_path = file_objs if file_objs is not None else url
pipeline_options = PdfPipelineOptions()
pipeline_options.do_ocr = do_ocr
pipeline_options.do_table_structure = do_table_structure
pdf_format_options = PdfFormatOption(
pipeline_options=pipeline_options,
backend=PyPdfiumDocumentBackend,
)
doc_converter = DocumentConverter(
allowed_formats=[InputFormat.PDF],
format_options={InputFormat.PDF: pdf_format_options}
)
loader = DoclingLoader(
file_path=file_path,
export_type=ExportType.MARKDOWN,
converter=doc_converter
)
try:
docs = loader.load()
return docs[0].page_content
except Exception as e:
raise RuntimeError(f"Failed to convert document to markdown: {e}")
def create_rag_index(collection_name, text_no_prefix):
text_splitter = RecursiveCharacterTextSplitter.from_huggingface_tokenizer(
tokenizer,
chunk_size=256,
chunk_overlap=0,
add_start_index=True,
strip_whitespace=True,
separators=["\n\n", "\n", ".", " ", ""],
)
docs = [Document(page_content=x) for x in text_splitter.split_text(text_no_prefix)]
vectorstore = Chroma.from_documents(collection_name=collection_name, persist_directory="./chroma_db", documents=docs, embedding=embedding_model)
return vectorstore
@spaces.GPU
def auto_convert(file_objs, url, do_ocr, do_table_structure):
# When a new file/URL is loaded, disable chat (compression not done)
chat_status = "Document not compressed yet. Please compress the document to enable chat."
if file_objs is None and (url is None or url.strip() == ""):
return (
gr.update(value=""),
"Number of tokens before compression: ",
gr.update(),
"Number of tokens after compression: ",
0,
gr.update(interactive=False),
False,
{},
chat_status
)
print("Converting to markdown")
try:
markdown = convert_to_markdown(file_objs, url, do_ocr, do_table_structure)
except RuntimeError as e:
return (
gr.update(value=f"{str(e)} Please try uploading another document format."),
"Number of tokens before compression: ",
gr.update(),
"Number of tokens after compression: ",
0,
gr.update(interactive=False),
False,
{},
chat_status
)
print("Done")
combined_text = prefix + markdown
print("Suggestioning Compression ratio")
token_count, suggestions, _ = calculate_tokens_suggest_compression_ratio(combined_text, tokenizer, model)
print("Done")
min_ratio = min(suggestions)
max_ratio = max(suggestions)
default_ratio = 4
retrieval_tokens = int(token_count / default_ratio)
token_count_str = f"Number of tokens before compression: {token_count}"
retrieval_str = f"Number of tokens after compression: {retrieval_tokens}"
slider_update = gr.update(value=default_ratio, minimum=min_ratio, maximum=max_ratio, step=1)
if combined_text.startswith(prefix):
rag_text = combined_text[len(prefix):]
else:
rag_text = combined_text
current_timestamp = int(time.time())
collection_name = f"default_{current_timestamp}_{uuid.uuid4().hex[:6]}"
rag_index = create_rag_index(collection_name, rag_text)
state = {"rag_index": collection_name}
print("Done")
return (
combined_text,
token_count_str,
slider_update,
retrieval_str,
token_count,
gr.update(interactive=True), # Enable compress button if conversion succeeds.
False,
state,
chat_status
)
def get_compressed_kv_cache(sink_tokens, step_size, target_token_size, context_ids, context_attention_mask, question_ids, question_attention_mask):
try:
device = model.device
dtype = model.dtype
sink_tokens = sink_tokens
num_chunks = step_size
context_ids = context_ids.to(device)
context_attention_mask = context_attention_mask.to(device)
question_ids = question_ids.to(device)
question_attention_mask = question_attention_mask.to(device)
question_len = question_ids.size(1)
total_len = context_ids.size(1)
max_context_tokens_allowed = model.config.max_position_embeddings - question_len
if total_len > max_context_tokens_allowed:
num_chunks = max(step_size, math.ceil(total_len / max_context_tokens_allowed))
if total_len <= sink_tokens or num_chunks == 1:
context_ids_list = [context_ids]
context_attention_mask_list = [context_attention_mask]
else:
remainder_len = total_len - sink_tokens
base = remainder_len // num_chunks
leftover = remainder_len % num_chunks
chunk_sizes = [sink_tokens + base]
for _ in range(num_chunks - 2):
chunk_sizes.append(base)
if num_chunks > 1:
chunk_sizes.append(base + leftover)
context_ids_list = []
context_attention_mask_list = []
offset = 0
for size in chunk_sizes:
end = offset + size
context_ids_list.append(context_ids[:, offset:end])
context_attention_mask_list.append(context_attention_mask[:, offset:end])
offset = end
len_rest = max(total_len - sink_tokens, 1)
compression_factor = len_rest // target_token_size
if compression_factor < 1:
compression_factor = 1
tokenized_doc_chunks = []
for ids_chunk, mask_chunk in zip(context_ids_list, context_attention_mask_list):
tokenized_doc_chunks.append({"input_ids": ids_chunk, "attention_mask": mask_chunk})
print("Number of chunks: ", len(tokenized_doc_chunks))
rotary_emb = model.model.rotary_emb.to(device)
inv_freq = rotary_emb.inv_freq
batch_size = question_ids.size(0)
ones_mask = torch.ones(batch_size, 1, dtype=question_attention_mask.dtype, device=device)
cache = FinchCache()
past_cache_len = 0
past_attention_mask = torch.zeros(batch_size, 0, dtype=question_attention_mask.dtype, device=device)
num_chunks = len(tokenized_doc_chunks)
query_context_matrices = {}
def query_hook_fn(module, input, output):
layer_idx = getattr(module, "layer_idx", None)
if layer_idx is not None:
query_states = output.detach()
bsz, seq_len, hidden_dim = query_states.size()
num_query_heads = module.num_query_heads
head_dim = hidden_dim // num_query_heads
query_states = (
query_states.view(bsz, seq_len, num_query_heads, head_dim)
.transpose(1, 2)
.contiguous()
)
query_context_matrices[layer_idx] = query_states[:, :, _current_chunk_offset:, :].clone()
hooks = []
for i, layer in enumerate(model.model.layers):
layer.self_attn.q_proj.layer_idx = i
layer.self_attn.q_proj.num_query_heads = layer.self_attn.config.num_attention_heads
hook = layer.self_attn.q_proj.register_forward_hook(query_hook_fn)
hooks.append(hook)
for j, tokenized_doc_chunk in enumerate(tokenized_doc_chunks):
current_seq_length = tokenized_doc_chunk["input_ids"].size(1)
_current_chunk_offset = current_seq_length
query_context_matrices.clear()
chunk_input_ids = tokenized_doc_chunk["input_ids"].contiguous()
chunk_attention_mask = tokenized_doc_chunk["attention_mask"].contiguous()
segment_attention_mask = torch.cat(
[past_attention_mask, chunk_attention_mask, ones_mask], dim=-1
).contiguous()
current_input_ids = torch.cat([chunk_input_ids, question_ids], dim=-1).contiguous()
current_attention_mask = torch.cat([segment_attention_mask, question_attention_mask], dim=-1).contiguous()
past_seen_tokens = cache.get_seq_length() if cache is not None else 0
cache_position = torch.arange(
past_seen_tokens + chunk_input_ids.shape[1],
past_seen_tokens + current_input_ids.shape[1],
device=device
)
causal_mask = model.model._prepare_4d_causal_attention_mask_with_cache_position(
current_attention_mask,
sequence_length=question_ids.size(1),
target_length=current_attention_mask.size(-1),
dtype=dtype,
device=device,
cache_position=cache_position,
batch_size=current_input_ids.size(0),
).contiguous()
with torch.no_grad():
outputs = model.model(
input_ids=current_input_ids,
use_cache=True,
past_key_values=cache,
)
cache = outputs.past_key_values
len_question = question_ids.size(1)
for layer_idx in range(len(model.model.layers)):
key_matrix = cache.key_cache[layer_idx]
query_matrix = query_context_matrices[layer_idx]
layer_cache_pos = torch.arange(
past_cache_len + current_seq_length,
past_cache_len + current_seq_length + len_question,
device=device
)
position_ids = layer_cache_pos.unsqueeze(0)
cos, sin = rotary_emb(query_matrix, position_ids)
cos = cos.unsqueeze(1)
sin = sin.unsqueeze(1)
query_matrix = (query_matrix * cos) + (rotate_half(query_matrix) * sin)
num_repeats = model.config.num_attention_heads // model.config.num_key_value_heads
key_matrix = repeat_kv(key_matrix, num_repeats)
scaling = math.sqrt(model.config.head_dim)
attention_matrix = torch.matmul(query_matrix, key_matrix.transpose(2, 3)) / scaling
causal_mask_sliced = causal_mask[:, :, :, : key_matrix.shape[-2]]
attention_matrix = attention_matrix + causal_mask_sliced
attention_matrix = torch.nn.functional.softmax(attention_matrix, dim=-1, dtype=torch.float32).to(query_matrix.dtype)
tol = 1e-8
binary_mask = (torch.abs(causal_mask_sliced.to(torch.float32)) < tol).to(torch.float32)
non_zero_counts = binary_mask.sum(dim=3, keepdim=True)
non_zero_counts = torch.clamp_min(non_zero_counts, 1.0).to(attention_matrix.dtype)
attention_matrix = attention_matrix / non_zero_counts
if j != num_chunks - 1:
attention_matrix = attention_matrix[:, :, :, : past_cache_len + current_seq_length].clone().contiguous()
else:
attention_matrix = attention_matrix[:, :, :, : past_cache_len + current_seq_length + len_question].clone().contiguous()
attention_matrix = torch.sum(attention_matrix, dim=-2)
attention_matrix = attention_matrix.view(
attention_matrix.size(0), model.config.num_key_value_heads, num_repeats, -1
).sum(dim=2)
full_context_size = attention_matrix.size(-1)
attention_matrix[..., :sink_tokens] = float("inf")
if j == num_chunks - 1:
attention_matrix[..., -len_question:] = float("inf")
if j == 0:
k = int(sink_tokens + (max(0, current_seq_length - sink_tokens) // compression_factor))
k = min(k + past_cache_len, full_context_size)
elif j < num_chunks - 1:
to_keep_new = int(current_seq_length // compression_factor)
k = min(past_cache_len + to_keep_new, full_context_size)
else:
desired_final = sink_tokens + target_token_size + len_question
k = desired_final if full_context_size >= desired_final else full_context_size
k = max(k, sink_tokens)
selected_indices = torch.topk(attention_matrix, k, dim=-1).indices
selected_indices, _ = torch.sort(selected_indices, dim=-1)
cache.compress_cache(layer_idx, selected_indices, inv_freq)
past_cache_len = cache._seen_tokens
past_attention_mask = torch.ones(1, past_cache_len, device=device)
for hook in hooks:
hook.remove()
return cache
except Exception as e:
raise RuntimeError(f"Failed to compress KV cache: {e}")
def run_naive_rag_query(collection_name, query, rag_token_size, prefix, task, few_shot_examples):
k = max(1, rag_token_size // 256)
vectorstore = Chroma(persist_directory="./chroma_db", embedding_function=embedding_model, collection_name=collection_name)
retriever = vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": k})
retrieved_docs = retriever.invoke(query)
for doc in retrieved_docs:
print("=================")
print(doc.page_content)
print("=================")
formatted_context = "\n\n".join([doc.page_content for doc in retrieved_docs])
rag_context = prefix + "Retrieved context: \n" + formatted_context + task + few_shot_examples
return rag_context
@spaces.GPU
def prepare_compression_and_rag(combined_text, retrieval_slider_value, global_local_value, task_description, few_shot, state, progress=gr.Progress()):
progress(0, desc="Starting compression process")
# percentage = int(global_local_value.replace('%', ''))
percentage = 0 if global_local_value == "RAG" else 100
progress(0.1, desc="Tokenizing text and preparing task")
question_text = task_description + "\n" + few_shot
context_encoding = tokenizer(combined_text, return_tensors="pt").to(device)
question_encoding = tokenizer(question_text, return_tensors="pt").to(device)
context_ids = context_encoding["input_ids"]
context_attention_mask = context_encoding["attention_mask"]
question_ids = question_encoding["input_ids"]
question_attention_mask = question_encoding["attention_mask"]
retrieval_context_length = int(context_ids.size(1) / retrieval_slider_value)
rag_tokens = int(retrieval_context_length * (1.0 - (percentage / 100)))
kv_tokens = retrieval_context_length - rag_tokens
progress(0.2, desc=f"Token breakdown computed: {kv_tokens} KV tokens, {rag_tokens} RAG tokens")
if percentage > 0:
target_token_size = int(retrieval_context_length * (percentage / 100))
progress(0.3, desc="Starting KV cache compression")
step_size = 2
try:
past_key_values = copy.deepcopy(get_compressed_kv_cache(sink_tokens, step_size, target_token_size,
context_ids, context_attention_mask,
question_ids, question_attention_mask))
except Exception as e:
progress(1, desc="Compression failed")
print("Error during KV cache compression:", e)
state["error"] = "Error during KV cache compression. Please try lowering the compression ratio and try again."
return state, False
compressed_length = past_key_values.get_seq_length()
progress(0.6, desc="KV cache compression completed")
else:
target_token_size = 0
past_key_values = FinchCache()
compressed_length = past_key_values.get_seq_length()
progress(0.3, desc="Skipping compression as percentage is 0")
current_timestamp = int(time.time())
cache_name = f"cache_{current_timestamp}_{uuid.uuid4().hex[:6]}.pt"
save_dir = "./cache_dir"
os.makedirs(save_dir, exist_ok=True)
save_path = os.path.join(save_dir, cache_name)
past_key_values.save(save_path)
progress(0.8, desc="Cache saved successfully")
collection_name = state.get("rag_index", None)
if collection_name is None:
print("Collection name not found; creating a new one.")
if combined_text.startswith(prefix):
rag_text = combined_text[len(prefix):]
else:
rag_text = combined_text
current_timestamp = int(time.time())
collection_name = f"default_{current_timestamp}_{uuid.uuid4().hex[:6]}"
rag_index = create_rag_index(collection_name, rag_text)
state.update({
"compressed_cache": save_path,
"rag_index": collection_name,
"global_local": percentage,
"task_description": task_description,
"few_shot": few_shot,
"retrieval_slider": retrieval_context_length,
})
progress(1, desc="Compression complete")
return state, "Document compressed successfully. You can now chat.", True
@spaces.GPU
def chat_response_stream(message: str, history: list, state: dict, compression_done: bool):
# Check if the document is compressed before allowing chat
if not compression_done or "compressed_cache" not in state:
yield "Document not compressed yet. Please compress the document first to enable chat."
return
user_message = message
save_path = state["compressed_cache"]
past_key_values = FinchCache.load(save_path, device=model.device)
compressed_length = past_key_values.get_seq_length()
collection_name = state["rag_index"]
retrieval_slider_value = state["retrieval_slider"]
percentage = state["global_local"]
rag_retrieval_size = int(retrieval_slider_value * (1.0 - (percentage / 100)))
print("RAG retrieval size: ", rag_retrieval_size)
print("Compressed cache: ", compressed_length)
if percentage == 0:
rag_prefix = prefix
rag_task = state["task_description"]
rag_few_shot = state["few_shot"]
else:
rag_prefix = ""
rag_task = ""
rag_few_shot = ""
print("user message: ", user_message)
if rag_retrieval_size != 0:
print("Running RAG query")
rag_context = run_naive_rag_query(collection_name, user_message, rag_retrieval_size, rag_prefix, rag_task, rag_few_shot)
new_input = rag_context + "\nquestion: " + user_message + suffix + "answer:"
else:
new_input = "\nquestion: " + user_message + suffix + "answer:"
tokenized_new_input = tokenizer(new_input, return_tensors="pt").to(device)
eos_block = torch.full((1, compressed_length), tokenizer.eos_token_id, device=device, dtype=torch.long)
new_input_ids = torch.cat([eos_block, tokenized_new_input["input_ids"]], dim=-1)
new_attention_mask = torch.cat([torch.ones((1, compressed_length), device=device), tokenized_new_input["attention_mask"]], dim=-1)
print("New input is: ", new_input)
streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
generate_kwargs = dict(
input_ids=new_input_ids,
attention_mask=new_attention_mask,
past_key_values=past_key_values,
streamer=streamer,
use_cache=True,
max_new_tokens=1024,
num_beams=1,
do_sample=False,
top_p=1.0,
top_k=None,
temperature=1.0,
# top_k=64,
# top_p=0.95,
# min_p=0.0
)
t = Thread(target=model.generate, kwargs=generate_kwargs)
t.start()
full_output = ""
for text in streamer:
full_output += text
time.sleep(0.05)
yield full_output
return full_output
def update_token_breakdown(token_count, retrieval_slider, global_local_value):
retrieval_context_length = int(token_count / retrieval_slider)
# percentage = int(global_local_value.replace('%', ''))
percentage = 0 if global_local_value == "RAG" else 100
rag_tokens = int(retrieval_context_length * (1.0 - (percentage / 100)))
kv_tokens = retrieval_context_length - rag_tokens
return f"Token Breakdown: {kv_tokens} tokens (KV compression), {rag_tokens} tokens (RAG retrieval)", f"Number of tokens after compression: {retrieval_context_length}"
##########################################################################
# Gradio Interface
##########################################################################
CSS = """
.main-container {
display: flex;
align-items: stretch;
}
.upload-section, .chatbot-container {
display: flex;
flex-direction: column;
height: 100%;
overflow-y: auto;
}
.upload-section {
padding: 10px;
border: 2px dashed #ccc;
border-radius: 10px;
}
.upload-button {
background: #34c759 !important;
color: white !important;
border-radius: 25px !important;
}
.chatbot-container {
margin-top: 0;
}
.status-output {
margin-top: 10px;
font-size: 14px;
}
.processing-info {
margin-top: 5px;
font-size: 12px;
color: #666;
}
.info-container {
margin-top: 10px;
padding: 10px;
border-radius: 5px;
}
.file-list {
margin-top: 0;
max-height: 200px;
overflow-y: auto;
padding: 5px;
border: 1px solid #eee;
border-radius: 5px;
}
.stats-box {
margin-top: 10px;
padding: 10px;
border-radius: 5px;
font-size: 12px;
}
.submit-btn {
background: #1a73e8 !important;
color: white !important;
border-radius: 25px !important;
margin-left: 10px;
padding: 5px 10px;
font-size: 16px;
}
.input-row {
display: flex;
align-items: center;
}
"""
def reset_chat_state():
return gr.update(value="Document not compressed yet. Please compress the document to enable chat."), False
with gr.Blocks(css=CSS, theme=gr.themes.Soft(font=["Arial", gr.themes.GoogleFont("Inconsolata"), "sans-serif"])) as demo:
# gr.HTML("<h1><center>Beyond RAG with LLama 3.1-8B-Instruct Model</center></h1>")
gr.HTML("<h1><center>Beyond RAG: Compress your document and chat with it.</center></h1>")
# Define chat_status_text as a Textbox with a set elem_id for custom styling.
chat_status_text = gr.Textbox(value="Document not compressed yet. Please compress the document to enable chat.", interactive=False, show_label=False, render=False, lines=5)
hidden_token_count = gr.State(value=0)
compression_done = gr.State(value=False)
compressed_doc_state = gr.State(value={})
with gr.Row(elem_classes="main-container"):
with gr.Column(elem_classes="upload-section"):
gr.Markdown("### Document Preprocessing")
with gr.Row():
file_input = gr.File(label="Drop file here or upload", file_count="multiple", elem_id="file-upload-area", height=120)
url_input = gr.Textbox(label="or enter a URL", placeholder="https://example.com/document.pdf", lines=2)
with gr.Row():
do_ocr = gr.Checkbox(label="Do OCR on Images", value=False, visible=False)
do_table = gr.Checkbox(label="Parse Tables", value=False, visible=False)
with gr.Accordion("Prompt Designer", open=False):
task_description_input = gr.Textbox(label="Task Description", value=default_task_description, lines=3, elem_id="task-description")
few_shot_input = gr.Textbox(label="Few-Shot Examples", value=default_few_shot, lines=10, elem_id="few-shot")
with gr.Accordion("Show Markdown Output", open=False):
markdown_output = gr.Textbox(label="Markdown Output", lines=20)
token_count_text = gr.Markdown("Number of tokens before compression: ")
retrieval_slider = gr.Slider(label="Select Compression Rate", minimum=1, maximum=32, step=1, value=2)
retrieval_info_text = gr.Markdown("Number of tokens after compression: ")
tokens_breakdown_text = gr.Markdown("Token breakdown will appear here.")
# global_local_slider = gr.Radio(label="Hybrid Retrieval (0 is all RAG, 100 is all global)",
# choices=["0%", "25%", "50%", "75%", "100%"], value="100%")
global_local_slider = gr.Radio(
label="Retrieval Mode",
choices=["RAG", "KVCompress"],
value="KVCompress"
)
compress_button = gr.Button("Compress Document", interactive=False, size="md", elem_classes="upload-button")
# File input: Run auto_convert then chain reset_chat_state.
file_input.change(
fn=auto_convert,
inputs=[file_input, url_input, do_ocr, do_table],
outputs=[markdown_output, token_count_text, retrieval_slider, retrieval_info_text,
hidden_token_count, compress_button, compression_done, compressed_doc_state, chat_status_text]
).then(
fn=reset_chat_state,
inputs=None,
outputs=[chat_status_text, compression_done]
)
# URL input: Run auto_convert then chain reset_chat_state.
url_input.change(
fn=auto_convert,
inputs=[file_input, url_input, do_ocr, do_table],
outputs=[markdown_output, token_count_text, retrieval_slider, retrieval_info_text,
hidden_token_count, compress_button, compression_done, compressed_doc_state, chat_status_text]
).then(
fn=reset_chat_state,
inputs=None,
outputs=[chat_status_text, compression_done]
)
# OCR checkbox: Run auto_convert then chain reset_chat_state.
do_ocr.change(
fn=auto_convert,
inputs=[file_input, url_input, do_ocr, do_table],
outputs=[markdown_output, token_count_text, retrieval_slider, retrieval_info_text,
hidden_token_count, compress_button, compression_done, compressed_doc_state, chat_status_text]
).then(
fn=reset_chat_state,
inputs=None,
outputs=[chat_status_text, compression_done]
)
# Table structure checkbox: Run auto_convert then chain reset_chat_state.
do_table.change(
fn=auto_convert,
inputs=[file_input, url_input, do_ocr, do_table],
outputs=[markdown_output, token_count_text, retrieval_slider, retrieval_info_text,
hidden_token_count, compress_button, compression_done, compressed_doc_state, chat_status_text]
).then(
fn=reset_chat_state,
inputs=None,
outputs=[chat_status_text, compression_done]
)
# Reset chat state when prompt designer fields change.
task_description_input.change(
fn=reset_chat_state,
inputs=None,
outputs=[chat_status_text, compression_done]
)
few_shot_input.change(
fn=reset_chat_state,
inputs=None,
outputs=[chat_status_text, compression_done]
)
# Reset chat state when the Markdown output changes.
markdown_output.change(
fn=reset_chat_state,
inputs=None,
outputs=[chat_status_text, compression_done]
)
# When sliders change, reset chat state.
retrieval_slider.change(
fn=reset_chat_state,
inputs=None,
outputs=[chat_status_text, compression_done]
)
global_local_slider.change(
fn=reset_chat_state,
inputs=None,
outputs=[chat_status_text, compression_done]
)
# Update token breakdown when sliders change.
retrieval_slider.change(
fn=update_token_breakdown,
inputs=[hidden_token_count, retrieval_slider, global_local_slider],
outputs=[tokens_breakdown_text, retrieval_info_text]
)
global_local_slider.change(
fn=update_token_breakdown,
inputs=[hidden_token_count, retrieval_slider, global_local_slider],
outputs=[tokens_breakdown_text, retrieval_info_text]
)
# Compress button: Prepare compression and then update chat status.
compress_button.click(
fn=prepare_compression_and_rag,
inputs=[markdown_output, retrieval_slider, global_local_slider, task_description_input, few_shot_input, compressed_doc_state],
outputs=[compressed_doc_state, chat_status_text, compression_done]
)
with gr.Column(elem_classes="chatbot-container"):
chat_status_text.render()
gr.Markdown("## Chat (LLama 3.1-8B-Instruct)")
gr.Markdown("**Note:** There is currently no chat history available.")
chat_interface = gr.ChatInterface(
fn=chat_response_stream,
additional_inputs=[compressed_doc_state, compression_done],
type="messages",
fill_height=True
)
demo.queue(max_size=16).launch() |