Spaces:
Sleeping
Sleeping
File size: 33,147 Bytes
c27ef68 76cfc68 c27ef68 94f7913 |
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 |
import gradio as gr
import spaces
import torch
import torch.nn as nn
from torch.nn import functional as F
import numpy as np
import math
import os
import pickle
import requests
import textwrap
import subprocess
import shutil
import time
from dataclasses import dataclass
from typing import Optional
from transformers import AutoTokenizer
# ==============================================================================
# ------------------------- VERSION 1: SHARED SETUP ----------------------------
# ==============================================================================
def setup_environment():
"""Checks for and sets up the necessary data for V1."""
nano_gpt_repo_path = 'nanoGPT'
data_dir_path = 'shakespeare_char'
meta_path = os.path.join(data_dir_path, 'meta.pkl')
if os.path.exists(meta_path):
return
print("Required data not found. Starting one-time setup...")
if not os.path.exists(nano_gpt_repo_path):
try:
subprocess.run(['git', 'clone', 'https://github.com/karpathy/nanoGPT.git'], check=True, capture_output=True, text=True)
except subprocess.CalledProcessError as e:
print(f"Error cloning repository: {e.stderr}")
pass
source_data_dir = os.path.join(nano_gpt_repo_path, 'data', 'shakespeare_char')
if not os.path.exists(data_dir_path) and os.path.exists(source_data_dir):
shutil.copytree(source_data_dir, data_dir_path)
# Check if we can run prepare
prepare_script_path = os.path.join(data_dir_path, 'prepare.py')
if os.path.exists(prepare_script_path) and not os.path.exists(meta_path):
subprocess.run(['python', 'prepare.py'], check=True, cwd=data_dir_path, capture_output=True, text=True)
setup_environment()
def download_file(url, filename):
if os.path.exists(filename):
return
print(f"Downloading '{filename}'...")
try:
response = requests.get(url, stream=True)
response.raise_for_status()
with open(filename, 'wb') as f:
for chunk in response.iter_content(chunk_size=8192):
f.write(chunk)
except requests.exceptions.RequestException as e:
print(f"Error downloading {url}: {e}")
# ==============================================================================
# ---------------------- VERSION 1: ARCHITECTURE & LOGIC -----------------------
# ==============================================================================
# V1 Constants and Meta Loading
v1_data_dir = './shakespeare_char/'
v1_meta_url = 'https://huggingface.co/spaces/thejagstudio/NanoDiffusion/resolve/main/meta.pkl'
v1_meta_path = 'meta.pkl'
download_file(v1_meta_url, v1_meta_path)
v1_vocab_size = 65 # Fallback
v1_itos = {}
v1_stoi = {}
if os.path.exists(v1_meta_path):
with open(v1_meta_path, 'rb') as f:
meta = pickle.load(f)
v1_vocab_size = meta['vocab_size']
v1_itos = meta['itos']
v1_stoi = meta['stoi']
v1_context_length = 256
device = 'cuda' if torch.cuda.is_available() else 'cpu'
def v1_decode(indices_tensor: torch.Tensor):
if indices_tensor.dim() > 1:
indices_tensor = indices_tensor.squeeze(0)
indices = indices_tensor.cpu().numpy()
return ''.join([v1_itos.get(i, '?') for i in indices])
def wrap_text(long_text, width=80):
paragraphs = long_text.splitlines()
wrapped = [textwrap.fill(p, width=width) if p else '' for p in paragraphs]
return "\n".join(wrapped)
@dataclass
class V1_GPTConfig:
block_size: int = 1024
vocab_size: int = 50304
n_layer: int = 12
n_head: int = 12
n_embd: int = 768
cond_dim: int = 64
dropout: float = 0.0
bias: bool = False
class V1_MLP(nn.Module):
def __init__(self, config):
super().__init__()
self.c_fc = nn.Linear(config.n_embd, 4 * config.n_embd, bias=config.bias)
self.gelu = nn.GELU()
self.c_proj = nn.Linear(4 * config.n_embd, config.n_embd, bias=config.bias)
self.dropout = nn.Dropout(config.dropout)
def forward(self, x):
x = self.c_fc(x)
x = self.gelu(x)
x = self.c_proj(x)
x = self.dropout(x)
return x
class V1_SelfAttention(nn.Module):
def __init__(self, config):
super().__init__()
assert config.n_embd % config.n_head == 0
self.c_attn = nn.Linear(config.n_embd, 3 * config.n_embd, bias=config.bias)
self.c_proj = nn.Linear(config.n_embd, config.n_embd, bias=config.bias)
self.attn_dropout = nn.Dropout(config.dropout)
self.resid_dropout = nn.Dropout(config.dropout)
self.n_head = config.n_head
self.n_embd = config.n_embd
self.dropout = config.dropout
self.flash = hasattr(torch.nn.functional, 'scaled_dot_product_attention')
def forward(self, x):
B, T, C = x.size()
q, k, v = self.c_attn(x).split(self.n_embd, dim=2)
k = k.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)
q = q.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)
v = v.view(B, T, self.n_head, C // self.n_head).transpose(1, 2)
if self.flash:
y = torch.nn.functional.scaled_dot_product_attention(q, k, v, attn_mask=None, dropout_p=self.dropout if self.training else 0, is_causal=False)
else:
att = (q @ k.transpose(-2, -1)) * (1.0 / math.sqrt(k.size(-1)))
att = F.softmax(att, dim=-1)
att = self.attn_dropout(att)
y = att @ v
y = y.transpose(1, 2).contiguous().view(B, T, C)
y = self.resid_dropout(self.c_proj(y))
return y
def v1_modulate(x: torch.Tensor, shift: torch.Tensor, scale: torch.Tensor) -> torch.Tensor:
return x * (1 + scale) + shift
def v1_bias_add_scale(x: torch.Tensor, bias: Optional[torch.Tensor], scale: torch.Tensor, residual: Optional[torch.Tensor]) -> torch.Tensor:
if bias is not None:
out = scale * (x + bias)
else:
out = scale * x
if residual is not None:
out = residual + out
return out
class V1_DDiTBlock(nn.Module):
def __init__(self, config):
super().__init__()
self.ln_1 = nn.LayerNorm(config.n_embd, bias=config.bias)
self.attn = V1_SelfAttention(config)
self.ln_2 = nn.LayerNorm(config.n_embd, bias=config.bias)
self.mlp = V1_MLP(config)
self.adaLN_modulation = nn.Linear(config.cond_dim, 6 * config.n_embd)
self.adaLN_modulation.weight.data.zero_()
self.adaLN_modulation.bias.data.zero_()
def forward(self, x, c):
shift_msa, scale_msa, gate_msa, shift_mlp, scale_mlp, gate_mlp = self.adaLN_modulation(c)[:, None].chunk(6, dim=2)
x_skip = x
x = v1_modulate(self.ln_1(x), shift_msa, scale_msa)
x = self.attn(x)
x = v1_bias_add_scale(self.attn(self.ln_1(x)), None, gate_msa, x_skip)
x = v1_bias_add_scale(self.mlp(v1_modulate(self.ln_2(x), shift_mlp, scale_mlp)), None, gate_mlp, x)
return x
class V1_DDitFinalLayer(nn.Module):
def __init__(self, config):
super().__init__()
self.norm_final = nn.LayerNorm(config.n_embd, bias=config.bias)
self.linear = nn.Linear(config.n_embd, config.vocab_size)
self.linear.weight.data.zero_()
self.linear.bias.data.zero_()
self.adaLN_modulation = nn.Linear(config.cond_dim, 2 * config.n_embd)
self.adaLN_modulation.weight.data.zero_()
self.adaLN_modulation.bias.data.zero_()
def forward(self, x, c):
shift, scale = self.adaLN_modulation(c)[:, None].chunk(2, dim=2)
x = v1_modulate(self.norm_final(x), shift, scale)
x = self.linear(x)
return x
class V1_TimestepEmbedder(nn.Module):
def __init__(self, hidden_size, frequency_embedding_size=256):
super().__init__()
self.mlp = nn.Sequential(
nn.Linear(frequency_embedding_size, hidden_size, bias=True),
nn.SiLU(),
nn.Linear(hidden_size, hidden_size, bias=True),
)
self.frequency_embedding_size = frequency_embedding_size
@staticmethod
def timestep_embedding(t, dim, max_period=10000):
half = dim // 2
freqs = torch.exp(
-math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half
).to(device=t.device)
args = t[:, None].float() * freqs[None]
embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
if dim % 2:
embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
return embedding
def forward(self, t):
t_freq = self.timestep_embedding(t, self.frequency_embedding_size)
t_emb = self.mlp(t_freq)
return t_emb
class V1_GPT(nn.Module):
def __init__(self, config):
super().__init__()
assert config.vocab_size is not None
assert config.block_size is not None
self.config = config
self.sigma_map = V1_TimestepEmbedder(config.cond_dim)
self.transformer = nn.ModuleDict(dict(
wte = nn.Embedding(config.vocab_size, config.n_embd),
wpe = nn.Embedding(config.block_size, config.n_embd),
drop = nn.Dropout(config.dropout),
h = nn.ModuleList([V1_DDiTBlock(config) for _ in range(config.n_layer)]),
ln_f = nn.LayerNorm(config.n_embd, bias=config.bias),
))
self.lm_head = V1_DDitFinalLayer(config)
self.apply(self._init_weights)
for pn, p in self.named_parameters():
if pn.endswith('c_proj.weight'):
torch.nn.init.normal_(p, mean=0.0, std=0.02/math.sqrt(2 * config.n_layer))
def _init_weights(self, module):
if isinstance(module, nn.Linear):
torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
if module.bias is not None:
torch.nn.init.zeros_(module.bias)
elif isinstance(module, nn.Embedding):
torch.nn.init.normal_(module.weight, mean=0.0, std=0.02)
def forward(self, idx, sigma):
sigma = sigma.reshape(-1)
b, t = idx.size()
c = F.silu(self.sigma_map(sigma))
assert t <= self.config.block_size, f"Cannot forward sequence of length {t}, block size is only {self.config.block_size}"
pos = torch.arange(0, t, dtype=torch.long, device=device)
tok_emb = self.transformer.wte(idx)
pos_emb = self.transformer.wpe(pos)
x = self.transformer.drop(tok_emb + pos_emb)
for block in self.transformer.h:
x = block(x, c)
x = self.transformer.ln_f(x)
x = self.lm_head(x, c)
x = torch.scatter(x, -1, idx[..., None], torch.zeros_like(x[..., :1]))
return x
class V1_GeometricNoise:
def __init__(self, sigma_min=1e-4, sigma_max=20):
self.sigmas = 1.0 * torch.tensor([sigma_min, sigma_max]).to(device)
def rate_noise(self, t):
return self.sigmas[0] ** (1 - t) * self.sigmas[1] ** t * (self.sigmas[1].log() - self.sigmas[0].log())
def total_noise(self, t):
return self.sigmas[0] ** (1 - t) * self.sigmas[1] ** t
def __call__(self, t):
return self.total_noise(t), self.rate_noise(t)
# --- V1 Inference Logic ---
def v1_transition(x_t: torch.Tensor, delta_sigma: torch.Tensor) -> torch.Tensor:
base_prob = (1 - torch.exp(-delta_sigma[..., None])) / v1_vocab_size
trans = torch.ones(*x_t.shape, v1_vocab_size, device=x_t.device) * base_prob
trans = trans.scatter(-1, x_t[..., None], torch.zeros_like(trans))
diag_fill = 1 - trans.sum(dim=-1, keepdim=True)
trans = trans.scatter(-1, x_t[..., None], diag_fill)
return trans
def v1_staggered_score(score, delta_sigma):
exp_factor = torch.exp(-delta_sigma)[..., None]
correction = ((exp_factor - 1) / (v1_vocab_size * exp_factor)) * score.sum(dim=-1, keepdim=True)
return correction + score / exp_factor
def v1_sample_categorical(probs: torch.Tensor) -> torch.Tensor:
eps = 1e-10
gumbel_noise = -torch.log(-torch.log(torch.rand_like(probs) + eps) + eps)
return torch.argmax(torch.log(probs + eps) + gumbel_noise, dim=-1)
# --- V1 Model Loading ---
print("Initializing V1 Model...")
v1_model_args = dict(n_layer=6, n_head=6, n_embd=384, cond_dim=64,
bias=False, vocab_size=v1_vocab_size, block_size=v1_context_length, dropout=0.2)
v1_config = V1_GPTConfig(**v1_model_args)
v1_model = V1_GPT(v1_config)
try:
v1_model.load_state_dict(
torch.hub.load_state_dict_from_url(
'https://huggingface.co/spaces/thejagstudio/NanoDiffusion/resolve/main/final_model.pth?download=true',
map_location=device
)
)
v1_model.to(device)
v1_model.eval()
print("V1 Model loaded successfully.")
except Exception as e:
print(f"Failed to load V1 model: {e}")
v1_model = None
v1_noise = V1_GeometricNoise(sigma_min=1e-4, sigma_max=20)
def v1_generate_stream(steps, speed):
"""
Generator function for V1 that yields frames directly.
Combined logic of generation and replay to allow for immediate stopping.
"""
if v1_model is None:
yield "Error: V1 Model not loaded"
return
steps = int(steps)
speed = float(speed)
eps = 1e-5
# Calculate delay based on speed slider (similar to V2)
# 0.5 is base constant, speed scales it down
delay = 0.5 / max(speed, 0.1)
x = torch.randint(0, v1_vocab_size, (1, v1_context_length), device=device)
initial_text = f"--- Initial Random Noise ---\n\n{wrap_text(v1_decode(x[0]))}"
yield initial_text
time.sleep(delay)
timesteps = torch.linspace(1, eps, steps + 1, device=device)
step_size = (1 - eps) / steps
with torch.no_grad():
for i in range(steps):
t = timesteps[i] * torch.ones(x.shape[0], 1, device=device)
curr_sigma_bar = v1_noise(t)[0]
next_sigma_bar = v1_noise(t - step_size)[0]
delta_sigma = curr_sigma_bar - next_sigma_bar
log_score = v1_model(x, curr_sigma_bar)
score = torch.exp(log_score)
stag_score = v1_staggered_score(score, delta_sigma)
probs = stag_score * v1_transition(x, delta_sigma)
x = v1_sample_categorical(probs)
progress_text = f"--- Denoising Step {i + 1}/{steps} ---\n\n{wrap_text(v1_decode(x[0]))}"
yield progress_text
# Artificial delay for visualization
if speed < 20:
time.sleep(delay)
t = timesteps[steps] * torch.ones(x.shape[0], 1, device=device)
curr_sigma_bar = v1_noise(t)[0]
delta_sigma = curr_sigma_bar
log_score = v1_model(x, curr_sigma_bar)
score = torch.exp(log_score)
stag_score = v1_staggered_score(score, delta_sigma)
probs = stag_score * v1_transition(x, delta_sigma)
x = v1_sample_categorical(probs)
final_text = f"--- Final Denoised Text (Step {steps}) ---\n\n{wrap_text(v1_decode(x[0]))}"
yield final_text
# ==============================================================================
# ---------------------- VERSION 2: ARCHITECTURE & LOGIC -----------------------
# ==============================================================================
# PLEASE UPDATE THIS PATH TO YOUR ACTUAL LOCAL FILE OR URL
V2_MODEL_PATH = "checkpoints/model_fp32.pt"
class V2_RMSNorm(nn.Module):
def __init__(self, dim: int, eps: float = 1e-6):
super().__init__()
self.eps = eps
self.weight = nn.Parameter(torch.ones(dim))
def forward(self, x):
var = x.pow(2).mean(-1, keepdim=True)
x = x * torch.rsqrt(var + self.eps)
return self.weight * x
class V2_RotaryEmbedding(nn.Module):
def __init__(self, dim, max_position_embeddings=16384, base=100000, scaling_factor=1.0):
super().__init__()
self.scaling_factor = scaling_factor
self.dim = dim
self.base = base
self.max_position_embeddings = max_position_embeddings
self.inv_freq = None
self._cache = {}
def _update_freqs(self, device):
base = self.base * (self.scaling_factor ** (self.dim / (self.dim - 2)))
inv_freq = 1.0 / (base ** (torch.arange(0, self.dim, 2).float().to(device) / self.dim))
self.inv_freq = inv_freq
def forward(self, x, seq_len=None):
if seq_len is None:
seq_len = x.shape[-2]
if self.inv_freq is None or self.inv_freq.device != x.device:
self._update_freqs(x.device)
cache_key = (seq_len, x.device, x.dtype)
if cache_key in self._cache:
return self._cache[cache_key]
t = torch.arange(seq_len, device=x.device, dtype=self.inv_freq.dtype)
freqs = torch.einsum("i,j->ij", t, self.inv_freq)
emb = torch.cat((freqs, freqs), dim=-1)
cos = emb.cos()[None, None, :, :]
sin = emb.sin()[None, None, :, :]
self._cache[cache_key] = (cos, sin)
if len(self._cache) > 10:
self._cache.pop(next(iter(self._cache)))
return cos, sin
def v2_apply_rotary_pos_emb(q, k, cos, sin):
def rotate_half(x):
x1 = x[..., : x.shape[-1] // 2]
x2 = x[..., x.shape[-1] // 2 :]
return torch.cat((-x2, x1), dim=-1)
q_embed = (q * cos) + (rotate_half(q) * sin)
k_embed = (k * cos) + (rotate_half(k) * sin)
return q_embed, k_embed
class V2_DiffusionAttention(nn.Module):
def __init__(self, config):
super().__init__()
self.hidden_size = config.hidden_size
self.num_heads = config.num_attention_heads
self.head_dim = self.hidden_size // self.num_heads
self.num_key_value_heads = config.num_key_value_heads
self.num_key_value_groups = self.num_heads // self.num_key_value_heads
self.use_flash_attn = config.use_flash_attn
self.q_proj = nn.Linear(self.hidden_size, self.num_heads * self.head_dim, bias=False)
self.k_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
self.v_proj = nn.Linear(self.hidden_size, self.num_key_value_heads * self.head_dim, bias=False)
self.o_proj = nn.Linear(self.num_heads * self.head_dim, self.hidden_size, bias=False)
def forward(self, hidden_states, freqs_cis, attention_mask=None, past_kv=None):
bsz, q_len, _ = hidden_states.size()
q = self.q_proj(hidden_states).view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
k = self.k_proj(hidden_states).view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
v = self.v_proj(hidden_states).view(bsz, q_len, self.num_key_value_heads, self.head_dim).transpose(1, 2)
cos, sin = freqs_cis
cos = cos[:, :, :q_len, :]
sin = sin[:, :, :q_len, :]
q, k = v2_apply_rotary_pos_emb(q, k, cos, sin)
if past_kv is not None:
cache_k, cache_v = past_kv
k = torch.cat([cache_k, k], dim=2)
v = torch.cat([cache_v, v], dim=2)
current_kv = (k, v)
k = k.repeat_interleave(self.num_key_value_groups, dim=1)
v = v.repeat_interleave(self.num_key_value_groups, dim=1)
attn_mask = None
if attention_mask is not None:
attn_mask = attention_mask[:, None, None, :].to(dtype=q.dtype)
attn_mask = (1.0 - attn_mask) * torch.finfo(q.dtype).min
output = F.scaled_dot_product_attention(
q, k, v, attn_mask=attn_mask, dropout_p=0.0, is_causal=False
)
output = output.transpose(1, 2).contiguous().view(bsz, q_len, self.hidden_size)
return self.o_proj(output), current_kv
class V2_MLP(nn.Module):
def __init__(self, config):
super().__init__()
self.gate_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
self.up_proj = nn.Linear(config.hidden_size, config.intermediate_size, bias=False)
self.down_proj = nn.Linear(config.intermediate_size, config.hidden_size, bias=False)
self.act_fn = nn.SiLU()
def forward(self, x):
return self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
class V2_BlockDiffusionBlock(nn.Module):
def __init__(self, config):
super().__init__()
self.self_attn = V2_DiffusionAttention(config)
self.mlp = V2_MLP(config)
self.input_layernorm = V2_RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.post_attention_layernorm = V2_RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.use_activation_checkpointing = config.use_activation_checkpointing
def forward(self, hidden_states, freqs_cis, attention_mask, past_kv):
return self._forward(hidden_states, freqs_cis, attention_mask, past_kv)
def _forward(self, hidden_states, freqs_cis, attention_mask, past_kv):
residual = hidden_states
hidden_states = self.input_layernorm(hidden_states)
attn_out, new_kv = self.self_attn(hidden_states, freqs_cis, attention_mask, past_kv)
hidden_states = residual + attn_out
residual = hidden_states
hidden_states = self.post_attention_layernorm(hidden_states)
hidden_states = residual + self.mlp(hidden_states)
return hidden_states, new_kv
@dataclass
class V2_ModelConfig:
vocab_size: int = 151936
hidden_size: int = 1024
intermediate_size: int = 2816
num_hidden_layers: int = 16
num_attention_heads: int = 16
num_key_value_heads: int = 4
max_position_embeddings: int = 16384
rms_norm_eps: float = 1e-6
rope_theta: float = 100000.0
pad_token_id: int = 0
mask_token_id: int = 1
use_flash_attn: bool = True
use_activation_checkpointing: bool = False
attention_dropout: float = 0.0
hidden_dropout: float = 0.0
ModelConfig = V2_ModelConfig
class V2_DiffusionLLM(nn.Module):
def __init__(self, config: V2_ModelConfig):
super().__init__()
self.config = config
pad_idx = config.pad_token_id if config.pad_token_id < config.vocab_size else None
self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, padding_idx=pad_idx)
self.layers = nn.ModuleList([V2_BlockDiffusionBlock(config) for _ in range(config.num_hidden_layers)])
self.norm = V2_RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
self.rotary_emb = V2_RotaryEmbedding(
config.hidden_size // config.num_attention_heads,
config.max_position_embeddings
)
self.lm_head.weight = self.embed_tokens.weight
def forward(self, input_ids, attention_mask=None, past_key_values=None):
bsz, seqlen = input_ids.shape
hidden_states = self.embed_tokens(input_ids)
freqs_cis = self.rotary_emb(hidden_states, seq_len=seqlen)
if past_key_values is None:
past_key_values = [None] * len(self.layers)
new_kvs = []
for i, layer in enumerate(self.layers):
hidden_states, kv = layer(hidden_states, freqs_cis, attention_mask, past_key_values[i])
new_kvs.append(kv)
hidden_states = self.norm(hidden_states)
logits = self.lm_head(hidden_states)
return logits, new_kvs
DiffusionLLM = V2_DiffusionLLM
# --- V2 Loading Logic ---
print("Initializing V2 components...")
v2_tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen2.5-0.5B")
if v2_tokenizer.pad_token is None:
v2_tokenizer.pad_token = v2_tokenizer.eos_token
v2_model = None
v2_config = None
if os.path.exists(V2_MODEL_PATH):
print(f"Loading V2 model from {V2_MODEL_PATH}...")
try:
checkpoint = torch.load(V2_MODEL_PATH, map_location=device, weights_only=False)
v2_config = checkpoint['config']
v2_model = V2_DiffusionLLM(v2_config)
state_dict = checkpoint['model_state']
state_dict = {k: v.float() for k, v in state_dict.items()}
v2_model.load_state_dict(state_dict)
v2_model = v2_model.to(device)
v2_model.eval()
print("V2 Model loaded.")
except Exception as e:
print(f"Error loading V2 model: {e}")
else:
print(f"V2 Model file not found at {V2_MODEL_PATH}. Version 2 tab will not work without it.")
@torch.no_grad()
def v2_generate_block_diffusion(prompt, steps, block_size, max_new_tokens, replay_speed):
"""
Refactored to yield frames for real-time streaming.
"""
if v2_model is None:
yield "Error: V2 Model not found. Check path."
return
v2_model.eval()
# Handle inputs
steps = int(steps)
block_size = int(block_size)
max_new_tokens = int(max_new_tokens)
speed = float(replay_speed)
prompt_ids = v2_tokenizer.encode(prompt, return_tensors="pt").to(device)
config = v2_model.config
num_blocks = max_new_tokens // block_size
context_ids = prompt_ids
# Helper params
temperature = 1.0
top_k = 40
top_p = 0.9
repetition_penalty = 1.2
# Calculate delay based on speed slider
delay = 0.5 / max(speed, 0.1)
for block_idx in range(num_blocks):
mask_block = torch.full((1, block_size), config.mask_token_id, device=device)
is_masked = torch.ones(1, block_size, dtype=torch.bool, device=device)
for step_idx in range(steps):
# --- SNAPSHOT & YIELD ---
# Decode context
ctx_str = v2_tokenizer.decode(context_ids[0], skip_special_tokens=True)
# Decode block with masking visual
block_tokens = mask_block[0].tolist()
block_vis = []
for i, tid in enumerate(block_tokens):
if is_masked[0, i]:
block_vis.append("β") # Mask symbol
else:
block_vis.append(v2_tokenizer.decode([tid], skip_special_tokens=False))
block_str = "".join(block_vis)
frame_text = (f"--- Generating Block {block_idx+1}/{num_blocks} | Step {step_idx+1}/{steps} ---\n\n"
f"{ctx_str}{block_str}")
yield frame_text
# Artificial delay to visualize the step
if speed < 20: # If max speed, skip sleep
time.sleep(delay)
# ------------------------
full_input = torch.cat([context_ids, mask_block], dim=1)
attention_mask = torch.ones_like(full_input, dtype=torch.float32)
logits, _ = v2_model(full_input, attention_mask=attention_mask)
block_logits = logits[:, -block_size:, :]
# Repetition penalty
if repetition_penalty != 1.0:
seen_tokens = set(context_ids[0].tolist())
for i in range(block_size):
if not is_masked[0, i]:
seen_tokens.add(mask_block[0, i].item())
for token_id in seen_tokens:
if token_id < block_logits.shape[-1]:
if block_logits[0, :, token_id].mean() > 0:
block_logits[:, :, token_id] /= repetition_penalty
else:
block_logits[:, :, token_id] *= repetition_penalty
block_logits = block_logits / temperature
# Top-K
if top_k > 0:
top_k_logits, top_k_indices = torch.topk(block_logits, top_k, dim=-1)
block_logits = torch.full_like(block_logits, float('-inf'))
block_logits.scatter_(-1, top_k_indices, top_k_logits)
# Top-P
if top_p < 1.0:
sorted_logits, sorted_indices = torch.sort(block_logits, descending=True, dim=-1)
cumulative_probs = torch.cumsum(F.softmax(sorted_logits, dim=-1), dim=-1)
sorted_indices_to_remove = cumulative_probs > top_p
sorted_indices_to_remove[..., 1:] = sorted_indices_to_remove[..., :-1].clone()
sorted_indices_to_remove[..., 0] = 0
indices_to_remove = sorted_indices_to_remove.scatter(-1, sorted_indices, sorted_indices_to_remove)
block_logits[indices_to_remove] = float('-inf')
probs = F.softmax(block_logits, dim=-1)
probs = torch.nan_to_num(probs, nan=0.0, posinf=0.0, neginf=0.0)
probs = probs.clamp(min=1e-10)
probs = probs / probs.sum(dim=-1, keepdim=True)
sampled_tokens = torch.multinomial(probs.view(-1, probs.size(-1)), num_samples=1)
sampled_tokens = sampled_tokens.view(1, block_size)
confidence = probs.gather(-1, sampled_tokens.unsqueeze(-1)).squeeze(-1)
tokens_to_unmask = max(1, block_size // steps)
if step_idx == steps - 1:
tokens_to_unmask = is_masked.sum().item()
if tokens_to_unmask > 0 and is_masked.sum() > 0:
masked_confidence = confidence.clone()
masked_confidence[~is_masked] = -1.0
num_to_unmask = min(tokens_to_unmask, is_masked.sum().item())
_, top_indices = torch.topk(masked_confidence.view(-1), num_to_unmask)
for idx in top_indices:
mask_block[0, idx] = sampled_tokens[0, idx]
is_masked[0, idx] = False
context_ids = torch.cat([context_ids, mask_block], dim=1)
generated_ids = context_ids[0].tolist()
final_text = v2_tokenizer.decode(generated_ids, skip_special_tokens=True)
yield final_text
# ==============================================================================
# ------------------------------- GRADIO UI ------------------------------------
# ==============================================================================
css = '''.gradio-container > .fillable {max-width: 900px !important}
h3{margin-top: 1em}
p{margin-top: 0}
textarea{font-family: monospace; background-color: #1a1b1e; color: #e0e0e0}
'''
with gr.Blocks(theme=gr.themes.Citrus(), css=css) as demo:
gr.Markdown("# Diffusion Language Models Playground")
with gr.Tabs():
# --- TAB 1: VERSION 1 (CHAR DIFFUSION) ---
with gr.Tab("Version 1: Character Diffusion (NanoGPT)"):
gr.Markdown("### Tiny 11M parameter character-based continuous diffusion.")
with gr.Row():
with gr.Column(scale=1):
v1_steps = gr.Slider(64, 512, 128, step=1, label="Denoising Steps")
v1_speed = gr.Slider(1, 20, 10, step=1, label="Generation/Replay Speed")
with gr.Row():
v1_btn = gr.Button("Generate", variant="primary")
v1_stop = gr.Button("Stop", variant="stop")
with gr.Column(scale=2):
v1_out = gr.Textbox(label="Generated Text", lines=15, interactive=False)
# V1 Logic: Merged generation and replay for proper stopping
v1_event = v1_btn.click(v1_generate_stream, inputs=[v1_steps, v1_speed], outputs=[v1_out])
v1_stop.click(fn=None, inputs=None, outputs=None, cancels=[v1_event])
# --- TAB 2: VERSION 2 (BLOCK DIFFUSION) ---
with gr.Tab("Version 2: Block Diffusion (LLaDA-style)"):
gr.Markdown("### Block-based diffusion using Qwen tokenizer.")
if v2_model is None:
gr.Warning(f"V2 Model not loaded. Please check path: {V2_MODEL_PATH}")
with gr.Row():
with gr.Column(scale=1):
v2_prompt = gr.Textbox(label="Prompt", value="The king went to the")
v2_steps = gr.Slider(4, 64, 16, step=1, label="Steps per Block")
v2_block_size = gr.Slider(8, 126, 32, step=8, label="Block Size")
v2_max_tokens = gr.Slider(32, 1024, 128, step=32, label="Total New Tokens")
v2_speed = gr.Slider(1, 20, 1, step=1, label="Generation/Replay Speed")
with gr.Row():
v2_btn = gr.Button("Generate", variant="primary")
v2_stop = gr.Button("Stop", variant="stop")
with gr.Column(scale=2):
v2_out = gr.Textbox(label="Generated Text", lines=15, interactive=False)
# V2 Logic
v2_event = v2_btn.click(
v2_generate_block_diffusion,
inputs=[v2_prompt, v2_steps, v2_block_size, v2_max_tokens, v2_speed],
outputs=[v2_out]
)
v2_stop.click(fn=None, inputs=None, outputs=None, cancels=[v2_event])
if __name__ == "__main__":
demo.launch() |