File size: 47,022 Bytes
19898f1 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 | """
Experiment 2-A (Modified): Image-conditioned Representation Analysis
Modification from original:
- Remove task format confound by unifying answer format
- All answers are pure spatial concepts: left, right, above, under, far, close
- Pairwise: "Is the {obj1} to the left or right of the {obj2}?" -> "left"
- Distance: "Compared to {ref}, is {target} far or close from you?" -> "far"
- 200 samples per category (up from 50)
Goal: Verify Hypothesis 4 - that above/far and under/close are mapped to similar
positions in embedding space, while left/right are well-separated.
"""
import os
import sys
import json
import argparse
import base64
import logging
import random
import re
from io import BytesIO
from collections import defaultdict
from typing import Dict, List, Tuple, Optional, Any
from abc import ABC, abstractmethod
import torch
import numpy as np
import pandas as pd
from PIL import Image
from tqdm import tqdm
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.metrics.pairwise import cosine_similarity
# Setup logging
logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
logger = logging.getLogger(__name__)
# Category order for output
CATEGORY_ORDER = ['left', 'right', 'above', 'under', 'far', 'close']
# Pair definitions for trajectory analysis
TRAJECTORY_PAIRS = {
'hypothesis': [
('above', 'far', 'above-far', '#d62728'), # red
('under', 'close', 'under-close', '#1f77b4'), # blue
],
'within_axis': [
('left', 'right', 'left-right', '#2ca02c'), # green
('above', 'under', 'above-under', '#ff7f0e'), # orange
('far', 'close', 'far-close', '#9467bd'), # purple
],
'counter_hypothesis': [
('above', 'close', 'above-close', '#e377c2'), # pink
('under', 'far', 'under-far', '#17becf'), # cyan
],
}
# Scale colors for cross-scale plots
SCALE_COLORS = {
'vanilla': '#1f77b4',
'80k': '#ff7f0e',
'400k': '#2ca02c',
'800k': '#d62728',
'2m': '#9467bd',
'roborefer': '#8c564b',
}
# ============================================================================
# Data Loading & Modification
# ============================================================================
# Regex patterns for extracting objects from pairwise questions
OBJECT_PATTERNS = [
re.compile(r'between\s+(.+?)\s+and\s+(.+?)\s+in', re.IGNORECASE),
re.compile(r'of\s+(.+?)\s+and\s+(.+?)\s+in', re.IGNORECASE),
re.compile(r'positions\s+of\s+(.+?)\s+and\s+(.+?)\s+interact', re.IGNORECASE),
re.compile(r'How\s+are\s+(.+?)\s+and\s+(.+?)\s+positioned', re.IGNORECASE),
re.compile(r'arrangement\s+of\s+(.+?)\s+and\s+(.+?)\s+in', re.IGNORECASE),
]
def extract_objects(question: str) -> Tuple[str, str]:
"""Extract two objects from a pairwise relation question."""
for pattern in OBJECT_PATTERNS:
m = pattern.search(question)
if m:
return m.group(1).strip(), m.group(2).strip()
raise ValueError(f"Could not extract objects from: {question}")
def modify_pairwise_sample(sample: dict) -> dict:
"""Modify a pairwise relation sample (left/right/above/under)."""
obj1, obj2 = extract_objects(sample['question'])
category = sample['category']
if category in ['left', 'right']:
new_question = f"Is the {obj1} to the left or right of the {obj2}?"
else: # above, under
new_question = f"Is the {obj1} above or under the {obj2}?"
return {
'index': sample['index'],
'image_base64': sample['image_base64'],
'question': new_question,
'answer': category,
'category': category,
}
def modify_distance_sample(sample: dict, rng: random.Random) -> dict:
"""Modify a distance relation sample (far/close)."""
category = sample['category']
answer_key = sample['answer'] # e.g. "C"
options = sample['options'] # {'A': 'table', 'B': 'towel', ...}
target_object = options[answer_key]
candidates = [v for k, v in options.items() if k != answer_key]
reference_object = rng.choice(candidates)
new_question = f"Compared to {reference_object}, is {target_object} far or close from you?"
return {
'index': sample['index'],
'image_base64': sample['image_base64'],
'question': new_question,
'answer': category,
'category': category,
}
def load_and_modify_data(
tsv_path: str,
samples_per_category: int = 200,
seed: int = 42
) -> Dict[str, List[dict]]:
"""
Load EmbSpatialBench data, modify questions to remove format confound.
"""
rng = random.Random(seed)
np.random.seed(seed)
df = pd.read_csv(tsv_path, sep='\t')
# Group by category
raw_grouped = defaultdict(list)
for _, row in df.iterrows():
category = row['category']
sample = {
'index': row['index'],
'image_base64': row['image'],
'question': row['question'],
'answer': row['answer'],
'category': category,
'options': {
'A': row['A'],
'B': row['B'],
'C': row['C'],
'D': row['D']
}
}
raw_grouped[category].append(sample)
# Sample and modify
modified_data = defaultdict(list)
stats = {'total': 0, 'success': 0, 'failed': 0}
for category in CATEGORY_ORDER:
samples = raw_grouped[category]
# Sample up to samples_per_category
if len(samples) > samples_per_category:
indices = np.random.choice(len(samples), samples_per_category, replace=False)
samples = [samples[i] for i in indices]
for sample in samples:
stats['total'] += 1
try:
if category in ['left', 'right', 'above', 'under']:
modified = modify_pairwise_sample(sample)
else: # far, close
modified = modify_distance_sample(sample, rng)
# Validate
assert modified['answer'] == modified['category']
modified_data[category].append(modified)
stats['success'] += 1
except Exception as e:
stats['failed'] += 1
logger.warning(f" Failed to modify sample {sample['index']}: {e}")
logger.info(f"Data modification: {stats['success']}/{stats['total']} success, {stats['failed']} failed")
for cat in CATEGORY_ORDER:
if cat in modified_data:
logger.info(f" {cat}: {len(modified_data[cat])} samples")
# Show first example
ex = modified_data[cat][0]
logger.info(f" Example Q: {ex['question']}")
logger.info(f" Example A: {ex['answer']}")
return dict(modified_data)
def decode_base64_image(base64_str: str) -> Image.Image:
"""Decode base64 string to PIL Image."""
image_data = base64.b64decode(base64_str)
return Image.open(BytesIO(image_data)).convert('RGB')
# ============================================================================
# Base Extractor
# ============================================================================
class BaseHiddenStateExtractor(ABC):
"""Base class for extracting hidden states from VLMs."""
def __init__(self, model_path: str, device: str = 'cuda', target_layers: List[int] = None):
self.model_path = model_path
self.device = device
self.hidden_states = {}
self.hooks = []
self._load_model()
num_layers = self._get_num_layers()
if target_layers is None:
self.target_layers = list(range(num_layers))
logger.info(f"Model has {num_layers} layers. Extracting ALL layers (0..{num_layers-1})")
else:
self.target_layers = target_layers
logger.info(f"Model has {num_layers} layers. Target layers: {self.target_layers}")
self._register_hooks()
def _register_hooks(self):
"""Register forward hooks on target layers."""
for layer_idx in self.target_layers:
module = self._get_layer_module(layer_idx)
if module is not None:
hook = module.register_forward_hook(self._make_hook(layer_idx))
self.hooks.append(hook)
logger.info(f" Registered hook on layer {layer_idx}")
def _make_hook(self, layer_idx: int):
"""Create a hook function for a specific layer."""
def hook_fn(module, input, output):
if isinstance(output, tuple):
hidden = output[0]
else:
hidden = output
# Last token pooling
last_token = hidden[:, -1, :].detach().cpu().float()
self.hidden_states[layer_idx] = last_token.squeeze(0)
return hook_fn
@abstractmethod
def _load_model(self):
pass
@abstractmethod
def _get_num_layers(self) -> int:
pass
@abstractmethod
def _get_layer_module(self, layer_idx: int):
pass
@abstractmethod
def extract(self, image: Image.Image, question: str) -> Dict[int, torch.Tensor]:
pass
def cleanup(self):
"""Remove hooks and free memory."""
for hook in self.hooks:
hook.remove()
self.hooks = []
if hasattr(self, 'model'):
del self.model
if hasattr(self, 'processor'):
del self.processor
torch.cuda.empty_cache()
# ============================================================================
# Molmo Extractor
# ============================================================================
class MolmoExtractor(BaseHiddenStateExtractor):
"""Hidden state extractor for Molmo models (native olmo format)."""
def _load_model(self):
config_path = os.path.join(self.model_path, "config.yaml")
checkpoint_path = os.path.join(self.model_path, "model.pt")
if os.path.exists(config_path) and os.path.exists(checkpoint_path):
self._load_native_model()
self.is_native = True
else:
self._load_hf_model()
self.is_native = False
def _load_native_model(self):
from olmo.config import ModelConfig
from olmo.model import Molmo as NativeMolmoModel
from olmo.data.model_preprocessor import MultiModalPreprocessor
from olmo.data.data_formatter import DataFormatter
_original_load = torch.load
def _unsafe_load_wrapper(*args, **kwargs):
if 'weights_only' not in kwargs:
kwargs['weights_only'] = False
return _original_load(*args, **kwargs)
torch.load = _unsafe_load_wrapper
config_path = os.path.join(self.model_path, "config.yaml")
checkpoint_path = os.path.join(self.model_path, "model.pt")
cfg = ModelConfig.load(config_path, key="model", validate_paths=False)
cfg.init_device = "cpu"
self.model = NativeMolmoModel(cfg)
state_dict = torch.load(checkpoint_path, map_location="cpu")
self.model.load_state_dict(state_dict)
self.model = self.model.to(self.device, dtype=torch.bfloat16).eval()
self.tokenizer = cfg.get_tokenizer()
v_cfg = cfg.vision_backbone
h, w = cfg.llm_patches_per_crop()
image_padding_mask = 2 if cfg.fix_image_padding else (1 if cfg.image_padding_embed else None)
class SafeDataFormatter(DataFormatter):
def get_system_prompt(self, style, for_inference, messages, rng=None):
if style is None:
style = "User"
return super().get_system_prompt(style, for_inference, messages, rng)
self.formatter = SafeDataFormatter(
prompt_templates=cfg.prompt_type,
message_format=cfg.message_formatting,
system_prompt=cfg.system_prompt_kind,
always_start_with_space=cfg.always_start_with_space,
default_inference_len=cfg.default_inference_len
)
self.preprocessor = MultiModalPreprocessor(
tokenizer=self.tokenizer,
normalize=str(v_cfg.image_model_type),
crop_mode=cfg.crop_mode,
max_crops=cfg.max_crops,
overlap_margins=cfg.overlap_margins,
resize=v_cfg.resize_mode,
use_col_tokens=cfg.use_col_tokens,
base_image_input_size=v_cfg.image_default_input_size,
image_pooling_w=cfg.image_pooling_w,
image_pooling_h=cfg.image_pooling_h,
image_token_length_w=w,
image_token_length_h=h,
image_patch_size=v_cfg.image_patch_size,
image_padding_mask=image_padding_mask,
pad_value=cfg.pad_value,
loss_token_weighting=cfg.multi_annotation_weighting,
)
logger.info(f"Loaded native Molmo model from {self.model_path}")
def _load_hf_model(self):
from transformers import AutoModelForCausalLM, AutoProcessor
self.model = AutoModelForCausalLM.from_pretrained(
self.model_path,
torch_dtype=torch.bfloat16,
trust_remote_code=True,
device_map=self.device
)
self.model.eval()
self.processor = AutoProcessor.from_pretrained(
self.model_path,
trust_remote_code=True
)
logger.info(f"Loaded HuggingFace Molmo model from {self.model_path}")
def _get_num_layers(self) -> int:
if self.is_native:
return len(self.model.transformer.blocks)
else:
if hasattr(self.model, 'model') and hasattr(self.model.model, 'transformer'):
return len(self.model.model.transformer.blocks)
return 32
def _get_layer_module(self, layer_idx: int):
if self.is_native:
return self.model.transformer.blocks[layer_idx]
else:
return self.model.model.transformer.blocks[layer_idx]
def extract(self, image: Image.Image, question: str) -> Dict[int, torch.Tensor]:
self.hidden_states = {}
if self.is_native:
example = {"messages": [question], "image": image}
messages, _ = self.formatter(example, is_training=False, for_inference=True, rng=np.random)
image_np = np.array(image)
batch = self.preprocessor(image_np, messages, is_training=False, require_image_features=True)
if 'input_ids' not in batch and 'input_tokens' in batch:
batch['input_ids'] = batch['input_tokens']
def to_tensor(x):
if isinstance(x, np.ndarray):
return torch.from_numpy(x)
return x
input_ids = to_tensor(batch['input_ids']).unsqueeze(0).to(self.device)
if input_ids.dtype not in [torch.long, torch.int64]:
input_ids = input_ids.long()
images_tensor = to_tensor(batch['images']).unsqueeze(0).to(self.device).to(dtype=torch.bfloat16)
image_masks = to_tensor(batch['image_masks']).unsqueeze(0).to(self.device).to(dtype=torch.bfloat16)
image_input_idx = to_tensor(batch['image_input_idx']).unsqueeze(0).to(self.device)
with torch.inference_mode():
with torch.autocast(device_type="cuda", enabled=True, dtype=torch.bfloat16):
_ = self.model(
input_ids=input_ids,
images=images_tensor,
image_masks=image_masks,
image_input_idx=image_input_idx,
)
else:
inputs = self.processor.process(images=[image], text=question)
processed_inputs = {}
for k, v in inputs.items():
v = v.to(self.device).unsqueeze(0)
if v.dtype == torch.float32:
v = v.to(dtype=torch.bfloat16)
processed_inputs[k] = v
with torch.no_grad():
_ = self.model(**processed_inputs)
return self.hidden_states.copy()
# ============================================================================
# NVILA Extractor
# ============================================================================
class NVILAExtractor(BaseHiddenStateExtractor):
"""Hidden state extractor for NVILA models."""
def _load_model(self):
original_sys_path = sys.path.copy()
sys.path = [p for p in sys.path if 'RoboRefer' not in p]
modules_to_remove = [key for key in list(sys.modules.keys()) if 'llava' in key.lower()]
removed_modules = {}
for mod in modules_to_remove:
removed_modules[mod] = sys.modules.pop(mod)
try:
import llava
from llava.media import Image as LLaVAImage
from llava import conversation as clib
except Exception as err:
sys.path = original_sys_path
for mod, module in removed_modules.items():
sys.modules[mod] = module
raise RuntimeError(f"Failed to import llava: {err}")
sys.path = original_sys_path
self.LLaVAImage = LLaVAImage
self.clib = clib
self.model = llava.load(self.model_path, model_base=None)
self._find_llm_backbone()
logger.info(f"Loaded NVILA model from {self.model_path}")
def _find_llm_backbone(self):
"""Find the LLM backbone module for hook registration."""
candidates = []
if hasattr(self.model, 'llm'):
if hasattr(self.model.llm, 'model') and hasattr(self.model.llm.model, 'layers'):
candidates.append(('model.llm.model.layers', self.model.llm.model.layers))
if hasattr(self.model.llm, 'layers'):
candidates.append(('model.llm.layers', self.model.llm.layers))
if hasattr(self.model, 'model'):
if hasattr(self.model.model, 'model') and hasattr(self.model.model.model, 'layers'):
candidates.append(('model.model.model.layers', self.model.model.model.layers))
if hasattr(self.model.model, 'layers'):
candidates.append(('model.model.layers', self.model.model.layers))
for name, module in self.model.named_modules():
if name.endswith('.layers') and hasattr(module, '__len__') and len(module) > 0:
candidates.append((name, module))
if candidates:
path, layers = candidates[0]
logger.info(f"Found LLM layers at: {path} (num_layers={len(layers)})")
self.llm_backbone = layers
self.layers_path = path
else:
logger.error("Could not find transformer layers in model!")
for name, _ in list(self.model.named_modules())[:20]:
logger.info(f" {name}")
raise ValueError("Could not locate transformer layers in NVILA model")
def _get_num_layers(self) -> int:
if hasattr(self, 'llm_backbone') and hasattr(self.llm_backbone, '__len__'):
return len(self.llm_backbone)
return 24
def _get_layer_module(self, layer_idx: int):
if hasattr(self, 'llm_backbone') and hasattr(self.llm_backbone, '__getitem__'):
module = self.llm_backbone[layer_idx]
logger.info(f" Accessing layer {layer_idx}: {type(module).__name__}")
return module
logger.error(f"Cannot access layer {layer_idx} - llm_backbone not properly initialized")
return None
def extract(self, image: Image.Image, question: str) -> Dict[int, torch.Tensor]:
self.hidden_states = {}
import tempfile
with tempfile.NamedTemporaryFile(suffix='.png', delete=False) as f:
temp_path = f.name
image.save(temp_path)
try:
prompt = [self.LLaVAImage(temp_path), question]
from transformers import GenerationConfig
gen_config = GenerationConfig(max_new_tokens=1, do_sample=False)
_ = self.model.generate_content(prompt, generation_config=gen_config)
finally:
os.unlink(temp_path)
return self.hidden_states.copy()
# ============================================================================
# RoboRefer Extractor (NVILA-based)
# ============================================================================
class RoboReferExtractor(NVILAExtractor):
"""Hidden state extractor for RoboRefer models (NVILA-based, different llava path)."""
ROBOREFER_PATH = '/data/shared/Qwen/RoboRefer'
def _load_model(self):
original_sys_path = sys.path.copy()
# Add RoboRefer path (opposite of NVILA which removes it)
if self.ROBOREFER_PATH not in sys.path:
sys.path.insert(0, self.ROBOREFER_PATH)
# Clear any existing llava modules to avoid conflicts
modules_to_remove = [key for key in list(sys.modules.keys()) if 'llava' in key.lower()]
removed_modules = {}
for mod in modules_to_remove:
removed_modules[mod] = sys.modules.pop(mod)
try:
import llava
from llava.media import Image as LLaVAImage
from llava import conversation as clib
except Exception as err:
sys.path = original_sys_path
for mod, module in removed_modules.items():
sys.modules[mod] = module
raise RuntimeError(f"Failed to import RoboRefer llava: {err}")
sys.path = original_sys_path
self.LLaVAImage = LLaVAImage
self.clib = clib
self.model = llava.load(self.model_path, model_base=None)
self._find_llm_backbone()
logger.info(f"Loaded RoboRefer model from {self.model_path}")
# ============================================================================
# Qwen2.5-VL Extractor
# ============================================================================
class Qwen25VLExtractor(BaseHiddenStateExtractor):
"""Hidden state extractor for Qwen2.5-VL models."""
BASE_MODEL = "Qwen/Qwen2.5-VL-3B-Instruct"
def _load_model(self):
from transformers import Qwen2_5_VLForConditionalGeneration, AutoProcessor
try:
self.model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
self.model_path,
torch_dtype=torch.bfloat16,
device_map=self.device
)
except ImportError:
logger.info("accelerate not available, loading model without device_map...")
self.model = Qwen2_5_VLForConditionalGeneration.from_pretrained(
self.model_path,
torch_dtype=torch.bfloat16,
)
self.model = self.model.to(self.device)
self.model.eval()
if self.model_path.startswith('/'):
logger.info(f"Fine-tuned model detected, loading processor from base model: {self.BASE_MODEL}")
self.processor = AutoProcessor.from_pretrained(self.BASE_MODEL)
else:
self.processor = AutoProcessor.from_pretrained(self.model_path)
logger.info(f"Loaded Qwen2.5-VL model from {self.model_path}")
def _get_num_layers(self) -> int:
return len(self.model.model.layers)
def _get_layer_module(self, layer_idx: int):
return self.model.model.layers[layer_idx]
def extract(self, image: Image.Image, question: str) -> Dict[int, torch.Tensor]:
self.hidden_states = {}
messages = [
{
"role": "user",
"content": [
{"type": "image", "image": image},
{"type": "text", "text": question}
]
}
]
text = self.processor.apply_chat_template(
messages, tokenize=False, add_generation_prompt=True
)
from qwen_vl_utils import process_vision_info
image_inputs, video_inputs = process_vision_info(messages)
inputs = self.processor(
text=[text],
images=image_inputs,
videos=video_inputs,
padding=True,
return_tensors="pt"
)
inputs = inputs.to(self.device)
with torch.no_grad():
_ = self.model(**inputs)
return self.hidden_states.copy()
# ============================================================================
# Factory Function
# ============================================================================
def get_extractor(model_type: str, model_path: str, scale: str = None, **kwargs) -> BaseHiddenStateExtractor:
# RoboRefer uses NVILA architecture but needs different llava import path
if model_type == 'nvila' and scale == 'roborefer':
return RoboReferExtractor(model_path, **kwargs)
extractors = {
'molmo': MolmoExtractor,
'nvila': NVILAExtractor,
'qwen': Qwen25VLExtractor,
}
if model_type not in extractors:
raise ValueError(f"Unknown model type: {model_type}. Available: {list(extractors.keys())}")
return extractors[model_type](model_path, **kwargs)
# ============================================================================
# Analysis Functions
# ============================================================================
def extract_all_layer_representations(
extractor: BaseHiddenStateExtractor,
data: Dict[str, List[dict]],
) -> Dict[int, Dict[str, np.ndarray]]:
"""Extract average hidden state representations for ALL target layers at once.
Returns:
Dict mapping layer_idx -> {category -> avg_vector}
"""
# category_states[layer_idx][category] = list of vectors
category_states = defaultdict(lambda: defaultdict(list))
for category in CATEGORY_ORDER:
if category not in data:
continue
samples = data[category]
logger.info(f"Processing category: {category}")
success_count = 0
for sample in tqdm(samples, desc=f" {category}"):
try:
image = decode_base64_image(sample['image_base64'])
hidden_states = extractor.extract(image, sample['question'])
for layer_idx in extractor.target_layers:
if layer_idx in hidden_states:
state = hidden_states[layer_idx].numpy().flatten()
if state.size > 0:
category_states[layer_idx][category].append(state)
if any(l in hidden_states for l in extractor.target_layers):
success_count += 1
else:
logger.warning(f" No target layers found. Available: {list(hidden_states.keys())}")
except Exception as e:
logger.warning(f" Error processing sample {sample['index']}: {e}")
continue
logger.info(f" {category}: Successfully extracted {success_count}/{len(samples)} samples")
# Average per category per layer
result = {}
for layer_idx in extractor.target_layers:
category_avg = {}
for category, states in category_states[layer_idx].items():
if states:
category_avg[category] = np.mean(states, axis=0)
if category_avg:
result[layer_idx] = category_avg
logger.info(f" Layer {layer_idx}: {len(category_avg)} categories collected")
else:
logger.error(f" Layer {layer_idx}: No states collected!")
if not result:
raise ValueError("No representations were extracted!")
return result
def compute_similarity_matrix(
representations: Dict[str, np.ndarray]
) -> pd.DataFrame:
"""Compute pairwise cosine similarity with fixed category order."""
available = [c for c in CATEGORY_ORDER if c in representations]
vectors = np.array([representations[cat] for cat in available])
sim_matrix = cosine_similarity(vectors)
return pd.DataFrame(sim_matrix, index=available, columns=available)
def analyze_hypothesis(sim_df: pd.DataFrame, model_name: str) -> dict:
"""Analyze the similarity matrix to test Hypothesis 4."""
results = {'model': model_name}
pairs_to_check = {
'above_far': ('above', 'far'),
'under_close': ('under', 'close'),
'left_right': ('left', 'right'),
}
for pair_name, (cat1, cat2) in pairs_to_check.items():
if cat1 in sim_df.index and cat2 in sim_df.columns:
sim = sim_df.loc[cat1, cat2]
results[f'sim_{pair_name}'] = sim
logger.info(f" {pair_name}: sim({cat1}, {cat2}) = {sim:.4f}")
else:
results[f'sim_{pair_name}'] = None
if results.get('sim_above_far') and results.get('sim_left_right'):
results['diff_above_far_vs_left_right'] = results['sim_above_far'] - results['sim_left_right']
if results.get('sim_under_close') and results.get('sim_left_right'):
results['diff_under_close_vs_left_right'] = results['sim_under_close'] - results['sim_left_right']
return results
# ============================================================================
# Visualization
# ============================================================================
def plot_similarity_heatmap(sim_df: pd.DataFrame, title: str, save_path: str):
"""Plot and save similarity heatmap with fixed category order."""
plt.figure(figsize=(10, 8))
available_order = [c for c in CATEGORY_ORDER if c in sim_df.index]
sim_df_ordered = sim_df.loc[available_order, available_order]
sns.heatmap(
sim_df_ordered,
annot=True,
fmt='.4f',
cmap='RdYlBu_r',
center=0.5,
vmin=0,
vmax=1,
square=True,
linewidths=0.5,
cbar_kws={'label': 'Cosine Similarity'}
)
plt.title(title, fontsize=14, fontweight='bold')
plt.tight_layout()
plt.savefig(save_path, dpi=300, bbox_inches='tight')
plt.close()
logger.info(f"Saved heatmap: {save_path}")
def plot_comparison(results_list: List[dict], save_path: str):
"""Plot comparison of similarity pairs across models."""
pairs = ['sim_above_far', 'sim_under_close', 'sim_left_right']
pair_labels = ['above-far', 'under-close', 'left-right']
fig, ax = plt.subplots(figsize=(12, 6))
x = np.arange(len(pairs))
width = 0.8 / len(results_list)
for i, result in enumerate(results_list):
model = result['model']
values = [result.get(p, 0) or 0 for p in pairs]
offset = (i - len(results_list) / 2 + 0.5) * width
bars = ax.bar(x + offset, values, width, label=model)
for bar, val in zip(bars, values):
if val:
ax.annotate(
f'{val:.3f}',
xy=(bar.get_x() + bar.get_width() / 2, bar.get_height()),
xytext=(0, 3),
textcoords='offset points',
ha='center',
va='bottom',
fontsize=8
)
ax.set_ylabel('Cosine Similarity')
ax.set_title('Spatial Concept Similarity Comparison (Modified Format)\n(Hypothesis 4: above-far & under-close should be > left-right for vanilla)')
ax.set_xticks(x)
ax.set_xticklabels(pair_labels)
ax.legend(loc='upper right', fontsize=8)
ax.set_ylim(0, 1)
ax.axhline(y=0.5, color='gray', linestyle='--', alpha=0.5)
plt.tight_layout()
plt.savefig(save_path, dpi=300, bbox_inches='tight')
plt.close()
logger.info(f"Saved comparison plot: {save_path}")
def _extract_pair_trajectory(
all_layer_sims: Dict[int, pd.DataFrame],
cat1: str, cat2: str,
) -> Tuple[List[int], List[float]]:
"""Extract similarity values for a pair across all layers."""
layers = sorted(all_layer_sims.keys())
valid_layers = []
values = []
for l in layers:
df = all_layer_sims[l]
if cat1 in df.index and cat2 in df.columns:
valid_layers.append(l)
values.append(df.loc[cat1, cat2])
return valid_layers, values
def get_representative_layers(all_layers: List[int], n: int = 5) -> List[int]:
"""Pick n representative layers (evenly spaced) for heatmap output."""
if len(all_layers) <= n:
return list(all_layers)
indices = np.linspace(0, len(all_layers) - 1, n, dtype=int)
return [all_layers[i] for i in indices]
def plot_similarity_trajectories(
all_layer_sims: Dict[int, pd.DataFrame],
title: str,
save_path: str,
):
"""Plot similarity of key category pairs across all layers.
Left panel: absolute cosine similarity per pair across layers.
Right panel: difference from left-right baseline (positive = more similar than L-R).
"""
fig, axes = plt.subplots(1, 2, figsize=(20, 7))
# --- Left panel: absolute similarity ---
ax = axes[0]
for cat1, cat2, label, color in TRAJECTORY_PAIRS['hypothesis']:
layers, vals = _extract_pair_trajectory(all_layer_sims, cat1, cat2)
ax.plot(layers, vals, '-', color=color, label=label, linewidth=2.5, markersize=0)
for cat1, cat2, label, color in TRAJECTORY_PAIRS['within_axis']:
layers, vals = _extract_pair_trajectory(all_layer_sims, cat1, cat2)
ax.plot(layers, vals, '--', color=color, label=label, linewidth=1.8, markersize=0)
for cat1, cat2, label, color in TRAJECTORY_PAIRS['counter_hypothesis']:
layers, vals = _extract_pair_trajectory(all_layer_sims, cat1, cat2)
ax.plot(layers, vals, ':', color=color, label=label, linewidth=1.5, alpha=0.8)
ax.set_xlabel('Layer Index', fontsize=12)
ax.set_ylabel('Cosine Similarity', fontsize=12)
ax.set_title(f'{title}\nPairwise Similarity Across Layers', fontsize=13)
ax.legend(fontsize=9, loc='best')
ax.grid(True, alpha=0.3)
# --- Right panel: difference from left-right ---
ax = axes[1]
lr_layers, lr_vals = _extract_pair_trajectory(all_layer_sims, 'left', 'right')
lr_dict = dict(zip(lr_layers, lr_vals))
for cat1, cat2, label, color in TRAJECTORY_PAIRS['hypothesis']:
layers, vals = _extract_pair_trajectory(all_layer_sims, cat1, cat2)
diffs = [v - lr_dict.get(l, 0) for l, v in zip(layers, vals)]
ax.plot(layers, diffs, '-', color=color, label=f'{label} - left-right',
linewidth=2.5, markersize=0)
for cat1, cat2, label, color in TRAJECTORY_PAIRS['counter_hypothesis']:
layers, vals = _extract_pair_trajectory(all_layer_sims, cat1, cat2)
diffs = [v - lr_dict.get(l, 0) for l, v in zip(layers, vals)]
ax.plot(layers, diffs, ':', color=color, label=f'{label} - left-right',
linewidth=1.5, alpha=0.8)
# Also show above-under and far-close as references
for cat1, cat2, label, color in TRAJECTORY_PAIRS['within_axis']:
if label == 'left-right':
continue
layers, vals = _extract_pair_trajectory(all_layer_sims, cat1, cat2)
diffs = [v - lr_dict.get(l, 0) for l, v in zip(layers, vals)]
ax.plot(layers, diffs, '--', color=color, label=f'{label} - left-right',
linewidth=1.5, alpha=0.7)
ax.axhline(y=0, color='gray', linestyle='-', linewidth=1, alpha=0.5)
ax.set_xlabel('Layer Index', fontsize=12)
ax.set_ylabel('Similarity Difference (pair - left-right)', fontsize=12)
ax.set_title(f'{title}\nRelative to Left-Right Baseline', fontsize=13)
ax.legend(fontsize=8, loc='best')
ax.grid(True, alpha=0.3)
plt.tight_layout()
plt.savefig(save_path, dpi=300, bbox_inches='tight')
plt.close()
logger.info(f"Saved trajectory plot: {save_path}")
def plot_cross_scale_trajectories(
cross_scale_data: Dict[str, Dict[int, pd.DataFrame]],
model_type: str,
save_path: str,
):
"""Compare layer-wise trajectories across training scales.
3 columns: above-far, under-close, left-right (control).
Each subplot shows one line per scale.
"""
pairs = [
('above', 'far', 'above-far (hypothesis)'),
('under', 'close', 'under-close (hypothesis)'),
('left', 'right', 'left-right (control)'),
]
fig, axes = plt.subplots(1, len(pairs), figsize=(7 * len(pairs), 6))
if len(pairs) == 1:
axes = [axes]
for idx, (cat1, cat2, label) in enumerate(pairs):
ax = axes[idx]
for scale in ['vanilla', '80k', '400k', '800k', '2m', 'roborefer']:
if scale not in cross_scale_data:
continue
layer_sims = cross_scale_data[scale]
layers, vals = _extract_pair_trajectory(layer_sims, cat1, cat2)
color = SCALE_COLORS.get(scale, 'gray')
ax.plot(layers, vals, '-', color=color, label=scale, linewidth=2, markersize=0)
ax.set_xlabel('Layer Index', fontsize=12)
ax.set_ylabel('Cosine Similarity', fontsize=12)
ax.set_title(label, fontsize=13, fontweight='bold')
ax.legend(fontsize=10)
ax.grid(True, alpha=0.3)
fig.suptitle(
f'{model_type.upper()} - Similarity Trajectory Across Scales',
fontsize=15, fontweight='bold', y=1.02
)
plt.tight_layout()
plt.savefig(save_path, dpi=300, bbox_inches='tight')
plt.close()
logger.info(f"Saved cross-scale trajectory: {save_path}")
def plot_similarity_evolution_heatmap(
cross_scale_data: Dict[str, Dict[int, pd.DataFrame]],
model_type: str,
save_path: str,
):
"""2D heatmap: x=layer, y=scale, color=similarity for each hypothesis pair.
Gives a bird's-eye view of how both network depth and training data scale
affect the similarity between hypothesis-relevant category pairs.
"""
pairs = [
('above', 'far', 'above-far'),
('under', 'close', 'under-close'),
('left', 'right', 'left-right'),
('above', 'under', 'above-under'),
('far', 'close', 'far-close'),
]
scale_order = ['vanilla', '80k', '400k', '800k', '2m', 'roborefer']
available_scales = [s for s in scale_order if s in cross_scale_data]
# Determine layer range from first available scale
first_scale = available_scales[0]
all_layers = sorted(cross_scale_data[first_scale].keys())
fig, axes = plt.subplots(len(pairs), 1, figsize=(max(14, len(all_layers) * 0.5), 3 * len(pairs)))
if len(pairs) == 1:
axes = [axes]
for idx, (cat1, cat2, label) in enumerate(pairs):
ax = axes[idx]
# Build matrix: rows=scales, cols=layers
matrix = np.full((len(available_scales), len(all_layers)), np.nan)
for si, scale in enumerate(available_scales):
layer_sims = cross_scale_data[scale]
for li, layer in enumerate(all_layers):
if layer in layer_sims:
df = layer_sims[layer]
if cat1 in df.index and cat2 in df.columns:
matrix[si, li] = df.loc[cat1, cat2]
im = ax.imshow(matrix, aspect='auto', cmap='RdYlBu_r', vmin=0.5, vmax=1.0)
ax.set_yticks(range(len(available_scales)))
ax.set_yticklabels(available_scales, fontsize=10)
# X-axis: show every Nth layer label to avoid crowding
step = max(1, len(all_layers) // 15)
ax.set_xticks(range(0, len(all_layers), step))
ax.set_xticklabels([str(all_layers[i]) for i in range(0, len(all_layers), step)], fontsize=8)
ax.set_title(label, fontsize=12, fontweight='bold')
ax.set_xlabel('Layer Index', fontsize=10)
fig.colorbar(im, ax=ax, label='Cosine Similarity', shrink=0.8)
fig.suptitle(
f'{model_type.upper()} - Similarity Evolution (Layer x Scale)',
fontsize=15, fontweight='bold', y=1.01
)
plt.tight_layout()
plt.savefig(save_path, dpi=300, bbox_inches='tight')
plt.close()
logger.info(f"Saved evolution heatmap: {save_path}")
# ============================================================================
# Model Configurations
# ============================================================================
MODEL_CONFIGS = {
'molmo': {
'vanilla': 'allenai/Molmo-7B-O-0924',
'80k': '/data/shared/Qwen/molmo/outputs/data_scale_exp_80k/unshared',
'400k': '/data/shared/Qwen/molmo/outputs/data_scale_exp_400k/unshared',
'800k': '/data/shared/Qwen/molmo/outputs/data_scale_exp_800k/unshared',
'2m': '/data/shared/Qwen/molmo/outputs/data_scale_exp_2m/unshared',
},
'nvila': {
'vanilla': '/data/shared/Qwen/mydisk/NVILA-Lite-2B',
'80k': '/data/shared/Qwen/mydisk/output/DATA/NVILA-Lite-2B-DATA_SCALE_EXP_80K-20251108_180221',
'400k': '/data/shared/Qwen/mydisk/output/DATA/NVILA-Lite-2B-DATA_SCALE_EXP_400K-20251108_180221',
'800k': '/data/shared/Qwen/mydisk/output/DATA/NVILA-Lite-2B-DATA_SCALE_EXP_800K-20251108_180221',
'2m': '/data/shared/Qwen/mydisk/output/DATA/NVILA-Lite-2B-DATA_SCALE_EXP_2M-20260205_003632',
'roborefer': '/data/shared/Qwen/mydisk/RoboRefer_model',
},
'qwen': {
'vanilla': 'Qwen/Qwen2.5-VL-3B-Instruct',
'80k': '/data/shared/Qwen/mydisk/output/Qwen/Qwen2.5-VL-3B-Instruct-data_scale_exp_80k-20251114_120221',
'400k': '/data/shared/Qwen/mydisk/output/Qwen/Qwen2.5-VL-3B-Instruct-data_scale_exp_400k-20251114_120221',
'800k': '/data/shared/Qwen/mydisk/output/Qwen/Qwen2.5-VL-3B-Instruct-data_scale_exp_800k-20251114_120221',
'2m': '/data/shared/Qwen/mydisk/output/Qwen/Qwen2.5-VL-3B-Instruct-data_scale_exp_2m-20260109_120517',
},
}
# ============================================================================
# Main
# ============================================================================
def main():
parser = argparse.ArgumentParser(description='Experiment 2-A (Modified): Embedding Space Analysis')
parser.add_argument('--data_path', type=str,
default='/data/shared/Qwen/EmbSpatial-Bench/EmbSpatial-Bench.tsv')
parser.add_argument('--model_type', type=str, required=True,
choices=['molmo', 'nvila', 'qwen'])
parser.add_argument('--scales', type=str, nargs='+',
default=['vanilla', '80k', '400k', '800k', '2m'])
parser.add_argument('--output_dir', type=str,
default='/data/shared/Qwen/experiments/exp2a_modified/results_all_layers')
parser.add_argument('--samples_per_category', type=int, default=200)
parser.add_argument('--device', type=str, default='cuda')
parser.add_argument('--seed', type=int, default=42)
args = parser.parse_args()
# Auto-include roborefer for nvila if not already specified
if args.model_type == 'nvila' and 'roborefer' not in args.scales:
args.scales.append('roborefer')
# Set random seed
np.random.seed(args.seed)
torch.manual_seed(args.seed)
random.seed(args.seed)
# Create output directory
output_dir = os.path.join(args.output_dir, args.model_type)
os.makedirs(output_dir, exist_ok=True)
# Load and modify data
logger.info("\n=== Loading & Modifying EmbSpatialBench Data ===")
data = load_and_modify_data(args.data_path, args.samples_per_category, args.seed)
results_list = []
cross_scale_data = {} # scale -> {layer_idx -> sim_df}
model_configs = MODEL_CONFIGS[args.model_type]
for scale in args.scales:
if scale not in model_configs:
logger.warning(f"Scale {scale} not available for {args.model_type}, skipping...")
continue
model_path = model_configs[scale]
if not os.path.exists(model_path) and not model_path.startswith('Qwen/') and not model_path.startswith('allenai/'):
logger.warning(f"Model path not found: {model_path}, skipping...")
continue
logger.info(f"\n=== Processing {args.model_type} - {scale} ===")
logger.info(f"Model path: {model_path}")
try:
extractor = get_extractor(
args.model_type,
model_path,
scale=scale,
device=args.device,
)
num_layers = len(extractor.target_layers)
# Extract representations for ALL layers in one pass
all_layer_reps = extract_all_layer_representations(extractor, data)
# Compute similarity matrices for all layers
scale_sims = {}
model_name = f"{args.model_type}_{scale}"
for layer_idx in sorted(all_layer_reps.keys()):
sim_df = compute_similarity_matrix(all_layer_reps[layer_idx])
scale_sims[layer_idx] = sim_df
results = analyze_hypothesis(sim_df, model_name)
results['layer_idx'] = layer_idx
results_list.append(results)
# Save CSV for every layer
sim_df.to_csv(os.path.join(output_dir, f'similarity_{scale}_L{layer_idx}.csv'))
cross_scale_data[scale] = scale_sims
logger.info(f" Computed similarity matrices for {len(scale_sims)} layers")
# Save heatmaps for representative layers only (to avoid hundreds of files)
rep_layers = get_representative_layers(sorted(scale_sims.keys()))
logger.info(f" Saving heatmaps for representative layers: {rep_layers}")
for layer_idx in rep_layers:
sim_df = scale_sims[layer_idx]
plot_similarity_heatmap(
sim_df,
f'{args.model_type.upper()} ({scale}) - Layer {layer_idx}/{num_layers-1}',
os.path.join(output_dir, f'heatmap_{scale}_L{layer_idx}.png')
)
# Per-scale trajectory plot
plot_similarity_trajectories(
scale_sims,
f'{args.model_type.upper()} ({scale})',
os.path.join(output_dir, f'trajectory_{scale}.png')
)
extractor.cleanup()
except Exception as e:
logger.error(f"Failed to process {args.model_type} - {scale}: {e}")
import traceback
traceback.print_exc()
continue
# Cross-scale comparison plots
if len(cross_scale_data) > 1:
plot_cross_scale_trajectories(
cross_scale_data,
args.model_type,
os.path.join(output_dir, 'trajectory_cross_scale.png')
)
plot_similarity_evolution_heatmap(
cross_scale_data,
args.model_type,
os.path.join(output_dir, 'evolution_heatmap.png')
)
# Save results summary
if results_list:
results_df = pd.DataFrame(results_list)
results_df.to_csv(os.path.join(output_dir, 'results_summary.csv'), index=False)
logger.info("\n=== Analysis Complete ===")
logger.info(f"Results saved to: {output_dir}")
logger.info(f"Total: {len(results_list)} (layer, scale) combinations across {len(cross_scale_data)} scales")
if __name__ == '__main__':
main()
|