Add files using upload-large-folder tool

models/qwen3_vl_compat.py

@@ -0,0 +1,49 @@
+"""
+Helpers for Qwen3-VL (transformers): vision lives on model.model.visual,
+and ViT+merger returns BaseModelOutputWithDeepstackFeatures.pooler_output.
+"""
+
+from __future__ import annotations
+
+import torch
+
+
+def get_visual_module(model: torch.nn.Module):
+    inner = getattr(model, "model", None)
+    if inner is not None and hasattr(inner, "visual"):
+        return inner.visual
+    return model.visual
+
+
+def set_visual_module(model: torch.nn.Module, value) -> None:
+    inner = getattr(model, "model", None)
+    if inner is not None and hasattr(inner, "visual"):
+        inner.visual = value
+        return
+    model.visual = value
+
+
+def merged_spatial_grid(image_grid_thw: torch.Tensor, spatial_merge_size: int) -> torch.Tensor:
+    """Processor grid (T,H,W) -> LLM-side grid after spatial merge on H,W."""
+    g = image_grid_thw.clone()
+    g[:, 1] = g[:, 1] // spatial_merge_size
+    g[:, 2] = g[:, 2] // spatial_merge_size
+    return g
+
+
+def pack_image_features(model, pixel_values: torch.Tensor, image_grid_thw: torch.Tensor):
+    """
+    Returns flat merger embeddings [total_tokens, D] and LLM grid [num_imgs, 3]
+    consistent with OpticalCompressor token counts.
+    """
+    inner = model.model if hasattr(model, "model") else model
+    out = inner.get_image_features(pixel_values, image_grid_thw)
+    parts = out.pooler_output
+    if isinstance(parts, (list, tuple)):
+        packed = torch.cat(parts, dim=0)
+    else:
+        packed = parts
+    vis = get_visual_module(model)
+    grid_llm = merged_spatial_grid(image_grid_thw, vis.spatial_merge_size)
+    return packed, grid_llm
+

results/benchmark/all_clip_scores.json

@@ -0,0 +1,56 @@
+{
+  "efficientui_prune60": {
+    "n": 50,
+    "avg_clip": 0.7523,
+    "min_clip": 0.4951,
+    "max_clip": 0.9579
+  },
+  "efficientui_prune80": {
+    "n": 50,
+    "avg_clip": 0.738,
+    "min_clip": 0.4485,
+    "max_clip": 0.9265
+  },
+  "qwen3_full": {
+    "n": 50,
+    "avg_clip": 0.7563,
+    "min_clip": 0.0,
+    "max_clip": 0.9357
+  },
+  "qwen3_res_1003520": {
+    "n": 50,
+    "avg_clip": 0.775,
+    "min_clip": 0.4805,
+    "max_clip": 0.9281
+  },
+  "qwen3_res_230400": {
+    "n": 50,
+    "avg_clip": 0.7768,
+    "min_clip": 0.4308,
+    "max_clip": 0.9515
+  },
+  "uipress_256": {
+    "n": 50,
+    "avg_clip": 0.7232,
+    "min_clip": 0.4485,
+    "max_clip": 0.8768
+  },
+  "visionzip_128": {
+    "n": 50,
+    "avg_clip": 0.7245,
+    "min_clip": 0.4485,
+    "max_clip": 0.8763
+  },
+  "visionzip_256": {
+    "n": 50,
+    "avg_clip": 0.7333,
+    "min_clip": 0.4485,
+    "max_clip": 0.881
+  },
+  "visionzip_64": {
+    "n": 50,
+    "avg_clip": 0.7197,
+    "min_clip": 0.4485,
+    "max_clip": 0.8768
+  }
+}

results/benchmark/bootstrap_ci.json

@@ -0,0 +1,74 @@
+{
+  "efficientui_prune60": {
+    "mean": 0.752298,
+    "ci_lower": 0.7238,
+    "ci_upper": 0.7804,
+    "ci_width": 0.0567,
+    "std": 0.1015,
+    "n": 50
+  },
+  "efficientui_prune80": {
+    "mean": 0.737994,
+    "ci_lower": 0.7076,
+    "ci_upper": 0.7678,
+    "ci_width": 0.0602,
+    "std": 0.1097,
+    "n": 50
+  },
+  "qwen3_full": {
+    "mean": 0.75634,
+    "ci_lower": 0.7099,
+    "ci_upper": 0.7941,
+    "ci_width": 0.0842,
+    "std": 0.1522,
+    "n": 50
+  },
+  "qwen3_res_1003520": {
+    "mean": 0.775018,
+    "ci_lower": 0.7453,
+    "ci_upper": 0.8028,
+    "ci_width": 0.0575,
+    "std": 0.1029,
+    "n": 50
+  },
+  "qwen3_res_230400": {
+    "mean": 0.7767660000000001,
+    "ci_lower": 0.7486,
+    "ci_upper": 0.8036,
+    "ci_width": 0.055,
+    "std": 0.1006,
+    "n": 50
+  },
+  "uipress_256": {
+    "mean": 0.7232,
+    "ci_lower": 0.695,
+    "ci_upper": 0.7511,
+    "ci_width": 0.0561,
+    "std": 0.1022,
+    "n": 50
+  },
+  "visionzip_128": {
+    "mean": 0.724536,
+    "ci_lower": 0.6958,
+    "ci_upper": 0.7524,
+    "ci_width": 0.0567,
+    "std": 0.1033,
+    "n": 50
+  },
+  "visionzip_256": {
+    "mean": 0.733302,
+    "ci_lower": 0.705,
+    "ci_upper": 0.7612,
+    "ci_width": 0.0562,
+    "std": 0.103,
+    "n": 50
+  },
+  "visionzip_64": {
+    "mean": 0.7196839999999999,
+    "ci_lower": 0.6927,
+    "ci_upper": 0.7457,
+    "ci_width": 0.053,
+    "std": 0.0976,
+    "n": 50
+  }
+}