Add files using upload-large-folder tool

runs_laq_lt/eval/cocolt_1ep_hires/summary.txt

@@ -0,0 +1,112 @@
+==============================================================================
+EVAL SUMMARY  |  cocolt_1ep_hires
+==============================================================================
+
+Metric: mean-over-IoU (fastevaluate). BPS = boxes/second, batch 1, 1 GPU.
+fallback% = switch_to_ar / forward_steps  (fraction of block-decisions; paper implies ~7-8%)
+
+      profile    mode |     F1      P      R  F1@.5 |    BPS  steps  boxes fallb%
+---------------------------------------------------------------------------------
+  paper_seed0    slow |   53.2   55.7   50.8   72.2 |   2.71   67.5   7.06    0.0
+  paper_seed0    fast |   51.5   61.5   44.4   66.9 |   8.82   15.1   6.88    0.0
+  paper_seed0  hybrid |   52.7   62.4   45.6   68.2 |   8.00   18.2   6.97    4.8
+
+  paper_seed1    slow |   53.4   56.0   51.0   72.2 |   2.71   67.3   7.05    0.0
+  paper_seed1    fast |   51.5   61.4   44.4   66.9 |   8.75   15.1   6.88    0.0
+  paper_seed1  hybrid |   52.5   62.2   45.4   68.0 |   8.02   18.1   6.95    4.8
+
+ paper_greedy    slow |   55.9   58.3   53.8   73.2 |   2.78   69.2   7.25    0.0
+ paper_greedy    fast |   53.5   58.4   49.3   70.2 |   9.17   15.0   7.00    0.0
+ paper_greedy  hybrid |   54.6   59.1   50.7   71.3 |   8.26   18.6   7.15    5.3
+
+       legacy    slow |   55.9   58.3   53.8   73.2 |   2.77   69.2   7.25    0.0
+       legacy    fast |   53.5   58.4   49.4   70.2 |   9.20   15.1   7.01    0.0
+       legacy  hybrid |   54.6   59.0   50.7   71.3 |   8.22   18.6   7.15    5.2
+
+paper is STOCHASTIC (nucleus sampling). Mean +/- half-spread over 2 seeds:
+---------------------------------------------------------------------------------
+ paper (mean)    slow | F1   53.3+/-0.1   BPS    2.7+/-0.0   fallback%    0.0+/-0.0
+ paper (mean)    fast | F1   51.5+/-0.0   BPS    8.8+/-0.0   fallback%    0.0+/-0.0
+ paper (mean)  hybrid | F1   52.6+/-0.1   BPS    8.0+/-0.0   fallback%    4.8+/-0.0
+
+  -> a gap smaller than the spread is NOT evidence. Use paper_greedy
+     (deterministic) for tight comparisons.
+
+INTEGRITY  legacy/slow (55.93) vs paper_greedy/slow (55.93) -> OK (identical, as required)
+
+B1 READ-OUT  hybrid fallback rate (switch_to_ar/forward_steps): legacy 5.2%  vs  paper 4.8%   (paper ~7-8%; the RELATIVE legacy>paper gap is the primary signal)
+  The 0.9 abnormality threshold is calibrated on the SHARPENED (T=0.7 +
+  top-p) distribution. Greedy's flatter softmax trips it far more often, so
+  legacy should be markedly higher. If they match, B1 has not taken effect.
+
+==============================================================================
+PROMPT-SHAPE (n_cats) OOD  --  profile: paper_greedy
+==============================================================================
+Training chunks each image's categories into queries of 1..3. The eval set
+asks for the image's FULL list, and 28.8% of eval queries have >3 categories
+-- a prompt shape the model never saw. Caveat: images with more categories
+also have more objects, so a drop may be difficulty, not novelty.
+
+--- paper_greedy/slow ---
+predictions: 4952  |  eval queries: 4952  |  unbucketed: 0
+
+    n_cats |     N |    F1 |     P |     R | F1@.5
+--------------------------------------------------
+         1 |  1025 |  74.0 |  78.3 |  70.1 |  86.0
+         2 |  1522 |  67.1 |  69.3 |  65.0 |  82.5
+         3 |   980 |  58.3 |  62.1 |  54.9 |  74.8
+       4-5 |   935 |  54.9 |  58.6 |  51.6 |  72.7
+        6+ |   490 |  50.4 |  54.3 |  47.0 |  68.5
+
+    n_cats |     N |    F1 |     P |     R | F1@.5
+--------------------------------------------------
+  <=3 (IN) |  3527 |  62.3 |  64.8 |  59.9 |  78.1
+  >3 (OOD) |  1425 |  51.6 |  54.8 |  48.7 |  69.7
+
+If '>3 (OOD)' is materially below '<=3 (IN)', the prompt-shape gap
+is real: training never showed the model a 4+-category query.
+Careful: bigger images have more categories AND more objects, so a
+small drop may just be task difficulty, not prompt-shape OOD.
+
+--- paper_greedy/fast ---
+predictions: 4952  |  eval queries: 4952  |  unbucketed: 0
+
+    n_cats |     N |    F1 |     P |     R | F1@.5
+--------------------------------------------------
+         1 |  1025 |  71.6 |  76.6 |  67.3 |  84.2
+         2 |  1522 |  64.4 |  68.1 |  61.1 |  79.7
+         3 |   980 |  55.5 |  60.7 |  51.2 |  72.1
+       4-5 |   935 |  52.4 |  59.2 |  47.1 |  69.3
+        6+ |   490 |  47.9 |  56.7 |  41.4 |  64.7
+
+    n_cats |     N |    F1 |     P |     R | F1@.5
+--------------------------------------------------
+  <=3 (IN) |  3527 |  59.5 |  63.4 |  56.0 |  75.2
+  >3 (OOD) |  1425 |  49.1 |  56.0 |  43.7 |  66.0
+
+If '>3 (OOD)' is materially below '<=3 (IN)', the prompt-shape gap
+is real: training never showed the model a 4+-category query.
+Careful: bigger images have more categories AND more objects, so a
+small drop may just be task difficulty, not prompt-shape OOD.
+
+--- paper_greedy/hybrid ---
+predictions: 4952  |  eval queries: 4952  |  unbucketed: 0
+
+    n_cats |     N |    F1 |     P |     R | F1@.5
+--------------------------------------------------
+         1 |  1025 |  73.3 |  78.6 |  68.6 |  86.0
+         2 |  1522 |  66.0 |  69.4 |  62.9 |  81.4
+         3 |   980 |  56.9 |  61.9 |  52.5 |  73.0
+       4-5 |   935 |  53.2 |  59.5 |  48.1 |  70.3
+        6+ |   490 |  48.4 |  56.8 |  42.1 |  65.2
+
+    n_cats |     N |    F1 |     P |     R | F1@.5
+--------------------------------------------------
+  <=3 (IN) |  3527 |  61.2 |  64.8 |  57.9 |  76.7
+  >3 (OOD) |  1425 |  49.6 |  56.1 |  44.5 |  66.7
+
+If '>3 (OOD)' is materially below '<=3 (IN)', the prompt-shape gap
+is real: training never showed the model a 4+-category query.
+Careful: bigger images have more categories AND more objects, so a
+small drop may just be task difficulty, not prompt-shape OOD.
+