org card: per-experiment hyperparameters and VQA/caption results

README.md

@@ -43,8 +43,10 @@ Winners are merged into the bf16 base and re-validated: adapters are evaluated o
 | Aspaklarya-011 | 16 / 16 | 0.05 | 2e-4 | linear | trained | 1c + 2q | 18,864 | dropout, wd 0.05, clip 1.0 |
 | Aspaklarya-012 | 16 / 16 | 0.0 | 2e-4 | linear | trained | 1c + 2q | 18,864 | as 011, no dropout |
 | Aspaklarya-013 | 16 / 16 | 0.0 | 1e-4 | linear | trained | 1c + 2q | 18,864 | low LR, long schedule |
+| Aspaklarya-014 | 16 / 16 | 0.0 | 2e-4 | linear | trained | 1c + 2q | 18,864 | effective batch 16 |
+| Aspaklarya-015 | 16 / 16 | 0.0 | 2e-4 | linear | trained | 1c + 2q | 18,864 | effective batch 64 |
 
-Constant everywhere: effective batch 32, warmup 0.03, grad clip 0.3 (1.0 for 011/012), seed 3407.
+Constant everywhere: effective batch 32 (except 014/015), warmup 0.03, grad clip 0.3 (1.0 for 011/012), seed 3407.
 
 ## Results
 
@@ -60,7 +62,9 @@ Best checkpoint per run, scored on 1000 images (2000 graded items). **Overall**
 | Aspaklarya-005 | 79.55 | 75.74 | **77.64** | 81.37 | 82.37 | **81.87** |
 | Aspaklarya-008 | 79.82 | 75.40 | **77.61** | 80.62 | 81.42 | **81.02** |
 | Aspaklarya-003 | 79.55 | 75.52 | **77.53** | 81.23 | 82.18 | **81.71** |
+| Aspaklarya-015 | 80.07 | 74.73 | **77.40** | 82.14 | 82.76 | **82.45** |
 | Aspaklarya-012 | 79.73 | 75.02 | **77.37** | 81.68 | 82.05 | **81.86** |
+| Aspaklarya-014 | 79.95 | 74.77 | **77.36** | 81.51 | 81.93 | **81.72** |
 | Aspaklarya-007 | 79.33 | 75.33 | **77.33** | 81.13 | 80.82 | **80.97** |
 | Aspaklarya-013 | 79.11 | 75.35 | **77.23** | 80.58 | 82.20 | **81.39** |
 | Aspaklarya-009 | 78.53 | 75.85 | **77.19** | 80.01 | 82.34 | **81.18** |
@@ -76,7 +80,7 @@ Paired bootstrap over images puts the between-run standard error at **~0.45 over
 
 ## Findings
 
-- **Optimization is saturated.** LR, rank, alpha, dropout, weight decay, clipping, batch geometry and schedule were all swept; none beat the noise floor.
+- **Optimization is saturated.** LR, rank, alpha, dropout, weight decay, clipping, batch geometry, batch size and schedule were all swept; none beat the noise floor.
 - **The vision tower matters.** Freezing it is the one ablation that consistently hurt.
 - **Extra QA per image does not help.** 4-QA and 6-QA mixtures tie with 2-QA.
 - **Data is the remaining lever**, not hyperparameters.

index.html

@@ -79,8 +79,10 @@ a { color: #0969da; }
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-011</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16 / 16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.05</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2e-4</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">linear</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">trained</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1c + 2q</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,864</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">dropout, wd 0.05, clip 1.0</td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-012</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16 / 16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2e-4</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">linear</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">trained</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1c + 2q</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,864</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">as 011, no dropout</td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-013</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16 / 16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1e-4</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">linear</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">trained</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1c + 2q</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,864</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">low LR, long schedule</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-014</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16 / 16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2e-4</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">linear</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">trained</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1c + 2q</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,864</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">effective batch 16</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-015</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16 / 16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2e-4</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">linear</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">trained</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1c + 2q</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,864</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">effective batch 64</td></tr>
 </tbody></table>
-<p>Constant everywhere: effective batch 32, warmup 0.03, grad clip 0.3 (1.0 for 011/012), seed 3407.</p>
+<p>Constant everywhere: effective batch 32 (except 014/015), warmup 0.03, grad clip 0.3 (1.0 for 011/012), seed 3407.</p>
 <h2>Results</h2>
 <p>Best checkpoint per run, scored on 1000 images (2000 graded items). <strong>Overall</strong> = mean of caption and VQA. Right block = in-domain subset, closest to the target distribution.</p>
 <table style="border-collapse:collapse;margin:1rem 0;font-size:.85rem;border:1px solid #d0d7de"><thead><tr>
@@ -100,7 +102,9 @@ a { color: #0969da; }
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-005</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.55</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.74</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.64</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.37</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.37</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.87</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-008</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.82</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.40</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.61</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.62</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.42</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.02</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-003</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.55</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.52</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.53</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.23</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.18</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.71</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-015</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.07</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.73</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.40</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.14</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.76</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>82.45</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-012</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.73</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.02</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.37</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.68</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.05</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.86</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-014</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.95</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.77</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.36</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.51</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.93</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.72</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-007</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.33</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.33</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.33</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.13</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.82</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>80.97</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-013</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.11</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.35</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.23</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.58</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.20</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.39</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-009</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.53</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">75.85</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>77.19</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.01</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.34</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>81.18</strong></td></tr>
@@ -113,7 +117,7 @@ a { color: #0969da; }
 <p>Paired bootstrap over images puts the between-run standard error at <strong>~0.45 overall</strong>, so the top eight runs are statistically tied — every confidence interval against the leader straddles zero. Doubling the eval set from 500 to 1000 images did not separate them, and it reshuffled the order: one run climbed six places purely on noise. Only the gaps to the two reference models are real.</p>
 <h2>Findings</h2>
 <ul>
-<li><strong>Optimization is saturated.</strong> LR, rank, alpha, dropout, weight decay, clipping, batch geometry and schedule were all swept; none beat the noise floor.</li>
+<li><strong>Optimization is saturated.</strong> LR, rank, alpha, dropout, weight decay, clipping, batch geometry, batch size and schedule were all swept; none beat the noise floor.</li>
 <li><strong>The vision tower matters.</strong> Freezing it is the one ablation that consistently hurt.</li>
 <li><strong>Extra QA per image does not help.</strong> 4-QA and 6-QA mixtures tie with 2-QA.</li>
 <li><strong>Data is the remaining lever</strong>, not hyperparameters.</li>