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

README.md

@@ -224,7 +224,7 @@ Batch size and learning rate illustrate that ceiling. Effective batch 16 came in
 Enlarging the training set did what none of those could, and training it longer did not extend the gain. That narrows the remaining question to **data composition**: the corpus added was RSICD-only, and the gain it produced is correspondingly RSICD-only while in-domain stays flat. Additional in-domain imagery is the targeted next step.
 
 
-## Round 1XX - verbose grounded answers (Aspaklarya-101 ... 107)
+## Round 1XX - verbose grounded answers (Aspaklarya-101 ... 108)
 
 ### Why a second round
 
@@ -254,7 +254,60 @@ max seq 2048, vision tower trained. **One variable each.**
 | Aspaklarya-104 | LoRA **scale** `alpha/r` 1.0 -> 2.0, rank 16 | `train` | 18,862 / 6,288 | 5.4 h |
 | Aspaklarya-105 | **objective balance**, `w_caption` 1.0 -> 0.4 | `train` | 18,862 / 6,288 | 5.4 h |
 | Aspaklarya-106 | **LoRA soup** of the four epoch-3 adapters | n/a | n/a | **0 h** |
-| Aspaklarya-107 | **cosine** schedule, 3 epochs | `train` | 18,862 / 6,288 | in progress |
+| Aspaklarya-107 | **cosine** schedule, lr 1e-4, warmup 0.10, 3 epochs | `train` | 18,862 / 6,288 | 3.8 h |
+| Aspaklarya-108 | **seed replication** of 102 (1234, 2024) | `train` | 18,862 / 6,288 | 2 x 5.4 h |
+
+### Hyperparameters in full
+
+Every value actually passed to the trainer, read back from each run's
+`run_config.json`. Cells that differ from the Aspaklarya-102 reference are
+**bold** - that is the single variable the run was built to test.
+
+| | 101 | 102 (ref) | 103 | 104 | 105 | 106 | 106b | 107 | 108 |
+|---|---|---|---|---|---|---|---|---|---|
+| split | **`train,other`** | `train` | `train` | `train` | `train` | n/a | n/a | `train` | `train` |
+| items | **35,979** | 18,862 | 18,862 | 18,862 | 18,862 | n/a | n/a | 18,862 | 18,862 |
+| images | **11,994** | 6,288 | 6,288 | 6,288 | 6,288 | n/a | n/a | 6,288 | 6,288 |
+| LoRA rank | 16 | 16 | **32** | 16 | 16 | 64 | 64 | 16 | 16 |
+| LoRA alpha | 16 | 16 | **32** | **32** | 16 | 64 | 64 | 16 | 16 |
+| scale `alpha/r` | 1.0 | 1.0 | 1.0 | **2.0** | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 |
+| LoRA dropout | 0.0 | 0.0 | 0.0 | 0.0 | 0.0 | n/a | n/a | 0.0 | 0.0 |
+| vision tower | trained | trained | trained | trained | trained | trained | trained | trained | trained |
+| `w_caption` | 1.0 | 1.0 | 1.0 | 1.0 | **0.4** | n/a | n/a | 1.0 | 1.0 |
+| `w_vqa` | 1.0 | 1.0 | 1.0 | 1.0 | 1.0 | n/a | n/a | 1.0 | 1.0 |
+| loss weighting | token | token | token | token | token | n/a | n/a | token | token |
+| epochs | 4 | 4 | 4 | 4 | 4 | n/a | n/a | **3** | 4 |
+| learning rate | 2e-4 | 2e-4 | 2e-4 | 2e-4 | 2e-4 | n/a | n/a | **1e-4** | 2e-4 |
+| scheduler | linear | linear | linear | linear | linear | n/a | n/a | **cosine** | linear |
+| warmup ratio | 0.03 | 0.03 | 0.03 | 0.03 | 0.03 | n/a | n/a | **0.10** | 0.03 |
+| weight decay | 0.001 | 0.001 | 0.001 | 0.001 | 0.001 | n/a | n/a | 0.001 | 0.001 |
+| grad clip | 0.3 | 0.3 | 0.3 | 0.3 | 0.3 | n/a | n/a | 0.3 | 0.3 |
+| batch x accum | 4 x 8 | 4 x 8 | 4 x 8 | 4 x 8 | 4 x 8 | n/a | n/a | 4 x 8 | 4 x 8 |
+| effective batch | 32 | 32 | 32 | 32 | 32 | n/a | n/a | 32 | 32 |
+| max seq length | 2048 | 2048 | 2048 | 2048 | 2048 | n/a | n/a | 2048 | 2048 |
+| max image side | 1024 | 1024 | 1024 | 1024 | 1024 | n/a | n/a | 1024 | 1024 |
+| seed | 3407 | 3407 | 3407 | 3407 | 3407 | n/a | n/a | 3407 | **1234 / 2024** |
+| base precision | bf16 | bf16 | bf16 | bf16 | bf16 | bf16 | bf16 | bf16 | bf16 |
+| save every | 590 | 295 | 295 | 295 | 295 | n/a | n/a | 295 | 295 |
+| steps / epoch | 1125 | 590 | 590 | 590 | 590 | n/a | n/a | 590 | 590 |
+| total steps | 4500 | 2360 | 2360 | 2360 | 2360 | n/a | n/a | 1770 | 2360 |
+| wall clock | 9.8 h | 5.4 h | 5.4 h | 5.4 h | 5.4 h | **0 h** | **0 h** | 3.8 h | 2 x 5.4 h |
+
+Constant across the whole round and therefore omitted above: base model
+`Qwen/Qwen3-VL-8B-Instruct`, no 4-bit quantisation, 1 caption + 2 QA per
+image, VQA column `VQA_1126`, caption prompts drawn from the 50-query bank,
+eval-loss probe of 400 held-out items (200 caption / 200 VQA).
+
+**Aspaklarya-106 and 106b are not training runs.** They are exact LoRA soups:
+four rank-16 adapters stacked into a single rank-64 adapter, which reproduces
+the arithmetic mean of the parent weight deltas to within 2.3e-9. 106 averages
+the epoch-3 parents, 106b the epoch-4 parents. Neither consumed GPU training
+time; both are listed here so their rank and alpha are not mistaken for a
+tuned choice.
+
+**Aspaklarya-108 is a replication, not a new configuration.** It is
+Aspaklarya-102 run again at two further seeds, to measure the run-to-run
+spread that every comparison in this table silently assumes is small.
 
 ### How these are scored
 
@@ -292,9 +345,11 @@ generated at a 48-token budget, which clips verbose answers and penalises the
 | Aspaklarya-104-step1770 | 74.6 | 68.78 | 81.92 | 78.56 | **75.97** |
 | Aspaklarya-103-step1770 | 74.46 | 69.0 | 81.81 | 78.52 | **75.95** |
 | Aspaklarya-101-step2360 | 72.38 | 69.28 | 82.21 | 79.72 | **75.9** |
+| Aspaklarya-106b-soup | 72.77 | 69.99 | 81.53 | 78.73 | **75.75** |
 | Aspaklarya-102-step1770 | 73.45 | 68.93 | 81.56 | 78.6 | **75.64** |
 | Aspaklarya-102-step2360 | 73.24 | 68.71 | 82.0 | 78.49 | **75.61** |
 | Aspaklarya-105-step2360 | 73.18 | 68.7 | 82.09 | 78.41 | **75.59** |
+| Aspaklarya-107-step1770 | 72.16 | 69.9 | 80.98 | 77.06 | **75.03** |
 | **Aspaklarya-020-step4500** | 66.18 | 70.65 | 82.79 | 79.99 | **74.9** |
 | in-house reference | 65.26 | 63.59 | 71.91 | 58.58 | **64.84** |
 | Qwen3-VL-8B (base) | 60.64 | 53.13 | 68.31 | 63.28 | **61.34** |

index.html

@@ -248,7 +248,7 @@ a { color: #0969da; }
 <p>The <strong>hyperparameter</strong> search is closed. Rank, alpha, dropout, learning rate, schedule, weight decay, gradient clipping, vision-tower freezing, QA-per-image, caption prompt pool, batch geometry, effective batch size, learning rate at large batch and base-weight precision were all swept across 19 recipes under identical conditions. None displaced the reference recipe by more than the paired error bar.</p>
 <p>Batch size and learning rate illustrate that ceiling. Effective batch 16 came in at <strong>-0.59</strong> overall, batch 64 at 2e-4 at <strong>-0.56</strong>, and batch 64 at 3e-4 at <strong>-0.34</strong> overall and <strong>-0.87</strong> in-domain; batch 64 at 4e-4 was stopped at epoch 4 while declining. In each case captioning held steady or improved while VQA fell.</p>
 <p>Enlarging the training set did what none of those could, and training it longer did not extend the gain. That narrows the remaining question to <strong>data composition</strong>: the corpus added was RSICD-only, and the gain it produced is correspondingly RSICD-only while in-domain stays flat. Additional in-domain imagery is the targeted next step.</p>
-<h2>Round 1XX - verbose grounded answers (Aspaklarya-101 ... 107)</h2>
+<h2>Round 1XX - verbose grounded answers (Aspaklarya-101 ... 108)</h2>
 <h3>Why a second round</h3>
 <p>The 0XX round closed with hyperparameters saturated and data identified as the only live lever. It also left a product problem: every 0XX model answers in one word, because its supervision was one word. <code>yes</code> is a correct answer to <em>&quot;are there any anomalies in the scene?&quot;</em> and a useless one.</p>
 <p>The dataset author then supplied a new column, <strong><code>VQA_1126</code></strong>, replacing the terse answers for the in-domain (DM) images with ~300-character grounded answers that name objects, locations and context. The RSICD rows are byte-identical to the old column; only the 5,148 DM rows changed, from a mean of 5.9 characters to ~304.</p>
@@ -267,8 +267,54 @@ a { color: #0969da; }
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-104</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">LoRA <strong>scale</strong> <code>alpha/r</code> 1.0 -&gt; 2.0, rank 16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>train</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,862 / 6,288</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">5.4 h</td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-105</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>objective balance</strong>, <code>w_caption</code> 1.0 -&gt; 0.4</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>train</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,862 / 6,288</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">5.4 h</td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-106</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>LoRA soup</strong> of the four epoch-3 adapters</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>0 h</strong></td></tr>
-<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-107</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>cosine</strong> schedule, 3 epochs</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>train</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,862 / 6,288</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">in progress</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-107</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>cosine</strong> schedule, lr 1e-4, warmup 0.10, 3 epochs</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>train</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,862 / 6,288</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">3.8 h</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-108</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>seed replication</strong> of 102 (1234, 2024)</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>train</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,862 / 6,288</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2 x 5.4 h</td></tr>
 </tbody></table>
+<h3>Hyperparameters in full</h3>
+<p>Every value actually passed to the trainer, read back from each run&#x27;s <code>run_config.json</code>. Cells that differ from the Aspaklarya-102 reference are <strong>bold</strong> - that is the single variable the run was built to test.</p>
+<table style="border-collapse:collapse;margin:1rem 0;font-size:.85rem;border:1px solid #d0d7de"><thead><tr>
+<th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap"></th>
+<th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">101</th>
+<th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">102 (ref)</th>
+<th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">103</th>
+<th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">104</th>
+<th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">105</th>
+<th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">106</th>
+<th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">106b</th>
+<th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">107</th>
+<th style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;background:#f6f8fa;font-weight:600;white-space:nowrap">108</th>
+</tr></thead><tbody>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">split</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong><code>train,other</code></strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>train</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>train</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>train</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>train</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>train</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>train</code></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">items</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>35,979</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,862</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,862</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,862</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,862</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,862</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">18,862</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">images</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>11,994</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">6,288</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">6,288</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">6,288</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">6,288</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">6,288</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">6,288</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">LoRA rank</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>32</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">64</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">64</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">LoRA alpha</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>32</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>32</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">64</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">64</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">16</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">scale <code>alpha/r</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>2.0</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">LoRA dropout</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">0.0</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">0.0</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">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</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">0.0</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">vision tower</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">trained</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">trained</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">trained</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">trained</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">trained</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>w_caption</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>0.4</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><code>w_vqa</code></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1.0</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">loss weighting</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">token</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">token</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">token</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">token</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">token</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">token</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">token</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">epochs</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>3</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">learning rate</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">2e-4</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">2e-4</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">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>1e-4</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2e-4</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">scheduler</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">linear</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">linear</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">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>cosine</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">linear</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">warmup ratio</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.03</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.03</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.03</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.03</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.03</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>0.10</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.03</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">weight decay</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.001</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.001</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.001</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.001</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.001</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.001</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.001</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">grad clip</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.3</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.3</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.3</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.3</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.3</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.3</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">0.3</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">batch x accum</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4 x 8</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4 x 8</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4 x 8</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4 x 8</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4 x 8</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4 x 8</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4 x 8</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">effective batch</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">32</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">32</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">32</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">32</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">32</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">32</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">32</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">max seq length</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2048</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2048</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2048</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2048</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2048</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2048</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2048</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">max image side</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1024</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1024</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1024</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1024</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1024</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1024</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1024</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">seed</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">3407</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">3407</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">3407</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">3407</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">3407</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">3407</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>1234 / 2024</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">base precision</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">bf16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">bf16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">bf16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">bf16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">bf16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">bf16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">bf16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">bf16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">bf16</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">save every</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">590</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">295</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">295</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">295</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">295</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">295</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">295</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">steps / epoch</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1125</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">590</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">590</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">590</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">590</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">590</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">590</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">total steps</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">4500</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">n/a</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2360</td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">wall clock</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">9.8 h</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">5.4 h</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">5.4 h</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">5.4 h</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">5.4 h</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>0 h</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>0 h</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">3.8 h</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">2 x 5.4 h</td></tr>
+</tbody></table>
+<p>Constant across the whole round and therefore omitted above: base model <code>Qwen/Qwen3-VL-8B-Instruct</code>, no 4-bit quantisation, 1 caption + 2 QA per image, VQA column <code>VQA_1126</code>, caption prompts drawn from the 50-query bank, eval-loss probe of 400 held-out items (200 caption / 200 VQA).</p>
+<p><strong>Aspaklarya-106 and 106b are not training runs.</strong> They are exact LoRA soups: four rank-16 adapters stacked into a single rank-64 adapter, which reproduces the arithmetic mean of the parent weight deltas to within 2.3e-9. 106 averages the epoch-3 parents, 106b the epoch-4 parents. Neither consumed GPU training time; both are listed here so their rank and alpha are not mistaken for a tuned choice.</p>
+<p><strong>Aspaklarya-108 is a replication, not a new configuration.</strong> It is Aspaklarya-102 run again at two further seeds, to measure the run-to-run spread that every comparison in this table silently assumes is small.</p>
 <h3>How these are scored</h3>
 <p>Round-1XX checkpoints were first ranked on the DM VQA slice alone - the slice the round was trained for. Every model in the programme has since been re-judged on <strong>one fixed sample drawn with seed 1105</strong>, drawn once outside the model loop so all 24 checkpoints see byte-identical items:</p>
 <table style="border-collapse:collapse;margin:1rem 0;font-size:.85rem;border:1px solid #d0d7de"><thead><tr>
@@ -302,9 +348,11 @@ a { color: #0969da; }
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-104-step1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.6</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.78</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.92</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.56</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.97</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-103-step1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">74.46</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.81</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.52</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.95</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-101-step2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.38</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.28</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.21</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.72</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.9</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-106b-soup</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.77</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.99</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.53</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.73</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.75</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-102-step1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.45</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.93</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">81.56</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.6</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.64</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-102-step2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.24</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.71</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.0</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.49</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.61</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-105-step2360</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">73.18</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.7</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.09</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">78.41</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.59</strong></td></tr>
+<tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Aspaklarya-107-step1770</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">72.16</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">69.9</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">80.98</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">77.06</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>75.03</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>Aspaklarya-020-step4500</strong></td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">66.18</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">70.65</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">82.79</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">79.99</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>74.9</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">in-house reference</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">65.26</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">63.59</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">71.91</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">58.58</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>64.84</strong></td></tr>
 <tr><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">Qwen3-VL-8B (base)</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">60.64</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">53.13</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">68.31</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap">63.28</td><td style="border:1px solid #d0d7de;padding:5px 10px;text-align:left;white-space:nowrap"><strong>61.34</strong></td></tr>