feat(evaluation): add the ViT-GPT2 Phase 3 greedy run

nlpconnect/vit-gpt2-image-captioning at the pinned revision dc68f91,
greedy (num_beams 1, no sampling, max_new_tokens 40, repetition penalty
1.0, no prompt, float32), scored on the committed 500-image slice under
EVAL_METHODOLOGY.md section 8. Written by scripts/compare_models.py.

results/phase3-vit-gpt2-greedy/comparison_meta.json

@@ -0,0 +1,30 @@
+{
+  "protocol": "docs/EVAL_METHODOLOGY.md § 8",
+  "model_id": "vit-gpt2",
+  "backend": "huggingface",
+  "captioner": "HFCaptioner",
+  "hub_repo": "nlpconnect/vit-gpt2-image-captioning",
+  "revision": "dc68f91c06a1ba6f15268e5b9c13ae7a7c514084",
+  "decode_settings": {
+    "num_beams": 1,
+    "do_sample": false,
+    "max_new_tokens": 40,
+    "repetition_penalty": 1.0,
+    "no_repeat_ngram_size": 0,
+    "precision": "float32"
+  },
+  "normalisation": "preprocess_caption -> strip_sentinels",
+  "slice": {
+    "source": "results/stabilized-greedy/predictions.jsonl",
+    "fingerprint_sha256": "6b5628bfa410ed233ef9603beed3c05c9e63acebc8bfa31d7e63e634f9e25116",
+    "images": 500,
+    "references": 732
+  },
+  "batch_size": 1,
+  "device": "cpu",
+  "seed": 42,
+  "metrics": {
+    "meteor": true,
+    "cider": true
+  }
+}

results/phase3-vit-gpt2-greedy/metrics.json

@@ -0,0 +1,11 @@
+{
+  "n_examples": 500,
+  "bleu1": 49.12276232739276,
+  "bleu2": 33.41039138807018,
+  "bleu3": 22.91337848185665,
+  "bleu4": 15.84152396302579,
+  "rouge_l": 44.509803731649235,
+  "meteor": 19.782713178028878,
+  "cider": 1.2629916931297422,
+  "errors": {}
+}

results/phase3-vit-gpt2-greedy/report.md

@@ -0,0 +1,20 @@
+# Evaluation run — vit-gpt2
+
+- Decode strategy: `greedy`
+- Weights: `nlpconnect/vit-gpt2-image-captioning@dc68f91c06a1ba6f15268e5b9c13ae7a7c514084`
+- Tokenizer dir: `nlpconnect/vit-gpt2-image-captioning@dc68f91c06a1ba6f15268e5b9c13ae7a7c514084`
+- Samples: **500**
+- Timestamp (UTC): 2026-10-06T14:46:39.934422+00:00
+- Repetition penalty: 1.0
+
+## Metrics
+
+| Metric | Value |
+|---|---|
+| BLEU-1 | 49.12 |
+| BLEU-2 | 33.41 |
+| BLEU-3 | 22.91 |
+| BLEU-4 | 15.84 |
+| ROUGE-L | 44.51 |
+| METEOR | 19.78 |
+| CIDEr | 1.26 |

results/phase3-vit-gpt2-greedy/run_meta.json

@@ -0,0 +1,12 @@
+{
+  "model_id": "vit-gpt2",
+  "decode_strategy": "greedy",
+  "weights_path": "nlpconnect/vit-gpt2-image-captioning@dc68f91c06a1ba6f15268e5b9c13ae7a7c514084",
+  "tokenizer_dir": "nlpconnect/vit-gpt2-image-captioning@dc68f91c06a1ba6f15268e5b9c13ae7a7c514084",
+  "n_samples": 500,
+  "max_length": 40,
+  "beam_width": null,
+  "length_penalty": null,
+  "repetition_penalty": 1.0,
+  "timestamp_utc": "2026-10-06T14:46:39.934422+00:00"
+}