apoorvrajdev commited on
Commit
2a5ee2b
·
1 Parent(s): cc71c5f

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

Browse files

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 ADDED
@@ -0,0 +1,30 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "protocol": "docs/EVAL_METHODOLOGY.md § 8",
3
+ "model_id": "vit-gpt2",
4
+ "backend": "huggingface",
5
+ "captioner": "HFCaptioner",
6
+ "hub_repo": "nlpconnect/vit-gpt2-image-captioning",
7
+ "revision": "dc68f91c06a1ba6f15268e5b9c13ae7a7c514084",
8
+ "decode_settings": {
9
+ "num_beams": 1,
10
+ "do_sample": false,
11
+ "max_new_tokens": 40,
12
+ "repetition_penalty": 1.0,
13
+ "no_repeat_ngram_size": 0,
14
+ "precision": "float32"
15
+ },
16
+ "normalisation": "preprocess_caption -> strip_sentinels",
17
+ "slice": {
18
+ "source": "results/stabilized-greedy/predictions.jsonl",
19
+ "fingerprint_sha256": "6b5628bfa410ed233ef9603beed3c05c9e63acebc8bfa31d7e63e634f9e25116",
20
+ "images": 500,
21
+ "references": 732
22
+ },
23
+ "batch_size": 1,
24
+ "device": "cpu",
25
+ "seed": 42,
26
+ "metrics": {
27
+ "meteor": true,
28
+ "cider": true
29
+ }
30
+ }
results/phase3-vit-gpt2-greedy/diagnostics.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
results/phase3-vit-gpt2-greedy/metrics.json ADDED
@@ -0,0 +1,11 @@
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "n_examples": 500,
3
+ "bleu1": 49.12276232739276,
4
+ "bleu2": 33.41039138807018,
5
+ "bleu3": 22.91337848185665,
6
+ "bleu4": 15.84152396302579,
7
+ "rouge_l": 44.509803731649235,
8
+ "meteor": 19.782713178028878,
9
+ "cider": 1.2629916931297422,
10
+ "errors": {}
11
+ }
results/phase3-vit-gpt2-greedy/predictions.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
results/phase3-vit-gpt2-greedy/report.md ADDED
@@ -0,0 +1,20 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Evaluation run — vit-gpt2
2
+
3
+ - Decode strategy: `greedy`
4
+ - Weights: `nlpconnect/vit-gpt2-image-captioning@dc68f91c06a1ba6f15268e5b9c13ae7a7c514084`
5
+ - Tokenizer dir: `nlpconnect/vit-gpt2-image-captioning@dc68f91c06a1ba6f15268e5b9c13ae7a7c514084`
6
+ - Samples: **500**
7
+ - Timestamp (UTC): 2026-10-06T14:46:39.934422+00:00
8
+ - Repetition penalty: 1.0
9
+
10
+ ## Metrics
11
+
12
+ | Metric | Value |
13
+ |---|---|
14
+ | BLEU-1 | 49.12 |
15
+ | BLEU-2 | 33.41 |
16
+ | BLEU-3 | 22.91 |
17
+ | BLEU-4 | 15.84 |
18
+ | ROUGE-L | 44.51 |
19
+ | METEOR | 19.78 |
20
+ | CIDEr | 1.26 |
results/phase3-vit-gpt2-greedy/run_meta.json ADDED
@@ -0,0 +1,12 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model_id": "vit-gpt2",
3
+ "decode_strategy": "greedy",
4
+ "weights_path": "nlpconnect/vit-gpt2-image-captioning@dc68f91c06a1ba6f15268e5b9c13ae7a7c514084",
5
+ "tokenizer_dir": "nlpconnect/vit-gpt2-image-captioning@dc68f91c06a1ba6f15268e5b9c13ae7a7c514084",
6
+ "n_samples": 500,
7
+ "max_length": 40,
8
+ "beam_width": null,
9
+ "length_penalty": null,
10
+ "repetition_penalty": 1.0,
11
+ "timestamp_utc": "2026-10-06T14:46:39.934422+00:00"
12
+ }