Image-Text-to-Text
Transformers
Safetensors
qwen2_5_vl
qwen3_vl
vision-language
visual-token-pruning
visionzip
scopd
Instructions to use enmingzhangzz/SCOPD with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use enmingzhangzz/SCOPD with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="enmingzhangzz/SCOPD")# pip install -U transformers accelerate # Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("enmingzhangzz/SCOPD", device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use enmingzhangzz/SCOPD with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "enmingzhangzz/SCOPD" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "enmingzhangzz/SCOPD", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/enmingzhangzz/SCOPD
- SGLang
How to use enmingzhangzz/SCOPD with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "enmingzhangzz/SCOPD" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "enmingzhangzz/SCOPD", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "enmingzhangzz/SCOPD" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "enmingzhangzz/SCOPD", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use enmingzhangzz/SCOPD with Docker Model Runner:
docker model run hf.co/enmingzhangzz/SCOPD
File size: 6,910 Bytes
ab18ce0 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 | import base64
import json
from pathlib import Path
import pytest
from scopd_eval import postprocess
from scopd_eval.datasets import make_record
from scopd_eval.infer import visionzip_settings
from scopd_eval.protocol import DEFAULT_MODEL, candidate_answer, exact_match, make_request
from scopd_eval.scoring import summarize
IMAGE = 'data:image/png;base64,' + base64.b64encode(b'test-image-bytes').decode()
def record(answer='B', candidate='<think>A is tempting</think><answer>B</answer>', **extra):
return {'dataset': 'MMStar', 'sample_id': '1', 'question': 'Which object?',
'options': {'A': 'red cube', 'B': 'blue sphere'}, 'reference': answer,
'images': [IMAGE], 'raw_prediction': candidate, **extra}
@pytest.mark.parametrize('candidate,match', [
('<think>A</think><answer>B</answer>', True),
('<answer>A</answer>', False), ('blue sphere', True),
('B or A', False), ('I think A, not B', False),
('<think>the answer is B', False), ('<answer>B', False),
('<answer>A</answer><answer>B</answer>', False),
('<answer>B</answer> Actually A', False),
])
def test_match_is_strict(candidate, match):
assert exact_match(record(candidate=candidate)) is match
def test_numeric_equivalence_without_substring_matching():
assert exact_match(record(answer='0.5', candidate='1/2', options={}))
assert not exact_match(record(answer='0.5', candidate='0.5 or 0.6', options={}))
assert not exact_match(record(answer='5', candidate='15', options={}))
def test_judge_sees_question_gt_candidate_and_all_images():
row = record(candidate='the sphere', images=[IMAGE, IMAGE])
payload, key, identity = make_request(row, '.', revision='revision1')
assert payload['model'] == DEFAULT_MODEL
content = payload['messages'][1]['content']
context = json.loads(content[0]['text'])
assert context['reference_answer'] == 'B' and context['candidate_final_answer'] == 'the sphere'
assert context['question'] == row['question'] and context['options'] == row['options']
assert [x['type'] for x in content] == ['text', 'image_url', 'image_url']
assert 'A is tempting' not in content[0]['text']
assert len(key) == 64 and len(identity['image_sha256']) == 2
assert payload['chat_template_kwargs']['enable_thinking'] is False
def test_cache_changes_with_gt_image_revision_and_candidate():
row = record()
key = make_request(row, '.', revision='a')[1]
for changed in [dict(row, reference='A'), dict(row, raw_prediction='A'),
dict(row, images=['data:image/png;base64,' + base64.b64encode(b'other').decode()])]:
assert make_request(changed, '.', revision='a')[1] != key
assert make_request(row, '.', revision='b')[1] != key
def test_missing_image_fails_closed(tmp_path):
with pytest.raises(FileNotFoundError):
make_request(record(images=['missing.png']), tmp_path, revision='a')
with pytest.raises(ValueError):
make_request(record(images=[]), tmp_path, revision='a')
def test_only_mismatches_call_judge_and_resume(tmp_path, monkeypatch):
rows = [record(), record(candidate='the sphere', sample_id='2'),
record(candidate='A', sample_id='3')]
path = tmp_path / 'predictions.jsonl'
path.write_text(''.join(json.dumps(row) + '\n' for row in rows))
calls = []
def query(url, payload):
text = json.loads(payload['messages'][1]['content'][0]['text'])['candidate_final_answer']
calls.append(text)
return text == 'the sphere', {'mock': True}
monkeypatch.setattr(postprocess, 'query', query)
kwargs = dict(image_root=tmp_path, base_url='http://localhost', model=DEFAULT_MODEL, revision='pin')
result = postprocess.run(path, tmp_path / 'out', **kwargs)
assert sorted(calls) == ['A', 'the sphere']
assert result['fallback_total'] == 2
assert result['scores']['MMStar']['accuracy_percent'] == pytest.approx(200 / 3)
postprocess.run(path, tmp_path / 'out', **kwargs)
assert len(calls) == 2
with pytest.raises(ValueError):
postprocess.run(path, tmp_path / 'out', **dict(kwargs, revision='other'))
def test_failure_not_converted_to_incorrect_score(tmp_path, monkeypatch):
path = tmp_path / 'predictions.jsonl'
path.write_text(json.dumps(record(candidate='unknown')) + '\n')
def query(*args):
raise TimeoutError('test failure')
monkeypatch.setattr(postprocess, 'query', query)
with pytest.raises(RuntimeError):
postprocess.run(path, tmp_path / 'out', image_root=tmp_path, base_url='http://localhost',
model=DEFAULT_MODEL, revision='pin')
summary = json.loads((tmp_path / 'out/summary.json').read_text())
assert summary['status'] == 'incomplete' and 'scores' not in summary
def test_mme_pair_and_mmvp_pair_scoring():
rows = [dict(record(sample_id=str(i), dataset='MME'), correct=i != 2, exact_match=i != 2,
metadata={'category': 'count', 'image_path': str(i // 2)}) for i in range(4)]
assert summarize(rows)['MME']['mme_score'] == 125
pairs = [dict(r, dataset='MMVP', sample_id=str(i + 1)) for i, r in enumerate(rows)]
assert summarize(pairs)['MMVP']['pair_accuracy_percent'] == 50
@pytest.mark.parametrize('retention,backend', [(.1, .95), (.2, .85), (.3, .75), (1., 0.)])
def test_retention_is_not_backend_ratio(retention, backend):
assert visionzip_settings(retention)['visionzip_ratio'] == backend
def test_hallusion_boolean_reference():
row = make_record('HallusionBench', {'index': 'x', 'question': 'Is it blue?', 'answer': 1}, 'Yes', ['x.png'])
assert row['reference'] == 'Yes' and exact_match(row)
def test_cvbench_keeps_source_macro_then_2d_3d_macro():
rows = []
for i, (source, split, correct) in enumerate([('a', '2D', True), ('b', '2D', False),
('b', '2D', False), ('b', '2D', False),
('x', '3D', True)]):
rows.append(dict(record(dataset='CVBench', sample_id=str(i)), correct=correct, exact_match=correct,
metadata={'source': source, 'split': split}))
report = summarize(rows)['CVBench']
assert report['2d_source_macro_percent'] == 50
assert report['3d_accuracy_percent'] == 100
assert report['cvbench_macro_percent'] == 75
def test_mmereal_duplicate_option_labels_are_not_repaired():
source = {'index': 1, 'question': 'Which?', 'answer': 'E',
'A': 'first', 'B': 'second', 'C': 'third', 'D': 'fourth', 'E': 'fifth',
'multi-choice options': '(A) first\n(B) second\n(C) third\n(D) fourth\n(D) fifth'}
row = make_record('MME-RealWorld-Lite', source, 'fifth', [IMAGE])
assert row['options']['E'] == 'E' and not exact_match(row)
payload, _, _ = make_request(row, '.', revision='pin')
assert '(D) fifth' in payload['messages'][1]['content'][0]['text']
|