0
# Text Scoring
1

2
Text similarity and likelihood scoring for comparing text pairs, ranking, and evaluation tasks. Supports various scoring methods including cosine similarity and likelihood-based metrics.
3

4
## Capabilities
5

6
### Text Scoring
7

8
Compare text pairs and compute similarity or likelihood scores for ranking, retrieval, and evaluation applications.
9

10
```python { .api }
11
def score(
12
    self,
13
    data_1: Union[SingletonPrompt, Sequence[SingletonPrompt], ScoreMultiModalParam],
14
    data_2: Union[SingletonPrompt, Sequence[SingletonPrompt], ScoreMultiModalParam],
15
    /,
16
    *,
17
    truncate_prompt_tokens: Optional[int] = None,
18
    use_tqdm: Union[bool, Callable[..., tqdm]] = True,
19
    pooling_params: Optional[PoolingParams] = None,
20
    lora_request: Optional[Union[List[LoRARequest], LoRARequest]] = None
21
) -> List[ScoringRequestOutput]:
22
    """
23
    Score similarity between text pairs.
24

25
    Parameters:
26
    - data_1: First set of texts (positional-only)
27
    - data_2: Second set of texts to compare (positional-only)
28
    - truncate_prompt_tokens: Maximum prompt length (keyword-only)
29
    - use_tqdm: Show progress bar (keyword-only)
30
    - pooling_params: Pooling configuration for embeddings (keyword-only)
31
    - lora_request: LoRA adapter configuration (keyword-only)
32

33
    Returns:
34
    List of ScoringRequestOutput with similarity scores
35

36
    Note: The scoring method is automatically determined by the model type.
37
    Embedding models use cosine similarity, cross-encoder models use likelihood.
38
    """
39
```
40

41
## Usage Examples
42

43
### Text Similarity Scoring
44

45
```python
46
from vllm import LLM, PoolingParams
47

48
llm = LLM(model="sentence-transformers/all-MiniLM-L6-v2")
49

50
text_pairs = [
51
    ("The cat sat on the mat.", "A cat was sitting on a mat."),
52
    ("Python is a programming language.", "Java is a programming language."),
53
    ("I love pizza.", "Pizza is the worst food ever.")
54
]
55

56
pooling_params = PoolingParams(pooling_type="MEAN", normalize=True)
57

58
for text1, text2 in text_pairs:
59
    outputs = llm.score(
60
        [text1], [text2],
61
        kind="cosine_similarity",
62
        pooling_params=pooling_params
63
    )
64
    score = outputs[0].outputs.score
65
    print(f"Similarity: {score:.3f} - '{text1}' vs '{text2}'")
66
```
67

68
## Types
69

70
```python { .api }
71
class ScoringRequestOutput:
72
    id: str
73
    outputs: ScoringOutput
74
    prompt_token_ids: List[int]
75
    finished: bool
76

77
class ScoringOutput:
78
    score: float  # Similarity or likelihood score
79
```