> ## Documentation Index
> Fetch the complete documentation index at: https://docs.bubench.lexmount.io/llms.txt
> Use this file to discover all available pages before exploring further.

# eval_utils

> Evaluation model and scoring utilities

# browseruse\_bench.utils.eval\_utils

Evaluation-related utility functions and classes.

## Import

```python theme={null}
from browseruse_bench.utils import (
    EvaluationModel,
    load_evaluation_model,
    encode_image,
    extract_score_from_response,
    calculate_success,
    normalized_results_file,
)
```

***

## EvaluationModel

OpenAI model wrapper class for task evaluation.

```python theme={null}
class EvaluationModel:
    def __init__(
        self,
        model: str = "gpt-4o",
        api_key: str = None,
        base_url: str = None
    )
```

<ParamField path="model" type="str" default="gpt-4o">
  Model name
</ParamField>

<ParamField path="api_key" type="str" default="None">
  API Key, defaults to environment variable
</ParamField>

<ParamField path="base_url" type="str" default="None">
  API Base URL, defaults to environment variable
</ParamField>

### generate

Generate evaluation response with automatic retry.

```python theme={null}
def generate(
    self,
    messages: List[Dict],
    max_tokens: int = 2048,
    temperature: float = 0.3,
    **kwargs
) -> str
```

***

## load\_evaluation\_model

Load evaluation model with environment variable fallback.

```python theme={null}
def load_evaluation_model(
    model: str = None,
    api_key: str = None,
    base_url: str = None
) -> EvaluationModel
```

### Configuration

The eval model is configured in the root `config.yaml` under the `eval:` section:

```yaml theme={null}
eval:
  model: gpt-5.4
  api_key: $OPENAI_API_KEY
  base_url: $OPENAI_BASE_URL
```

`api_key` and `base_url` accept `$VAR` placeholders resolved from `.env`.

***

## encode\_image

Convert a PIL image to base64 string.

```python theme={null}
def encode_image(
    image: PIL.Image,
    scale_factor: float = 1.0
) -> str
```

<ParamField path="image" type="PIL.Image" required>
  PIL Image object
</ParamField>

<ParamField path="scale_factor" type="float" default="1.0">
  Image scale factor (between 0.0 and 1.0), e.g., 0.5 means 50% size
</ParamField>

***

## extract\_score\_from\_response

Extract numerical score from evaluation response.

```python theme={null}
def extract_score_from_response(response: str) -> int
```

<ParamField path="response" type="str" required>
  Evaluation response text
</ParamField>

<ResponseField name="return" type="int">
  Extracted score (0 if not found)
</ResponseField>

***

## calculate\_success

Determine if task is successful based on score threshold.

```python theme={null}
def calculate_success(
    score: int,
    threshold: int = 60
) -> bool
```

<ParamField path="score" type="int" required>
  Task score
</ParamField>

<ParamField path="threshold" type="int" default="60">
  Success threshold
</ParamField>

<ResponseField name="return" type="bool">
  True if score meets or exceeds threshold
</ResponseField>

***

## normalized\_results\_file

Context manager that yields a path guaranteed to be JSONL format.

```python theme={null}
@contextmanager
def normalized_results_file(results_file: Path) -> Generator[Path, None, None]
```

<ParamField path="results_file" type="Path" required>
  Results file path
</ParamField>

### Usage Example

```python theme={null}
from browseruse_bench.utils import normalized_results_file

with normalized_results_file(Path("results.json")) as jsonl_path:
    # jsonl_path is guaranteed to be JSONL format
    with open(jsonl_path) as f:
        for line in f:
            record = json.loads(line)
```