Skip to content

llama

Llama generation module.

Classes:

LlamaGen 

LlamaGen(
    model_name: Optional[str] = None,
    temperature: Optional[float] = None,
    prompt_template: str = '',
    output_max_length: int = 500,
    device: str = 'auto',
    structured_output: Optional[Type[BaseModel]] = None,
    system_message: str = '',
    api_params: dict[str, Any] = DEFAULT_API_PARAMS,
    api_key: str = '',
    cache: Optional[Cache] = None,
    logs: dict[str, Any] = DEFAULT_LOGS,
)

Bases: GenerationBase

Llama Generation class.

Methods:

  • generate –

    Generate text using Llama model with language support.

Source code in src/rago/generation/base.py 
 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
def __init__(
    self,
    model_name: Optional[str] = None,
    temperature: Optional[float] = None,
    prompt_template: str = '',
    output_max_length: int = 500,
    device: str = 'auto',
    structured_output: Optional[Type[BaseModel]] = None,
    system_message: str = '',
    api_params: dict[str, Any] = DEFAULT_API_PARAMS,
    api_key: str = '',
    cache: Optional[Cache] = None,
    logs: dict[str, Any] = DEFAULT_LOGS,
) -> None:
    """Initialize Generation class."""
    if logs is DEFAULT_LOGS:
        logs = {}
    super().__init__(api_key=api_key, cache=cache, logs=logs)

    self.model_name: str = (
        model_name if model_name is not None else self.default_model_name
    )
    self.output_max_length: int = (
        output_max_length or self.default_output_max_length
    )
    self.temperature: float = (
        temperature
        if temperature is not None
        else self.default_temperature
    )

    self.prompt_template: str = (
        prompt_template or self.default_prompt_template
    )
    self.structured_output: Optional[Type[BaseModel]] = structured_output
    if api_params is DEFAULT_API_PARAMS:
        api_params = deepcopy(self.default_api_params or {})

    self.system_message = system_message
    self.api_params = api_params

    if device not in ['cpu', 'cuda', 'auto']:
        raise Exception(
            f'Device {device} not supported. Options: cpu, cuda, auto.'
        )

    cuda_available = torch.cuda.is_available()
    self.device_name: str = (
        'cpu' if device == 'cpu' or not cuda_available else 'cuda'
    )
    self.device = torch.device(self.device_name)

    self._validate()
    self._setup()

generate 

generate(query: str, context: list[str]) -> str

Generate text using Llama model with language support.

Source code in src/rago/generation/llama.py 
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
def generate(self, query: str, context: list[str]) -> str:
    """Generate text using Llama model with language support."""
    input_text = self.prompt_template.format(
        query=query, context=' '.join(context)
    )

    # Detect and set the language code for multilingual models (optional)
    language = str(detect(query)) or 'en'
    self.tokenizer.lang_code = language

    api_params = (
        self.api_params if self.api_params else self.default_api_params
    )

    # Generate the response with adjusted parameters

    model_params = dict(
        text_inputs=input_text,
        max_new_tokens=self.output_max_length,
        do_sample=True,
        temperature=self.temperature,
        eos_token_id=self.tokenizer.eos_token_id,
        **api_params,
    )
    response = self.generator(**model_params)

    self.logs['model_params'] = model_params

    # Extract and return the answer only
    answer = str(response[0].get('generated_text', ''))
    # Strip off any redundant text after the answer itself
    return answer.split('Answer:')[-1].strip()