tessl/pypi-farm-haystack

LLM framework to build customizable, production-ready LLM applications with pipelines connecting models, vector DBs, and data processors.

—

Pending

Overview

Eval results

Files

Reader Components

Name: tessl/pypi-farm-haystack
Author: tessl

Reading comprehension components for extractive question answering using FARM, Transformers, and specialized table readers.

Core Imports

from haystack.nodes.reader import FARMReader, TransformersReader, TableReader
from haystack.nodes.reader.base import BaseReader

Base Reader

from haystack.nodes.reader.base import BaseReader
from haystack.schema import Document, Answer
from typing import List, Optional, Dict, Any

class BaseReader:
    def predict(self, query: str, documents: List[Document], top_k: Optional[int] = None) -> List[Answer]:
        """
        Extract answers from documents for the given query.
        
        Args:
            query: Question text
            documents: List of documents to search for answers
            top_k: Maximum number of answers to return
            
        Returns:
            List of Answer objects with extracted text and confidence scores
        """

FARM Reader

from haystack.nodes.reader.farm import FARMReader

class FARMReader(BaseReader):
    def __init__(self, model_name_or_path: str = "deepset/roberta-base-squad2",
                 use_gpu: bool = True, no_ans_boost: float = 0.0,
                 return_no_answer: bool = False, top_k: int = 10,
                 max_seq_len: int = 256, doc_stride: int = 128):
        """
        Initialize FARM-based QA reader.
        
        Args:
            model_name_or_path: HuggingFace model name or local path
            use_gpu: Whether to use GPU acceleration
            no_ans_boost: Boost for "no answer" predictions
            return_no_answer: Whether to return "no answer" predictions
            top_k: Number of answers to return per document
            max_seq_len: Maximum sequence length for input
            doc_stride: Stride for sliding window over long documents
        """

Transformers Reader

from haystack.nodes.reader.transformers import TransformersReader

class TransformersReader(BaseReader):
    def __init__(self, model_name_or_path: str = "deepset/roberta-base-squad2",
                 tokenizer: Optional[str] = None, use_gpu: bool = True,
                 top_k: int = 10, max_seq_len: int = 256, doc_stride: int = 128):
        """
        Initialize Transformers-based QA reader.
        
        Args:
            model_name_or_path: HuggingFace model name or local path
            tokenizer: Tokenizer name (defaults to model tokenizer)
            use_gpu: Whether to use GPU acceleration
            top_k: Number of answers to return per document
            max_seq_len: Maximum sequence length for input
            doc_stride: Stride for sliding window over long documents
        """

Table Reader

from haystack.nodes.reader.table import TableReader

class TableReader(BaseReader):
    def __init__(self, model_name_or_path: str = "google/tapas-base-finetuned-wtq",
                 use_gpu: bool = True, top_k: int = 10):
        """
        Initialize table-based QA reader for structured data.
        
        Args:
            model_name_or_path: TAPAS model name or local path
            use_gpu: Whether to use GPU acceleration
            top_k: Number of answers to return
        """

Install with Tessl CLI