swarmauri_parser_bertembedding 0.8.0

Swarmauri Bert Embedding Parser

---

Swarmauri Parser Bert Embedding

Parser that converts text into embeddings using a Hugging Face BERT encoder. Produces Document objects whose metadata carries the averaged token embedding so downstream Swarmauri pipelines can work with dense vectors.

Features

• Uses transformers.BertModel + BertTokenizer (default bert-base-uncased).
• Accepts single strings or lists of strings and emits Document instances with original text and embedding metadata.
• Runs in inference (eval) mode with automatic torch.no_grad() handling.
• Works on CPU by default; configure PyTorch device settings to leverage GPU.

Prerequisites

• Python 3.10 or newer.
• PyTorch compatible with your hardware (installed automatically via transformers if not present; install CUDA-enabled wheels manually when needed).
• Internet access on first run so Hugging Face downloads tokenizer/model weights (or warm the cache ahead of time).

Installation

# pip
pip install swarmauri_parser_bertembedding

# poetry
poetry add swarmauri_parser_bertembedding

# uv (pyproject-based projects)
uv add swarmauri_parser_bertembedding

Quickstart

from swarmauri_parser_bertembedding import BERTEmbeddingParser

parser = BERTEmbeddingParser(parser_model_name="bert-base-uncased")

documents = parser.parse([
    "Swarmauri agents cooperate over shared memory.",
    "Dense embeddings power semantic search.",
])

for doc in documents:
    vector = doc.metadata["embedding"]
    print(doc.content)
    print(len(vector), vector[:5])

Custom Models & Devices

import torch
from swarmauri_parser_bertembedding import BERTEmbeddingParser
from transformers import BertModel

class GPUParser(BERTEmbeddingParser):
    def __init__(self, **kwargs):
        super().__init__(**kwargs)
        self._model = BertModel.from_pretrained(self.parser_model_name).to("cuda")

parser = GPUParser(parser_model_name="bert-base-multilingual-cased")
parser._model.eval()

Batch Embeddings at Scale

from tqdm import tqdm
from swarmauri_parser_bertembedding import BERTEmbeddingParser

texts = [f"Paragraph {i}" for i in range(1000)]
parser = BERTEmbeddingParser()

batched_docs = []
batch_size = 32
for start in tqdm(range(0, len(texts), batch_size)):
    batch = texts[start:start + batch_size]
    batched_docs.extend(parser.parse(batch))

Persist the resulting vectors into Swarmauri vector stores (Redis, Qdrant, etc.) via the metadata field.

Tips

• Preprocess text to match model expectations (lowercase for uncased BERT, language-specific cleanup for multilingual models).
• For extremely long documents, consider chunking before calling parse to respect the 512 token limit.
• Use PyTorch's to("cuda") or to("mps") to execute on GPUs or Apple silicon accelerators.
• Cache Hugging Face weights in CI/CD environments (HF_HOME=/cache/hf) to avoid repeated downloads.

Want to help?

If you want to contribute to swarmauri-sdk, read up on our guidelines for contributing that will help you get started.