DeepSearch Graph Language Models

1.0.0 · active · verified Fri Apr 17

deepsearch-glm is a Python library that enables the creation and querying of Graph Language Models (GLMs). It integrates Large Language Models (LLMs) with knowledge graphs derived from unstructured documents, primarily leveraging the `deepsearch-toolkit` library for graph processing. The current version is 1.0.0, and it's actively developed by DeepSearch AI, with releases expected to follow a feature-driven cadence.

Common errors

```
ModuleNotFoundError: No module named 'torch'
```
cause The underlying `transformers` library, used by `deepsearch-glm`, requires a machine learning backend (PyTorch or TensorFlow) which was not installed.

fix Install the necessary backend: `pip install torch` (recommended) or `pip install tensorflow`. Alternatively, install `deepsearch-glm` with all extras: `pip install deepsearch-glm[all]`.
```
ImportError: cannot import name 'Document' from 'deepsearch.glm.core.models'
```
cause The `Document` model, crucial for representing input documents, originates from the `deepsearch-toolkit` library, not directly from `deepsearch-glm`'s core models.

fix Use the correct import path: `from deepsearch.documents.core.models import Document`.
```
OSError: Can't load 'deepsearch-ai/deepsearch-glm-base' model. Make sure that...
```
cause This error typically indicates issues during the model download or loading process, such as network problems, insufficient disk space, corrupted cache, or an incorrect model ID.

fix Check your internet connection, ensure sufficient disk space, and verify the model name (`deepsearch-ai/deepsearch-glm-base`). If issues persist, try clearing the Hugging Face cache (e.g., `rm -rf ~/.cache/huggingface/hub`) and retrying.

Warnings

gotcha The `GraphLanguageModel.from_pretrained()` method will download a large language model (e.g., `deepsearch-glm-base` is ~1.5GB) on its first invocation. This requires significant disk space, memory, and an active internet connection, and may take considerable time.
Fix: Ensure you have sufficient disk space, RAM, and a stable internet connection. Consider pre-downloading models or using a shared cache in production environments if applicable.
gotcha While `deepsearch-glm` simplifies interaction with GLMs, its core functionality for processing documents into knowledge graphs heavily relies on `deepsearch-toolkit`. For advanced graph manipulation or custom document processing, direct usage and understanding of `deepsearch-toolkit` is often necessary.
Fix: Familiarize yourself with `deepsearch-toolkit` documentation for advanced graph-related tasks. Ensure `deepsearch-toolkit` is installed (it's a core dependency).
gotcha The underlying Large Language Models (via `transformers`) require a machine learning backend like PyTorch (`torch`) or TensorFlow (`tensorflow`). While `deepsearch-glm` lists these as optional dependencies, practically, one of them is required for the library to function. You might encounter `ModuleNotFoundError` if neither is installed.
Fix: Install `deepsearch-glm` with the `[all]` extra to include common optional dependencies (e.g., `pip install deepsearch-glm[all]`), or manually install your preferred backend: `pip install torch` or `pip install tensorflow`.

Install

pip install deepsearch-glm Basic Install
pip install deepsearch-glm[all] Install with all optional dependencies (includes ML backends)

Imports

GraphLanguageModel

from deepsearch.glm.core.glm import GraphLanguageModel

GlmQuery

from deepsearch.glm.queries import GlmQuery

Document
wrong:
```
from deepsearch.glm.core.models import Document
```
correct:
```
from deepsearch.documents.core.models import Document
```
The Document model is part of deepsearch-toolkit, which is a dependency of deepsearch-glm.

Quickstart

This quickstart demonstrates how to load a GraphLanguageModel and query it with a simple document. It highlights the main components and interactions within `deepsearch-glm`. Note that the first run will download a large language model.

import os
from deepsearch.glm.core.glm import GraphLanguageModel
from deepsearch.glm.queries import GlmQuery
from deepsearch.documents.core.models import Document

# Note: The deepsearch-glm-base model is large (~1.5GB) and will be downloaded on first run.
# This example requires deepsearch-toolkit to implicitly process documents into a graph.

try:
    # 1. Initialize the GraphLanguageModel (this will download a model if not cached)
    # This might take a while and uses significant memory/disk space.
    print("Loading GraphLanguageModel... (first run may download ~1.5GB model)")
    glm_model = GraphLanguageModel.from_pretrained("deepsearch-ai/deepsearch-glm-base")
    print("GraphLanguageModel loaded.")

    # 2. Prepare a dummy document. In a real scenario, this would be processed
    # by deepsearch-toolkit to extract a knowledge graph before querying.
    doc_content = "The DeepSearch AI platform helps enterprises extract insights from unstructured data using advanced AI."
    dummy_document = Document(name="deepsearch_info.txt", content=doc_content)

    # 3. Create a query
    question = "What is the DeepSearch AI platform used for?"
    query = GlmQuery(
        model=glm_model,
        documents=[dummy_document], # Provide documents directly for implicit graph creation
        query_text=question
    )

    # 4. Run the query
    print(f"\nQuerying: {question}")
    result = query.run()

    # 5. Print the result
    print("\nQuery Result:")
    print(result.answer)

except ImportError as e:
    print(f"Error: {e}. Ensure all core and optional dependencies (like 'transformers', 'torch' or 'tensorflow') are installed.")
    print("Try: pip install deepsearch-glm[all]")
except Exception as e:
    print(f"An error occurred during execution: {e}")
    print("Common issues include insufficient disk space, network problems during model download, or missing ML backend (torch/tensorflow).")

view raw JSON →