LangTorch 0.9.9

Framework for intuitive LLM application development with tensors.

LangTorch

LangTorch is a Python package designed to simplify the development of LLM applications by leveraging familiar PyTorch concepts.

Installation

pip install langtorch

Overview

LangTorch provides a structured approach to LLM applications, offering:

• TextTensors: A unified way to handle prompt templates, completion dictionaries, and chat histories.
• TextModules: Building blocks, derived from torch.nn.Module, specifically tailored for text operations and LLM calls both locally and via an API.
• other things that are also better than langchain

Examples

TextTensors

Creating and manipulating textual data as tensors:

template = TextTensor([["Explain {theory} in terms of {framework}"],  
                       ["Argue how {framework} can prove {theory}"]])  

result = template * TextTensor({"theory": "active inference", "framework": "thermodynamics" })

print(result)
# Outputs: [[Explain active inference in terms of thermodynamics]
#           [Argue how thermodynamics can prove active inference]]

TextModules

Building sequences of operations on text data:

chain = torch.nn.Sequential(
    TextModule("Calculate this equation: {}"),
    langtorch.methods.CoT,
    GPT4
    TextModule("Is this reasoning correct? {}", activation = GPT4)
)

output = chain(TextTensor(["170*32 =", "4*20 =", "123*45/10 =", "2**10*5 ="]))

Cosine Similarities

Compute similarities between entries:

from langtorch.tt import CosineSimilarity

cos = CosineSimilarity()
similarities = cos(TextTensor([["Yes"], ["No"]]), TextTensor(["1", "0", "Noo", "Yees"]))

Contribute

Your feedback and contributions are valued. Feel free to check out our contribution guidelines.

License

LangTorch is MIT licensed. See the LICENSE file for details.