Skip to main content

Compiler for the Sutra programming language — tensor ops on frozen-LLM embedding vectors.

Project description

sutra-compiler

The compiler for the Sutra programming language — a purely functional language whose primitives are tensor operations on frozen-LLM embedding vectors.

Install

pip install sutra-dev                     # validator + codegen only
pip install "sutra-dev[runtime]"          # adds torch so --emit / --run can execute the generated module
pip install "sutra-dev[runtime,embed]"    # adds the embedding client for LLM-substrate codebooks

Requires Python 3.11+.

Use

After install, the sutrac command is on your $PATH.

# Validate a .su source file.
sutrac path/to/program.su

# Validate every .su file under a directory.
sutrac path/to/directory

# Compile to self-contained PyTorch Python and print to stdout.
sutrac --emit path/to/program.su

# Compile and execute in one step (requires the [runtime] extra installed).
sutrac --run path/to/program.su

# Step-by-step compilation review (source → AST → inline → simplify → emit).
sutrac --review path/to/program.su

# Cross-file class-name casing check.
sutrac --consistency path/to/directory

sutrac --help lists every flag.

What is Sutra

A .su source file looks like TypeScript — functions, classes, && / ||, string and numeric literals — but every value is a vector in a frozen LLM embedding space, and every operation is a tensor op on those vectors. The compiler emits self-contained Python that calls a small runtime; chains of bind / unbind / bundle / similarity reduce to chains of matrix multiplies that the simplifier folds into cached matrices at compile time.

The result is a program where there are no Python branches, no host-side if / while on data values, and no JIT — just one straight-line tensor expression.

For a full introduction, see sutra.topazcomputing.com. For the language specification and design rationale, see the Sutra repository.

Status

Research-grade. Versions before 1.0 may break source compatibility. The grammar is stable; the codegen and the standard library still move.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

sutra_dev-1.1.0.tar.gz (541.6 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

sutra_dev-1.1.0-py3-none-any.whl (389.7 kB view details)

Uploaded Python 3

File details

Details for the file sutra_dev-1.1.0.tar.gz.

File metadata

  • Download URL: sutra_dev-1.1.0.tar.gz
  • Upload date:
  • Size: 541.6 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for sutra_dev-1.1.0.tar.gz
Algorithm Hash digest
SHA256 ec85a4e49814de968f434dde493f80f9f47595ba8e29612e190099cdfb3aee9c
MD5 3d43af65fa388bcd9c8c3f6f5ea0c24b
BLAKE2b-256 f7de496211aeb37418c4359c619d75c5ea92226c880799b898fe088a04254a1c

See more details on using hashes here.

Provenance

The following attestation bundles were made for sutra_dev-1.1.0.tar.gz:

Publisher: publish-sutra-compiler.yml on EmmaLeonhart/Sutra

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

File details

Details for the file sutra_dev-1.1.0-py3-none-any.whl.

File metadata

  • Download URL: sutra_dev-1.1.0-py3-none-any.whl
  • Upload date:
  • Size: 389.7 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? Yes
  • Uploaded via: twine/6.1.0 CPython/3.13.14

File hashes

Hashes for sutra_dev-1.1.0-py3-none-any.whl
Algorithm Hash digest
SHA256 dae15eb5e30eaa970393fb3e7e71e12028def60ad8b584f7f08364468de97cad
MD5 948b602bb0989bcca51c5c0190dc3c75
BLAKE2b-256 e9f1c6cfd9f99eb0872d433152a814ac76ee9ebbfeb3ed1fd6e459b03eeba8f7

See more details on using hashes here.

Provenance

The following attestation bundles were made for sutra_dev-1.1.0-py3-none-any.whl:

Publisher: publish-sutra-compiler.yml on EmmaLeonhart/Sutra

Attestations: Values shown here reflect the state when the release was signed and may no longer be current.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page