onnxruntime-bnlang

Bnlang binding for ONNX Runtime.

Unofficial third-party binding. Not affiliated with Microsoft. Consumes official ORT prebuilt releases.

এই README-এর বাংলা সংস্করণ — README.md

Quick start

import "onnxruntime-bnlang" as ort;

print(ort.version);
print(ort.providers());

var sess = ort.InferenceSession.open("model.onnx", { log_severity_level: 3 });

var info = sess.info();
var out  = sess.run({
    "input_ids":      { dtype: "int64", shape: [1, 16], data: [...] },
    "attention_mask": { dtype: "int64", shape: [1, 16], data: [...] }
});
// out["logits"] is { dtype, shape, handle } — materialize with tensor_to_data:
var logits = ort.tensor_to_data(out["logits"]["handle"]);
ort.tensor_close(out["logits"]["handle"]);

sess.close();

Exports

Status (v1.0.0)

Working:

• Session lifecycle: open / info / run / close
• Tensor dtypes: float32, float64, float16, int64, int32, uint8, bool
• Session options: log_severity_level, intra_op_num_threads, inter_op_num_threads, execution_providers
• Tensor handle table (zero-copy KV-cache reuse for downstream LLM loops)

Coming next:

• More execution providers (CUDA, CoreML)
• Fetch-list filtering (pick which outputs to compute)
• Profiling + IO binding

Build (local)

# Windows
bnl script/install.bnl    # download ORT prebuilt into deps/windows-x64/
.\build.ps1                # cmake configure + build  ->  build/windows-x64/*.dll

# macOS / Linux
bnl script/install.bnl
./build.sh

Cross-platform packaging

Built per platform; the bnl runtime picks the right artifact from bnl.json's targets map at import time:

LOAD_WITH_ALTERED_SEARCH_PATH in bnl core ensures the plugin's own directory is searched first for its dependencies, ahead of System32 / PATH.

Layout

bnl.json                 manifest (main + targets map)
CMakeLists.txt           build config
CMakePresets.json        one preset per platform

lib/
  index.bnl              public API (English + Bangla re-exports)
  session.bnl            InferenceSession + session-instance methods

src/                     C++ binding (excluded from published tarball)
  bnl/plugin.h           C ABI contract
  main.cpp               bnl_load entry
  session.{h,cpp}        Ort::Session handle table
  tensor.{h,cpp}         bnl_value <-> Ort::Value + handle table
  session_options.{h,cpp}

deps/<triple>/           ORT prebuilt (gitignored; install populates)
build/<triple>/          cmake output (gitignored)

script/
  install.bnl            download + verify + extract ORT prebuilt
  install-metadata.bnl   URLs + sha256 per platform
  build.bnl              thin wrapper around `cmake --preset`

License

MIT. ORT prebuilts are MIT-licensed by Microsoft. See NOTICES.md for third-party attribution.

MIT License

Copyright (c) 2026 Bnlang | Mamun

Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to furnish persons to whom the Software is
furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.

THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.

Third-party notices — onnxruntime-bnlang

This package's binary plugin dynamically links to and consumes prebuilt binaries from the projects below. They are downloaded at install time by script/install.bnl from their official distribution channels (NuGet, GitHub Releases) and remain governed by their original licenses.

ONNX Runtime

• Project: Microsoft ONNX Runtime
• Upstream: https://github.com/microsoft/onnxruntime
• License: MIT
• Copyright: Copyright (c) Microsoft Corporation

Used as the inference engine. We download Microsoft's prebuilt onnxruntime.dll / onnxruntime_providers_shared.dll at install time and dynamically link to it from our plugin. We do not redistribute the ORT binaries; users fetch them from Microsoft's official packages.

The full MIT license text is available at https://github.com/microsoft/onnxruntime/blob/main/LICENSE

DirectML

• Project: DirectML
• Upstream: part of Microsoft Windows (since Windows 10 1903)
• License: Microsoft Software License Terms (Windows component)

DirectML.dll is shipped with Windows itself and is not redistributed by this package. It is loaded from the system when execution_providers includes "DML".