<<<<<<< HEAD

solo-export

A simple CLI for exporting AI models in multiple formats, quantization schemes, and device targets, with built-in evaluation.

---

Features

• Export Formats:

  • ONNX
  • TorchScript
  • Hugging Face Hub
  • TensorRT (NVIDIA)
  • TFLite (TensorFlow Lite)
  • Core ML (Apple)

• Quantization:

  • FP16
  • INT8

• Device Targets:

  • CPU
  • CUDA
  • (extendable to other backends)

• CLI Scripts:

  • solo-export — run all exports + eval
  • solo-export-onnx — ONNX only
  • solo-export-ts — TorchScript only
  • solo-export-eval — Evaluate exports

• Evaluation: stub evaluations per export artifact for metrics like accuracy, latency, perplexity

---

Installation

pip install solo-exporter

---

Configuration (config/export.yaml)

Create config/export.yaml:

model:
  base: gpt2

export:
  formats:
    - onnx
    - torchscript
    - hf
    - tensorrt
    - tflite
    - coreml
  quantization:
    - fp16
    - int8

devices:
  - cpu
  - cuda

output_dir: output

evaluation:
  dataset: path/to/eval_dataset.jsonl
  metrics:
    - accuracy
    - latency
    - perplexity

---

Usage

Run all exports and evaluations

solo-export --config config/export.yaml

Export ONNX only

solo-export-onnx --config config/export.yaml

Export TorchScript only

solo-export-ts --config config/export.yaml

Run evaluations only

solo-export-eval --config config/export.yaml

Results will be saved under output/ in subfolders by format, quantization, and device.

---

Example Output Structure

output/
├── gpt2_fp16_onnx_cpu/
│   ├── model.onnx
│   └── metrics.json
├── gpt2_int8_ts_cuda/
│   ├── model.pt
│   └── metrics.json
└── ...

---

=======

Solo-Export

6b862b48f4da0a4e9869226c9334978206af944f