liodon-ai/Puro-2B-Base-ONNX

Puro-2B-Base — ONNX

ONNX export of thu-pacman/Puro-2B-Base, published by Liodon AI. Exported with optimum (optimum.exporters.onnx.main_export, task text-generation-with-past, so the graph exposes past-key-value inputs/outputs for KV-cached autoregressive decoding).

Files

File	Size	Notes
model.onnx	8.13 GB	FP32, full precision
model_fp16.onnx	4.30 GB	FP16, for GPU execution providers
model_quantized.onnx	2.03 GB	Dynamic INT8 (weight-only, no calibration)

Quick Start

import onnxruntime as ort
from transformers import AutoTokenizer

tok = AutoTokenizer.from_pretrained("liodon-ai/Puro-2B-Base-ONNX")
sess = ort.InferenceSession("model_quantized.onnx", providers=["CPUExecutionProvider"])
# past_key_values.*.key / .value inputs must be supplied (zero-length tensors
# for the first forward pass) -- see optimum's ORTModelForCausalLM for a
# ready-made wrapper that handles KV-cache bookkeeping automatically:
#   from optimum.onnxruntime import ORTModelForCausalLM
#   model = ORTModelForCausalLM.from_pretrained("liodon-ai/Puro-2B-Base-ONNX", file_name="model_quantized.onnx")

Source

• Model: thu-pacman/Puro-2B-Base
• License: other

Citation

@misc{liodonai_puro_2b_base_onnx,
  title        = {Puro-2B-Base — ONNX},
  author       = {{Liodon AI}},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/liodon-ai/Puro-2B-Base-ONNX}},
  note         = {ONNX export of thu-pacman/Puro-2B-Base}
}

---

Exported by Liodon AI