andrzej/SmolLM2-135M-Instruct-ONNX

SmolLM2-135M-Instruct (ONNX, mirror)

Source: HuggingFaceTB/SmolLM2-135M-Instruct by Hugging Face TB. Upstream describes it as supervised fine-tuning followed by Direct Preference Optimization (DPO) on UltraFeedback.

Conversion: none here. Every file is copied unchanged from HuggingFaceTB/SmolLM2-135M-Instruct at revision 12fd25f77366fa6b3b4b768ec3050bf629380bac, using the ONNX files published in that repository. Only the files the app loads are included: the int8 build (onnx/model_int8.onnx) and the fp32 build (onnx/model.onnx), plus config and tokenizer files. SHA-256 values were checked against the source repository.

This repository exists for in-browser demos with Transformers.js 4.3.2 (an educational app that compares base, SFT and SFT + DPO versions of SmolLM2-135M side by side). Language: English. The model can produce inaccurate or made-up text.

Licence: Apache 2.0, as the source repository. All credit for the model goes to its authors.

File	Bytes	SHA-256
config.json	861	8eb740e8bbe4cff95ea7b4588d17a2432deb16e8075bc5828ff7ba9be94d982a
generation_config.json	132	87b916edaaab66b3899b9d0dd0752727dff6666686da0504d89ae0a6e055a013
tokenizer.json	2,104,556	9ca9acddb6525a194ec8ac7a87f24fbba7232a9a15ffa1af0c1224fcd888e47c
tokenizer_config.json	3,764	4ec77d44f62efeb38d7e044a1db318f6a939438425312dfa333b8382dbad98df
onnx/model_int8.onnx	137,147,867	a7c33f9ef85d06734cc9d1f943f7e2ba57c77e769df727f4d3e217d9f672b0cc
onnx/model.onnx	540,345,794	1e790f1d6fac94097a6516ed1f07c29049097271d536d538562cd31ffc011a71

Citation (upstream)

@misc{allal2025smollm2smolgoesbig,
  title={SmolLM2: When Smol Goes Big -- Data-Centric Training of a Small Language Model},
  author={Loubna Ben Allal and Anton Lozhkov and Elie Bakouch and Gabriel Martín Blázquez and Guilherme Penedo and Lewis Tunstall and Andrés Marafioti and Hynek Kydlíček and Agustín Piqueres Lajarín and Vaibhav Srivastav and Joshua Lochner and Caleb Fahlgren and Xuan-Son Nguyen and Clémentine Fourrier and Ben Burtenshaw and Hugo Larcher and Haojun Zhao and Cyril Zakka and Mathieu Morlon and Colin Raffel and Leandro von Werra and Thomas Wolf},
  year={2025},
  eprint={2502.02737},
  archivePrefix={arXiv},
  primaryClass={cs.CL},
  url={https://arxiv.org/abs/2502.02737},
}