Younes Belkada

Open-source work

Several of Belkada's papers concern running and fine-tuning large models with fewer resources. Belkada co-authored LLM.int8() (2022) with Tim Dettmers, Mike Lewis and Luke Zettlemoyer, which introduced 8-bit matrix multiplication for transformer inference, and Petals (2022), a system for running and fine-tuning large models collaboratively across many machines. Belkada also co-taught the DeepLearning.AI short course "Quantization Fundamentals with Hugging Face" with Marc Sun.⁵⁷⁴

Belkada is listed among the authors of TRL, Hugging Face's library for post-training transformer models with methods such as supervised fine-tuning, reward modeling, DPO and GRPO, and co-authored the Zephyr paper (2023), which aligned a smaller chat model using distilled direct preference optimization. Belkada was also a co-author of "StarCoder 2 and The Stack v2" (2024), on open code models and their training data.¹²³

Falcon models

Later papers concern the Falcon family of language models. Belkada co-authored the technical reports for Falcon Mamba 7B (2024), a base model built on the Mamba architecture without attention, and Falcon-H1 (2025), a series of models with a parallel hybrid of attention and state-space layers. A 2026 paper on learnable multipliers for matrix layers lists Belkada with the Falcon LLM Team.⁸⁹⁶

Work and contributions

• TRL (Transformers Reinforcement Learning) Open-source code

  Listed among the authors in TRL's citation file; TRL is an Apache-2.0 toolkit for post-training transformer models with SFT, reward modeling, DPO, and GRPO.¹

• StarCoder2 Published research

  Co-author of "StarCoder 2 and The Stack v2: The Next Generation" (2024).³

• Zephyr Published research

  Co-author of "Zephyr: Direct Distillation of LM Alignment" (2023), on aligning a smaller chat model with distilled direct preference optimization.²

• LLM.int8() Published research

  Co-author, with Tim Dettmers, Mike Lewis, and Luke Zettlemoyer, of "LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale" (2022); the abstract says the software is open source.⁵

Timeline

1. Aug 2022Co-authored "LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale".⁵
2. Sep 2022Co-authored "Petals", on collaborative inference and fine-tuning of large models.⁷
3. Oct 2023Co-authored "Zephyr: Direct Distillation of LM Alignment".²
4. Feb 2024Co-authored "StarCoder 2 and The Stack v2: The Next Generation".³
5. Oct 2024Co-authored the Falcon Mamba 7B technical report.⁸
6. Jul 2025Co-authored the Falcon-H1 technical report on hybrid-head language models.⁹
7. Jan 2026Co-authored "Learnable Multipliers" with a Falcon LLM Team affiliation.⁶

Documented roles

The sources do not confirm a current role; these are the roles they document.

• Machine Learning Engineer, Hugging Face⁴
• Listed affiliation on the 2026 paper on learnable multipliers for language model matrix layers, Falcon LLM Team⁶

Companies in their history

• Hugging Face 34 people in the wiki

Related people

• Edward BeechingConnected to Hugging Face
• Leandro von WerraConnected to Hugging Face
• Lewis TunstallConnected to Hugging Face
• Loubna Ben AllalConnected to Hugging Face
• Hynek KydlíčekConnected to Hugging Face
• Waseem AlShikhLanguage models · Inference and serving
• Azalia MirhoseiniLanguage models · Inference and serving
• Wojciech ZarembaReinforcement learning · Language models
• Susan ZhangLanguage models · Reinforcement learning
• Alexander WettigLanguage models · Reinforcement learning