Skip to main content
TALENT247 // SPECIALIST AI EXECUTIVE SEARCH & TECHNICAL RECRUITMENT
ML · Machine Learning

Senior Open-Source Post-Training & Alignment ML Engineer

Paris, France · Full-time · Hybrid

AI Specialism

Machine Learning

Workplace

Hybrid

Location

Paris, France

Compensation

€95k – €135k

annual package

Role Dossier & Specification

REF: SENIOR-OPEN-SOURCE

THE CLIENT & MANDATE

Our client is the Paris-anchored steward of the open-source AI ecosystem whose model hub, dataset repositories, and post-training libraries form the daily foundation for hundreds of thousands of ML engineers and academic researchers worldwide. Whenever a new alignment paper, reasoning RL algorithm, or memory-saving kernel drops on arXiv, their open-source post-training team turns it into clean, reproducible, multi-GPU production code adopted by the entire industry within days.

They have retained Talent247 to recruit a Senior Open-Source Post-Training & Alignment ML Engineer in Paris. Joining their Open Post-Training & Alignment Libraries team, you will co-maintain flagship open-source alignment toolkits, run reference post-training recipes on dedicated H100 clusters, and democratize frontier reasoning and multimodal RL techniques for the global builder community.

KEY RESPONSIBILITIES

  • Design, implement, and maintain modular, high-performance PyTorch implementations of state-of-the-art post-training algorithms—including SFT, DPO, ORPO, GRPO, online RLOO/PPO, and verifiable-reward reasoning loops—across text and vision-language models.
  • Integrate memory- and compute-efficiency breakthroughs (FSDP2, DeepSpeed ZeRO, LoRA/QLoRA, Liger/Triton fused cross-entropy kernels, sequence packing, and vLLM async rollout co-location) so users train larger models on fewer GPUs.
  • Execute end-to-end open post-training reproductions on multi-node H100 Slurm clusters, publishing open recipes, training logs, datasets, and aligned checkpoints that match or beat closed baselines.
  • Build clean integration bridges between post-training trainers, sandboxed code/math verifiers, and standardized evaluation harnesses.
  • Review community pull requests, shape clean Pythonic APIs used by millions of monthly downloads, and author clear technical documentation and research blog posts.
  • Collaborate directly with hardware vendors and open-model labs to ensure day-zero post-training support for newly released model architectures.

TECHNICAL ENVIRONMENT & OFFERING

Their stack centers on Python, PyTorch, FSDP/Accelerate/DeepSpeed, Triton fused kernels, vLLM, Slurm H100 clusters, and open GitHub development. Working hybrid in Paris, our client offers €95,000–€135,000 EUR base salary plus equity options, full French benefits, conference travel, and global visibility for your technical work.

Core Requirements

ESSENTIAL QUALIFICATIONS

  • 5+ years of hands-on machine learning engineering or applied research experience with deep expertise in PyTorch internals and transformer training dynamics.
  • Direct experience implementing and debugging LLM post-training methods (SFT, DPO, RLHF, GRPO/PPO, distillation) across multi-GPU / multi-node setups.
  • Proven open-source library authorship or core maintainership—you care deeply about clean abstractions, backward compatibility, unit tests, and crisp documentation.
  • Strong command of distributed GPU memory optimization (mixed precision BF16/FP8, activation checkpointing, FSDP/DeepSpeed, PEFT/quantization).
  • Clear technical writing skills for authoring reproducible recipes and engaging constructively with open-source contributors.

PREFERRED / BONUS EXPERIENCE

  • Experience co-locating fast inference engines (vLLM/SGLang) inside online RL training loops or writing custom Triton kernels for post-training losses.
  • Public portfolio of widely used open-source ML repos, aligned model releases, or technical papers/reports.

Position Telemetry

Location Paris, France (Hybrid)
Department Machine Learning
Employment type Full-time
Workplace type Hybrid
Salary range
€95k – €135k annual package