Senior Machine Learning Engineer - Large Language Models

Key Skills

C Programming, C++, ChatGPT, Data science techniques, Machine learning techniques, Natural Language Processing (NLP), Python Programming, PyTorch, TensorFlow, Transformer

Job Description

SoundHound AI believes every person should be able to interact naturally with the products around them–by simply talking. With a global reach spanning two dozen languages, we build Voice AI products with conversational intelligence for cars, restaurant ordering, and more, allowing our customers to extend their brand in new and meaningful ways.

The work location for this role is limited to France. The workplace options for this role include in-office or hybrid (we have an office in Paris), or virtual. The compensation for this role includes base salary, equity (RSUs), and benefits.

In this role, you will:

• Develop and implement innovative neural architectures and pipelines for end-to-end ASR and large language models.
• Innovate in developing novel approaches and architectures for integration of large language models with end-to-end ASR to support prompting
• Perform data science/analytics 

We would love to hear from you if:

• You have strong foundations in the following areas
• AI/ML Fundamentals 
• Neural networks and transformers models in particular as applied to NLP
• Classical NLP
• Hyperparameter optimization methods such as Bayesian Optimization
• End-To-End-ASR
• Deep understanding of transformer models and associated architectures
• Knowledge of optimization techniques for improving the inference-time performance of transformer models is highly desirable
• Deep understanding of GPT (“Generative Pre-Trained Transformer”) models
• Familiarity with reinforcement learning (“RL”) techniques in LLM models (such as ChatGPT) and in particular Instruction following training methods such as those described in the InstructGPT paper
• Prompt engineering
• Large Language Models, training validation, testing and deployment
• GPT-3, ChatGPT, ChatGPT
• Knowledge of distributed training is highly desirable
• Prompt Engineering
• The ideal candidate will be on top of the latest research developments in the field.
• Demonstrated ability to develop efficient training pipelines using PyTorch (TensorFlow knowledge and experience is highly desirable)
• Experience with distributed training and inference (Model and Data parallelism) and the associated techniques necessary to train LLMs having tens of billions to 100s of billions of parameters
• E.g., familiarity with associated libraries such as DeepSpeed is highly desirable
• Knowledge of ideal deployment methods and technologies and relative tradeoffs MLOps
• General ML, Statistics and Data Science
• Experience and interest in statistical language models (“SLMs”) such as n-gram models is highly desirable
• Strong foundations in probability theory and statistics and understanding of concepts such as:
• bias/variance tradeoff
• model generalization
• bootstrapping methods
• Coding and Implementation

• Demonstrated facility with PyTorch (TensorFlow knowledge and experience is highly desirable
• Ability to write clean code in Python with strong design skills leveraging design patterns and object-oriented methods
• C/C++ coding skills are a strong plus