ML Design Engineer

About this role:

We are looking for an exceptional machine learning software engineer or computer scientist with experience building, deploying, and supporting machine learning kernels and models as part of our Design Acceleration Team (DAT). This team will deliver state of the art results on Groq's Tensor Streaming Processor using the metrics which are of importance to our customers. As a result, you'll work closely with our compiler team to estimate and optimize performance, our business and sales teams to assist Groq in focusing its efforts, and the customer to demonstrate results in their environment. Given the broad nature of skills needed to exceed customer expectations, we expect you to excel at either ML optimization or 'full stack' skills, but any mix is also welcome.

In order to provide the best customer experience in your role, you will be tasked with the following responsibilities:

Responsibilities & opportunities in this role:

• Lead technical engagements with customers by taking their models and associated runtimes and get it running within the Groq environment. Subsequently you will need to estimate performance and identify opportunities for improvement.
• Understand and become an expert in the Groq architecture, either at a system level for those with a full stack background, or being able to optimally program a spatial architecture for those with an optimization background.
• Collaborate with internal Compiler teams or Kernel engineers to improve performance of key customer designs.
• Work with customers to prepare models for acceleration with Groq, including the partitioning of models between CPU and TSP, convert between various model formats, identify system bottlenecks and assisting customers implementing model parallel approaches across multiple devices.

Ideal candidates have/are:

• Technically curious problem solver with the ability to quickly understand complex technical challenges, synthesize and articulate back. Understands the fundamentals of AI (CNN, RNN, transformers, MLPs, etc) and has experience applying them broadly. Knows how to structure problems to maximize the performance for at least one accelerator architecture.
• Demonstrated problem solving skills with the ability to formulate a strategy to incrementally achieve a goal while operating as part of a team
• Prior experience working with CV, L4/L5 Autonomous, NLP, DLRM and/or GANs a plus.
• Comfortable engaging directly with customers and operating with integrity and drive transparency, openness, and effective communication.
• The ideal skills for the optimization and full stack roles are shown as an example. An applicant can have one or the other, or a mix of the two:
• For optimization centric roles: (Focus on ML Framework and below)
• Demonstrated capability to take a model at the algorithm level and implement it in a low level language at a high level of performance
• Ability to split a model between CPU and TSP, or break models into pieces for model parallel approaches. Work with customers to make necessary adjustments.
• Be able to retrain models to accommodate any model changes for optimization
• Algorithm programming using languages such as OpenCL, CUDA, OneAPI, DSA languages or HDL and optimizing algorithms in tools like Numpy or Matlab.
• For full stack centric roles: (Focus on ML Framework and above)
• Comfortable working with either TF, Pytorch or ONNX, with an expertise in one.
• Full stack experience with any of a variety of frameworks (e.g. OpenVINO, gRPC, Spark.ML). Key is the ability to adapt quickly to a customer's flow and iterate.
• Desired capability to take a model at the algorithm level and implement it in any ML framework.
• Understanding and comfort using a variety of deployment approaches like Tensorflow Serving. Comfortable with containers/SLURM.
• Build test applications to measure performance, accuracy. Support retraining/fine tuning of models for the Groq Architecture.

Qualifications for this role:

• BS in CS, CE/EE, Math, or Physics or equivalent work experience.
• 5+ years machine learning or similar algorithmic/computational experience in model development or deployment.
• Recent MS/PhD computer science, math or engineering graduates with extensive hands-on experience with coding machine learning or DSP algorithms will also be considered.
• Customer first mindset. Driven to meet or exceed customer expectations, resolve issues expeditiously and, when possible, proactively.
• Excellent leadership, mentoring and cross-functional collaborative and influencing skills.
• Effective communication & presentation skills and comfortable in a customer-facing environment.
• Able to work in a very dynamic start-up environment.
• Pending reopening of businesses, willingness to travel to customer locations to support bring-up, debug, and porting efforts. Some roles may require up to 25% travel during peak on-site bring-up and support periods (i.e. one week per month).