About the job
Cerebras Systems is at the forefront of AI technology, creating the world's largest AI chip, 56 times the size of traditional GPUs. Our innovative wafer-scale architecture delivers the computational power of dozens of GPUs on a single chip while simplifying programming to the ease of a single device. This groundbreaking approach enables us to achieve unparalleled training and inference speeds, empowering machine learning professionals to seamlessly execute large-scale ML applications without the complexities of managing numerous GPUs or TPUs.
Cerebras serves an impressive clientele that includes top model laboratories, multinational corporations, and pioneering AI-native startups. Recently, OpenAI announced a multi-year partnership with Cerebras, deploying 750 megawatts of scale to revolutionize critical workloads with ultra-high-speed inference.
Our wafer-scale architecture also powers the fastest Generative AI inference solution globally, exceeding GPU-based hyperscale cloud inference services by over 10 times. This significant enhancement in speed is transforming the user experience of AI applications, facilitating real-time iteration and boosting intelligence through additional computational capabilities.
About The Role
We are in search of a talented Compiler Engineer to contribute to the design and implementation of new features within our CSL (Cerebras Software Language) compiler. CSL is a Zig-like programming language utilized both internally and externally to program our wafer-scale engine (WSE).
The language offers high-level abstractions to simplify programming the wafer WSE while providing low-level access to hardware internals for optimal hardware utilization. The compiler leverages MLIR infrastructure to translate CSL into LLVM IR, which is further compiled by a dedicated LLVM mid-end/backend into executable files.
Responsibilities
- Design and implement front-end language features, semantic analysis, intermediate representations, and lowering pipelines from CSL to MLIR dialect(s) and LLVM IR.
- Develop and enhance abstraction layers between the CSL language and the underlying hardware.
