About the role
We are looking
for a Senior Software Engineer to develop and optimize deep learning models,
including CNNs, LLMs, and MoE, for efficient inference across CPU, GPU,
hardware accelerators, and edge devices. The role focuses on quantization,
model compression, high-performance kernel implementation, transformer
optimization, and production deployment.
Responsibilities:
• Develop and optimize deep learning models (CNNs,
LLMs, MoE) for efficient inference across CPU, GPU, hardware accelerators, and
edge devices.
• Design and implement quantization algorithms
(PTQ, QAT, GPTQ, AWQ) from scratch.
• Apply model compression techniques such as
pruning, decomposition, and distillation.
• Implement and optimize quantized kernels (INT8,
INT4, FP8) using C++ for high performance.
• Translate research papers into production-ready
implementations.
• Optimize latency, throughput, and memory usage
for real-world deployment.
• Work on transformer optimization including
KV-cache, PEFT (LoRA/QLoRA), and MoE models.
• Profile, benchmark, and debug model performance
across different hardware platforms.
Requirements
Education:
• 4+ years of relevant experience.
• Strong programming skills in Python and C++.
• Proven experience in quantization algorithms
(PTQ, QAT, GPTQ, AWQ).
• Hands-on experience in pruning, model
compression, and inference optimization.
• Experience implementing quantization or
optimization techniques from scratch.
• Strong understanding of CNNs, Transformers, and
LLM architectures.
• Experience with PyTorch / ONNX and model
deployment pipelines.
• Strong problem-solving and performance
optimization skills.
Need to have (Can be bridged):
No
additional bridged skills were specified.
Good to have (Not essential):
• Experience with MoE architectures and PEFT
techniques (LoRA, QLoRA).
• Knowledge of TensorRT, ONNX Runtime, TVM, and
MLIR.
• Familiarity with hardware-aware optimization
across GPU, NPU, and edge devices.
• Experience in research paper implementation or
open-source contributions.
Preferred Qualifications (Optional):
No
additional preferred qualifications were specified.