Daniel Han

Unsloth AI

Sources checked
ABOUT

Co-founder and Chief Scientist at Unsloth AI. Pioneered manually derived mathematical backward-pass GPU kernels in OpenAI Triton, enabling 80% lower VRAM usage and 2-5x faster training speeds for fine-tuning Llama, Mistral, Gemma, and DeepSeek foundation models on single GPUs.

Areas of focus

Professional niches

THE WORK BEHIND THE PROFILE

Proof of Work

explanationChecked Sep 21, 2026

Manual Backpropagation Triton Kernel Derivations for Parameter-Efficient Cross-Entropy

Published mathematical explanations and Triton code demonstrating how computing gradient chain rules directly inside custom SRAM kernels avoids materializing multi-gigabyte intermediate tensors in PyTorch.

Scope & limitations

Custom Triton kernels require deep mathematical derivations for each distinct attention and normalization layer.

Context: Triton cross-entropy loss, RoPE embedding kernels, RMSNorm gradient fusion.

View mission
implementationChecked Sep 21, 2026

Unsloth: Fast and Memory-Efficient Open-Source LLM Fine-Tuning

Created an open-source library that rewrites PyTorch forward and backward passes into handwritten Triton kernels, supporting LoRA, QLoRA, and full fine-tuning with 0% accuracy loss while preventing out-of-memory errors.

Scope & limitations

Heavily optimized for NVIDIA Ampere, Ada Lovelace, and Hopper architectures; AMD ROCm support is emerging.

Context: Llama 3.1/3.2, Mistral NeMo, DeepSeek-R1-Distill, Gemma 2; 4-bit bitsandbytes and 16-bit mixed precision.

View mission

Guides to evaluating AI expertise