Skip to content
@firnlabs

Firn Labs

Minimal reference systems and RL implementations for LLM post-training
  • United States of America
README.md

Firn Labs

Independent research workspace exploring modern LLM post-training, distributed systems, reinforcement learning, and tooling for benchmarking and analysis.

Pinned Loading

  1. vrollout vrollout Public

    Asynchronous LLM rollout engine for distributed RL training.

    Python

  2. rl-kernels rl-kernels Public

    Fused Triton GPU kernels for RL post-training (DPO, cross-entropy).

    Python

  3. minigpt minigpt Public

    Decoder-only transformer with KV caching in PyTorch.

    Python

  4. minigrad minigrad Public

    Lightweight reverse-mode autograd engine and neural network library from scratch.

    Python

  5. minifsdp minifsdp Public

    Fully Sharded Data Parallel (FSDP2 / ZeRO-3) sharding engine in PyTorch.

    Python

Repositories

Showing 7 of 7 repositories

Top languages

Loading…

Most used topics

Loading…