RLsys-Foundation / APRILLinks
APRIL: Active Partial Rollouts in Reinforcement Learning to Tame Long-tail Generation. A system-level optimization for scalable LLM training.
☆44Updated 2 months ago
Alternatives and similar repositories for APRIL
Users that are interested in APRIL are comparing it to the libraries listed below
Sorting:
- 🔥 LLM-powered GPU kernel synthesis: Train models to convert PyTorch ops into optimized Triton kernels via SFT+RL. Multi-turn compilation…☆105Updated last month
- Official Implementation of SAM-Decoding: Speculative Decoding via Suffix Automaton☆38Updated 10 months ago
- Implementation for FP8/INT8 Rollout for RL training without performence drop.☆280Updated last month
- LongSpec: Long-Context Lossless Speculative Decoding with Efficient Drafting and Verification☆68Updated 5 months ago
- Accelerate LLM preference tuning via prefix sharing with a single line of code☆51Updated 5 months ago
- The official implementation of paper: SimLayerKV: A Simple Framework for Layer-Level KV Cache Reduction.☆52Updated last year
- Ouroboros: Speculative Decoding with Large Model Enhanced Drafting (EMNLP 2024 main)☆112Updated 8 months ago
- ☆132Updated 6 months ago
- ☆122Updated 6 months ago
- A lightweight reinforcement learning framework that integrates seamlessly into your codebase, empowering developers to focus on algorithm…☆91Updated 3 months ago
- Odysseus: Playground of LLM Sequence Parallelism☆78Updated last year
- GEAR: An Efficient KV Cache Compression Recipefor Near-Lossless Generative Inference of LLM☆170Updated last year
- Pytorch implementation of our paper accepted by ICML 2024 -- CaM: Cache Merging for Memory-efficient LLMs Inference☆47Updated last year
- ☆47Updated 6 months ago
- This repo contains the source code for: Model Tells You What to Discard: Adaptive KV Cache Compression for LLMs☆42Updated last year
- Code for paper: [ICLR2025 Oral] FlexPrefill: A Context-Aware Sparse Attention Mechanism for Efficient Long-Sequence Inference☆156Updated 2 months ago
- 16-fold memory access reduction with nearly no loss☆109Updated 8 months ago
- An efficient implementation of the NSA (Native Sparse Attention) kernel☆126Updated 5 months ago
- Source code for the paper "LongGenBench: Long-context Generation Benchmark"☆24Updated last year
- ☆105Updated 3 months ago
- Async pipelined version of Verl☆125Updated 8 months ago
- PoC for "SpecReason: Fast and Accurate Inference-Time Compute via Speculative Reasoning" [NeurIPS '25]☆60Updated 2 months ago
- [ICLR 2025] PEARL: Parallel Speculative Decoding with Adaptive Draft Length☆135Updated last month
- ☆203Updated 3 weeks ago
- [ICLR2025] Breaking Throughput-Latency Trade-off for Long Sequences with Speculative Decoding☆135Updated last year
- ☆56Updated 6 months ago
- Bridge Megatron-Core to Hugging Face/Reinforcement Learning☆172Updated this week
- [ICLR 2025] TidalDecode: A Fast and Accurate LLM Decoding with Position Persistent Sparse Attention☆49Updated 4 months ago
- Kinetics: Rethinking Test-Time Scaling Laws☆84Updated 5 months ago
- [ICLR 2025] DeFT: Decoding with Flash Tree-attention for Efficient Tree-structured LLM Inference☆45Updated 5 months ago