lishunyang12/vllm-omni-pr-dashboard
vLLM-Omni PR dashboard: contributors, recent submissions, and topics
Repositories
lishunyang12/vllm-omni-rankings
lishunyang12/vllm-omni-sm120-skip-softmax-results
Reproducible SM120 FlashInfer skip-softmax validation for vLLM-Omni: H3 and Wan videos, gated threshold 0.5 / until 0.94, Nsight traces and raw test data.
lishunyang12/h3-optimizations
H3/FastH3 inference optimizations: AdaLN, MXFP8/SwiGLU, Sage/VSA, VAE and RDMA, with DGX Spark experiment guidance.
lishunyang12/fasth3-lossy-ablation
FastH3 single-factor lossy ablations: seven prompts, five variants, 35 original videos.
lishunyang12/openvdn-lossy-ablation
OpenVDN single-factor lossy ablations: 40 videos, 2 baseline repeats, and 32 paired comparisons.
lishunyang12/FastVideo
A unified inference and post-training framework for accelerated video generation.
lishunyang12/vllm-project.github.io
lishunyang12/fast-ulysses
Ulysses sequence-parallel all-to-all as a torch custom op, moved by the GPU copy engines into torch symmetric memory. Zero SM usage; 1.66-2.17x over torch.distributed on NVLink.
lishunyang12/vllm-gguf-plugin
vLLM Quantization plugin for GGUF
lishunyang12/live-streaming-tools
lishunyang12/diffuse-compressor
Model-agnostic SVDQuant toolkit for diffusion transformers
lishunyang12/lishunyang12.github.io
lishunyang 的技术博客 — vLLM-Omni / 推理
lishunyang12/recipes
Common recipes to run vLLM
lishunyang12/dynamo
A Datacenter Scale Distributed Inference Serving Framework
lishunyang12/flash-attention
Fast and memory-efficient exact attention
lishunyang12/cosmos3-playground
Web playground (vLLM-Omni plugin) for NVIDIA Cosmos3 world-foundation models
lishunyang12/JoyAI-VL-Interaction
lishunyang12/Lance
A 3B-active-parameter native unified multimodal model for image and video understanding, generation, and editing.
lishunyang12/SRAM-Analysis---Static-Noise-Margin-Generation-tool
This Python-based tool provides visualization support for Cadence Virtuoso analog simulations by generating Voltage Transfer Characteristic (VTC) graphs and calculating Static Noise Margin (SNM) for SRAM cells.
lishunyang12/vllm-omni
A framework for efficient model inference with omni-modality models
lishunyang12/how-to-optim-algorithm-in-cuda
how to optimize some algorithm in cuda.
lishunyang12/vllm
A high-throughput and memory-efficient inference and serving engine for LLMs
lishunyang12/flashinfer
FlashInfer: Kernel Library for LLM Serving
lishunyang12/fa-fwd
Flash-Attention-3 forward kernel
lishunyang12/vllm-omni-skills
a collection of skills for vllm-omni
lishunyang12/auto-round
SOTA rounding-based quantization for high-accuracy low-bit LLM inference, seamlessly optimized for CPU/XPU/CUDA, with multi-datatype support and full compatibility with vLLM, SGLang, and Transformers.
lishunyang12/Model-Optimizer
A unified library of SOTA model optimization techniques like quantization, pruning, distillation, speculative decoding, etc. It compresses deep learning models for downstream deployment frameworks like TensorRT-LLM, TensorRT, vLLM, etc. to optimize inference speed.
lishunyang12/Capstone_ML
lishunyang12/sglang
SGLang is a high-performance serving framework for large language models and multimodal models.