a100

Using the famous cnn model in Pytorch, we run benchmarks on various gpu.

Open source compiler for your whole ML repo (support python, CUDA C/C++) into sass and cuda binary (which are closed sourced by Nvidia); 3-40% faster than Openai Triton for Neural Network (primarily LLM) based training/inferencing

Qwen3.8-27B on one A100: up to 14.2x single-turn and 8.2x multi-turn vs bf16 on stock vLLM. Four profiles tuned per use case (c1m, c1s, c32m, c32s). Drop-in plugin for vLLM 0.31.0. Models: huggingface.co/atdev-company

DeepSeek-V4.1-Flash inference on NVIDIA Ampere with vLLM, SM80 patches, reproducible throughput and coding-agent benchmarks. Tested on 8x A800 80GB.

Remove Backgroud from a video 🧤🎥

A kubernetes operator for managing nvidia MIG instances.

Z-AI V1.4 发行版本

LLM benchmarking, GPU workload orchestration backend server | Deepseek-R1, Qwen2.5, Llama3.1 | 4xRTX-5090 inside PRU2500, 2xH100 inside PRU2500, 8xMI210 in SuperMicro

Dashboard for AI Studio, Open Source Continuous Inference | Deepseek-R1, Qwen2.5, Llama3.1 | 4xRTX-5090 inside PRU2500, 2xH100 inside PRU2500, 8xMI210 in SuperMicro

CLI tool to check Oracle Cloud compute shape availability - find capacity across regions

Dataset and code for "Coarse-Grained Density Functional Theory Predictions via Deep Kernel Learning"

ClusterOps is an enterprise-grade Python library developed and maintained by the Swarms Team to help you manage and execute agents on specific CPUs and GPUs across clusters. This tool enables advanced CPU and GPU selection, dynamic task allocation, and resource monitoring, making it ideal for high-performance distributed computing environments.

Offline DeepSeek-V4.1-Flash deployment for 8x A100: DSpark k5, 256K context, local APIs and measured benchmarks.

Implement named entity recognition (NER) using regex and fine-tuned LLM, with a total of 15 categories. The ultimate goal is to apply the model to detect personally identifiable information (PII) in student writing.

Verified Indiana University Big Red 200 and Quartz HPC reference for AI coding agents — SLURM, modules, workflow recipes.

DeepSeek-V4-Flash-Vision-Exp multimodal serving on 4× CMP 170HX (unlocked GA100, sm_80, PP4, 512k ctx, DSpark)

DeepSeek V4.1 Flash 1017 on 8xA100/SM80: vLLM adaptations, official DeepSeek recipe tool protocol, C32/512K acceptance and reported benchmarks.

Text to Image Gen [ Demo ]

Pinned DeepSeek-V4.1-Flash-1017 adaptation and evidence-based performance experiments for 8x A100 SXM4 80GB.

Found out that using A100 and V100 on Vicuna and Llama2 have a different result, while other model such as Falcon doesn't has such question.