Shush is an app that deploys a WhisperV3 model with Flash Attention v2 on Modal and makes requests to it via a NextJS app
-
Updated
Jun 7, 2024 - TypeScript
#
flash-attention-2
Here are 29 public repositories matching this topic...
Implementation of FlashAttention (FA1-FA4) in PyTorch for educational and algorithmic clarity
-
Updated
Apr 12, 2026 - Python
Triton implementation of FlashAttention2 that adds Custom Masks.
deep-learning triton attention cuda-kernels attention-mechanism triton-lang flash-attention flash-attention-2
-
Updated
Aug 14, 2024 - Python
🎭 Real-time voice-controlled 3D avatar with multimodal AI - speak naturally and watch your AI companion respond with perfect lip-sync
websocket nextjs vlm fastapi huggingface whisper-ai flash-attention-2 multimodal-ai kokoro-tts smolvlm
-
Updated
May 23, 2026 - TypeScript
Performance of the C++ interface of flash attention and flash attention v2 in large language model (LLM) inference scenarios.
gpu cuda inference nvidia cutlass mha multi-head-attention llm tensor-core large-language-model flash-attention flash-attention-2
-
Updated
Feb 27, 2025 - C++
A flexible and efficient implementation of Flash Attention 2.0 for JAX, supporting multiple backends (GPU/TPU/CPU) and platforms (Triton/Pallas/JAX).
-
Updated
Mar 4, 2025 - Python
Cross-platform FlashAttention-2 Triton implementation for Turing+ GPUs with custom configuration mode
machine-learning deep-learning gpu optimization pytorch triton attention hopper implementation attention-mechanism turing ampere implementation-from-scratch blackwell transfromers large-language-models llm flash-attention flash-attention-2 flashattention
-
Updated
Jan 12, 2026 - Python
A cutlass cute implementation of headdim-64 flashattentionv2 TensorRT plugin for LightGlue. Run on Jetson Orin NX 8GB with TensorRT 8.5.2.
cuda transformer cutlass cute tensorrt feature-matching multihead-attention superpoint lightglue flash-attention flash-attention-2
-
Updated
Mar 3, 2025 - Cuda
Vulkan & GLSL implementation of FlashAttention-2
vulkan glsl artificial-intelligence gpu-acceleration attention gpu-computing deel-learning tensor-cores large-language-models llm flash-attention flash-attention-2
-
Updated
Jan 19, 2025 - C++
FlashAttention for sliding window attention in Triton (fwd + bwd pass)
-
Updated
Jun 25, 2025 - Python
A Gradio-based demo application for comparing state-of-the-art OCR models: DeepSeek-OCR, Dots.OCR, HunyuanOCR, and Nanonets-OCR2-3B.
python ocr pillow torch accelerate supervision gradio opencv-python nanonets torchvision sentencepiece huggingface-transformers huggingface-spaces flash-attention-2 hunyuan qwen2-5-vl dots-ocr deepseek-ocr easydict
-
Updated
May 13, 2026 - Python
DeepSeek-OCR-2-Unlimited-OCR is an advanced, experimental visual document processing and open-ended text localization dashboard. This application establishes a unified interface that allows users to swap between two premier vision-language document models: deepseek-ai/DeepSeek-OCR-2 and baidu/Unlimited-OCR.
python ocr torch pytorch addict gradio ocr-text-reader torchvision huggingface-transformers tokenizers einops flash-attention-2 deepseek-v3 deepseek-ocr easydict deepseek-ocr-2
-
Updated
Jun 25, 2026 - Python
Tiny VLMs Lab is a Hugging Face Space and open-source project showcasing lightweight Vision-Language Models for image captioning, OCR, reasoning, and multimodal understanding. It offers a simple Gradio interface to upload images, query models, adjust generation settings, and export results in Markdown or PDF.
ocr cuda optical-character-recognition gradio multimodality captioning-images huggingface-transformers vision-transformer hugging-face huggingface-spaces vision-language-model flash-attention-2 vlms qwen2-5-vl
-
Updated
Nov 26, 2025 - Python
SAM3-Plus-Qwen3.5 is an advanced, experimental computer vision suite that seamlessly integrates Facebook's Segment Anything Model 3 (SAM3) with the Qwen3.5 multimodal reasoning engine.
python scikit-learn ninja pillow torch supervision gradio kernels opencv-python torchvision sam3 huggingface-transformers timm einops flash-attention-2 multimodal-model qwen3-5
-
Updated
May 13, 2026 - Python
A high-performance C++ backend for extreme-context LLM inference. It replaces item-count batching with dynamic, VRAM-aware First-Fit Decreasing (FFD) bin packing. By using PyBind11 for async queueing, 16-token alignment, and `cudaHostAlloc` for zero-copy FlashAttention-2 transfers, it mathematically eliminates OOMs and maximizes GPU throughput.
zero-copy bin-packing cpp17 memory-management pybind11 gpu-optimization llm-inference flash-attention-2
-
Updated
Jun 5, 2026 - Python
Transcribe audio in minutes with OpenAI's WhisperV3 and Flash Attention v2 + Transformers without relying on third-party providers and APIs. Host it yourself or try it out.
python modal transformers transcription wavesurfer-js nvidia-cuda bun nvidia-gpu virtual-environment fastapi huggingface-transformers flash-attention-2 next14 whisper- whisperv3
-
Updated
Jun 18, 2024 - TypeScript
Poplar implementation of FlashAttention for IPU
-
Updated
Mar 12, 2024 - C++
Core-OCR is an advanced, experimental Optical Character Recognition (OCR) and document analysis suite designed for highly accurate text extraction, table reconstruction, and complex visual reasoning. Built on the robust Qwen2.5-VL and Qwen2-VL multimodal architectures.
python torch pytorch gradio kernels ocr-recognition torchvision huggingface-transformers vision-transformer huggingface-models huggingface-spaces vision-language-model flash-attention-2 qwen2-vl qwen2-5-vl
-
Updated
Mar 23, 2026 - Python
TRELLIS.2-Text-to-3D-FA2 is an advanced, experimental 3D asset generation suite that couples high-speed text-to-image synthesis with structured 3D geometry reasoning. By linking Alibaba's rapid Tongyi-MAI/Z-Image-Turbo pipeline with Microsoft's state-of-the-art microsoft/TRELLIS.2-4B foundational 3D asset mode.
torch trellis triton gradio 3d-models torchvision image-to-3d huggingface-transformers huggingface-spaces diffusers text-to-3d flash-attention-2 rerun-sdk trellis2
-
Updated
May 24, 2026 - Python
Systematically train and benchmark Mistral, Qwen2.5, and SmolLM2 on essay grading across 39 experiments through data analysis and engineering, structured preprocessing, instruction tuning, postprocessing, and leakage aware evaluation for robust score and rationale generation
modeling evaluation transformers data-engineering data-analysis lora preprocessing postprocessing automatic-essay-scoring instruction-tuning supervised-finetuning flash-attention-2 mistral-7b unsloth qwen2-5 smollm2
-
Updated
Aug 14, 2025 - Jupyter Notebook
Improve this page
Add a description, image, and links to the flash-attention-2 topic page so that developers can more easily learn about it.
Add this topic to your repo
To associate your repository with the flash-attention-2 topic, visit your repo's landing page and select "manage topics."