RTL implementation of a performance/area optimized bfloat16 adder and multiplication.
-
Updated
Mar 28, 2026 - Verilog
#
bf16
Here are 49 public repositories matching this topic...
Over 60 figures and diagrams of LLMs, quantization, low-rank adapters (LoRA), and chat templates FREE TO USE in your blog posts, slides, presentations, or papers.
supervised-learning data-types lora quantization fine-tuning sft chat-template hugging-face bf16 llm llms low-rank-adaptation fine-tuning-llm
-
Updated
Feb 18, 2025
Convert PyTorch checkpoints, safetensors files and whole Hugging Face model folders to fp16, bf16, fp8 or ComfyUI fp8_scaled, int8_convrot and int8, turn FP8 models (DeepSeek, Qwen) back into BF16 on a CPU, merge or split sharded safetensors, and extract EMA weights, with one command and without loading the model into RAM.
machine-learning deep-learning pytorch lora quantization ema fp16 model-compression huggingface model-conversion model-checkpoint bf16 llm fp8 stable-diffusion safetensors comfyui tensor-precision deepseek fp8-scaled
-
Updated
Oct 29, 2026 - Python
Lossless compression of BF16 LLM weights, outputs bit-identical to the original. MLX on Apple Silicon and NVIDIA GPUs.
compression metal cuda inference gemma mlx huggingface lossless-compression apple-silicon bf16 llm quantization-alternative
-
Updated
Nov 2, 2026 - Python
SplitZip: Ultra Fast Lossless KV Compression for Disaggregated LLM Serving
gpu triton mooncake lossless-compression kv-cache bf16 llm-serving llm-inference sglang kv-cache-compression disaggregated-inference neurips-2026 prefill-decode-disaggregation
-
Updated
Sep 28, 2026 - Python
Fast BF16 PyTorch training operators for standard and low-rank Attention Residuals
-
Updated
Oct 9, 2026 - Python
Systematic 24-hour benchmark study of Qwen3.6-27B inference on dual NVIDIA RTX PRO 6000 Blackwell SM120 (TP=2). 8 experiments comparing repne/vllm fork vs upstream vLLM across FP8/BF16/NVFP4/Q8_0 quants and MTP/DFlash speculative decoding. Peak: 2,083 tok/s at c=32. Quality: KLD vs BF16 = 0.0018 (noise floor).
-
Updated
Jun 3, 2026 - Python
Converts a floating-point number or hexadecimal representation of a floating-point numbers into various formats and displays them into binary/hexadecimal.
-
Updated
Dec 23, 2021 - C
Neural Weight Compression: lossless LLM weight compression decoded inside the CUDA matvec. BF16 31 % smaller and faster than cuBLAS; fp8 13 % smaller at native fp8 speed. Bit-identical weights.
compression gpu cuda transformers inference pytorch nvidia lossless lossless-compression bf16 llm fp8 llm-inference weight-compression
-
Updated
Sep 19, 2026 - Cuda
Rust based converter for BF16 safetensors to Halogen v2 format
-
Updated
Oct 20, 2026 - Rust
ComfyUI Qwen-Image 2.1 multi-GPU inference: Ulysses sequence parallelism for 1/2/4/8 GPUs, T2I, image editing, INT8/FP8, and distributed prefix cache.
ulysses image-editing nvidia image-generation text-to-image multi-gpu int8 nccl multi-reference multi-gpu-inference bf16 fp8 comfyui qwen sequence-parallelism distributed-inference comfyui-custom-node qwen-image prefix-cache qwen-image-2-1
-
Updated
Sep 29, 2026 - Python
Python implementations for multi-precision quantization in computer vision and sensor fusion workloads, targeting the XR-NPE Mixed-Precision SIMD Neural Processing Engine. The code includes visual inertial odometry (VIO), object classification, and eye gaze extraction code in FP4, FP8, Posit4, Posit8, and BF16 formats.
-
Updated
Aug 17, 2025 - Jupyter Notebook
Auto GGUF Converter for HuggingFace Hub Models with Multiple Quantizations (GGUF Format)
machine-learning cmake ai tensorflow transformers huggingface ai-models huggingface-models bf16 safetensors llama-cpp gguf gguf-models gguf-quantization gguf-editor convert-gguf
-
Updated
Feb 11, 2026 - Python
DeepSeek-OCR-experimental is an advanced, multi-purpose visual document intelligence and object localization sandbox. Powered by the unredacted prithivMLmods/DeepSeek-OCR-Latest-BF16.I64-v2.0 architecture, this suite is designed to deliver highly accurate, structure-aware image text extractions.
torch pytorch accelerate matplotlib addict gradio opencv-python torchvision i64 huggingface-transformers timm bf16 huggingface-spaces flash-attention-2 deepseek-ocr easydict
-
Updated
Jun 27, 2026 - Python
Matrix-free 3D SIMP topology optimization with fused gather-GEMM-scatter CUDA kernels on NVIDIA RTX 4090. Companion code for arXiv:2604.18020.
cuda scientific-computing gpu-computing topology-optimization cupy finite-element-methods matrix-free tensorcore bf16 rtx4090
-
Updated
Aug 26, 2026 - Python
Open, reproducible AMD XDNA1 + Strix Point XDNA2 compute on Linux: pinned builds, CPU-reference checks, persistent C/Python runners, and an LLM roadmap.
linux machine-learning amd hardware-acceleration aie npu edge-ai matmul mlir iree bf16 llm xdna ryzen-ai strix-point neural-processing-unit amdxdna xdna2 xdna1 mlir-aie
-
Updated
Aug 16, 2026 - Shell
Lossless AI model compression - ~34% smaller with bit-identical weights; the autopilot profiles your machine, picks the highest fidelity that runs, and streams models bigger than your RAM.
pytorch autopilot entropy-coding model-compression huggingface lossless-compression kv-cache bf16 llm fp8 safetensors streaming-inference
-
Updated
Jun 13, 2026 - Python
Reproducible benchmark suite and tuned Triton fused-MoE configs for NVIDIA H20 LLM inference. 24 configs, 36 perf data points, geomean 1.09× / peak 1.74× speedup.
benchmark triton moe hopper h20 kernel-tuning mixture-of-experts bf16 llm fp8 vllm llm-inference qwen mixtral deepseek sglang triton-kernels nvidia-h20 fused-moe
-
Updated
Jun 2, 2026 - Python
Add this topic to your repo
To associate your repository with the bf16 topic, visit your repo's landing page and select "manage topics."