Pushing the bf16 multiplication clock frequency to the max on the nominal corner on IHP 130nm 5L node.
-
Updated
Mar 29, 2026 - Python
#
bfloat16
Here are 18 public repositories matching this topic...
A LLaMA2-7b chatbot with memory running on CPU, and optimized using smooth quantization, 4-bit quantization or Intel® Extension For PyTorch with bfloat16.
meta cpu optimization chatbot intel llama numa int8 ipex 4-bit-cpu huggingface streamlit bfloat16 neural-compression chatgpt langchain llama2 meta-ai smooth-quantization chatbot-memory
-
Updated
Feb 27, 2024 - Python
A JAX implementation of stochastic addition.
-
Updated
Aug 15, 2022 - Python
A Pytorch implementation of stochastic addition.
-
Updated
Aug 11, 2022 - Python
Drop-in exact bf16 flash-attention for CUDA with a deterministic backward, tuned for Blackwell (sm_120 / RTX 5090).
deep-learning gpu cuda inference pytorch transformer triton attention memory-efficient attention-mechanism deterministic memory-optimization bfloat16 tensor-cores blackwell kv-cache llm-inference flash-attention rtx-5090 sm120
-
Updated
Jul 8, 2026 - Python
In bfloat16 a trainer gives the same token a different log probability depending on its batch shape; measured on 8 models, with the controls that survived
reinforcement-learning reproducibility numerics importance-sampling ppo bfloat16 trl llm rlhf vllm llm-training grpo verl batch-invariance
-
Updated
Oct 1, 2026 - Python
AWS deployment stack for Gemma 3 on SageMaker with HuggingFace TGI, OpenAI-compatible API (Lambda + API Gateway), and OpenWebUI chat interface
aws lambda cloudformation sagemaker tgi huggingface bfloat16 openai-api llm-inference openwebui gemma3
-
Updated
Mar 25, 2026 - Python
English Whisper Turbo transcription on Google Colab TPU in 3 cells. JAX/XLA, reusable model worker, TXT/SRT/VTT, resumable audio, and clear error logs. No API key.
python subtitles srt automatic-speech-recognition speech-to-text whisper tpu jax google-colab colab-notebook xla audio-transcription bfloat16 openai-whisper whisper-turbo whisper-large-v3-turbo tpu-v5e english-transcription tpu-v6e
-
Updated
Sep 20, 2026 - Python
FP64-accurate linear algebra on BF16/FP32 hardware (TPU) via the Ozaki scheme, in JAX.
-
Updated
Jul 9, 2026 - Python
Empirically calibrated false-positive bound for one-sided checksum ABFT on bfloat16 GEMM, where the classical τ = Ku threshold is undefined. Catches 76% of exponent-bit flips against a pre-registered 90% bar.
fault-tolerance reproducible-research gpu reliability cuda pytorch gemm error-detection abft pre-registration bfloat16 numerical-computing extreme-value-theory silent-data-corruption
-
Updated
Sep 29, 2026 - Python
Generate Spike extensions, assembly tests, SVA assertions & docs for custom RISC-V AI vector instructions from YAML specs. Bit-accurate FP8/BF16/INT4 numerics.
verification code-generation quantization risc-v hardware-verification bfloat16 fp8 spike-simulator ai-accelerator vector-instructions
-
Updated
Mar 31, 2026 - Python
BF16 x BF16 -> FP32 multiplier and two-stage FP32 adder in synthesizable SystemVerilog, bit-exact against two independent reference models
-
Updated
Oct 4, 2026 - Python
Float accumulation order alone flips RL reward verdicts and sampler/trainer probabilities — reproduce it on real GPT-2, then remove it with an order-independent reduction. numpy-only, runs in seconds.
reinforcement-learning numpy floating-point reproducibility posit quire numerical-stability post-training determinism computer-arithmetic gpt-2 bfloat16 numerical-computing nondeterminism rlhf reward-model llm-training llm-evaluation grpo bit-exact
-
Updated
Aug 16, 2026 - Python
Modly extension for research-only Cube3D INT4 text-to-mesh generation with UI-managed weights and GLB output.
cuda pytorch mesh glb bfloat16 cube3d int4 research-only text-to-3d local-ai text-to-mesh modly modly-extension geometry-only
-
Updated
Jul 17, 2026 - Python
Experimental research on bit flips in language-model weights: from faults nobody chose to faults an attacker did.
reproducible-research quantization fault-injection ai-safety rowhammer bit-flip-attack bfloat16 llm-security gguf hardware-fault-injection
-
Updated
Sep 5, 2026 - Python
Fast FP32-to-BF16 stochastic rounding for PyTorch and Triton, with BF16 AdamW and momentum SGD optimizer states.
-
Updated
Oct 3, 2026 - Python
A research toolbox for running and studying LLMs on constrained hardware, with memory, placement, inference, and benchmarking methods.
reproducible-research gpu cuda pytorch model-compression vram lossless-compression bfloat16 local-llm llm-inference speculative-decoding llm-benchmarking openai-compatible-api weight-streaming model-offloading
-
Updated
Oct 4, 2026 - Python
Add this topic to your repo
To associate your repository with the bfloat16 topic, visit your repo's landing page and select "manage topics."