End-to-end recipes for optimizing diffusion models with torchao and diffusers (inference and FP8 training).
-
Updated
Jan 8, 2026 - Python
#
torch-compile
Here are 35 public repositories matching this topic...
Profine automatically profiles and optimizes PyTorch training jobs on real GPUs, delivering measurable speedups and lower GPU costs before teams waste days tuning configs by hand.
python cli benchmark machine-learning deep-learning gpu modal pytorch command-line-tool profiling performance-optimization ai-agents model-training mixed-precision mlops mingpt llm-agents gpu-profiling torch-compile automated-optimization
-
Updated
May 20, 2026 - Python
PyTorch extension for Rebellions NPU
python machine-learning deep-learning neural-network inference pytorch npu hardware-backend ai-accelerator torch-compile rbln rebellions-ai rebellions eager-mode
-
Updated
Sep 21, 2026 - Python
How does torch.compile() achieve massive speedups despite highly optimized torch functions? See how operator fusion works!
-
Updated
Aug 3, 2026 - Jupyter Notebook
Wide-model collective ensemble system with fractal, geometric, and heavy compilation optimizations.
ai deep-learning experimental collective fractal torch pytorch ensemble geometric-shapes experimental-paradigm geometric-algorithms geometric deep-learning-framework geometric-deep-learning ai-experiments torchvision torch-geometric agentic torch-compile geofractal
-
Updated
Apr 2, 2026 - Jupyter Notebook
Optimized CSM-1B TTS pipeline for RTX 5090 (Blackwell sm_120). CUDA graph replay via patched HF Transformers. ~0.46x RTF. Topics (tags): csm text-to-speech rtx-5090 blackwell cuda-graphs torch-compile sesame streaming pytorch
text-to-speech streaming pytorch tts sesame csm huggingface blackwell torch-compile rtx-5090 sm-120 cuda-graphs
-
Updated
Apr 5, 2026 - Python
Leveraging torch.compile to accelerate cross-encoder inference
-
Updated
Mar 3, 2025 - Python
Compression-aware Qwen3-VL inference engine with torch.compile, CUDA Graph, scaled-FP8 KV, visual compaction, and multimodal prefix caching.
-
Updated
Sep 8, 2026 - Python
Benchmarking, optimising and serving the TabFM tabular foundation model online: exact KV cache, torch.compile, FastAPI, Triton, KServe on kind and GKE
kubernetes machine-learning pytorch gke model-serving inference-optimization kserve nvidia-triton torch-compile claude-code nvidia-l4 tabfm tabular-foundation-model
-
Updated
Sep 16, 2026 - Python
Distilling Florence-2 into a compact ~60M-param VQA student (feature-matching + attention transfer) — ~70% teacher-accuracy retention at 138x inference throughput with bf16 + torch.compile on H100.
computer-vision deep-learning pytorch vqa knowledge-distillation model-compression visual-question-answering inference-optimization vision-language-model florence-2 torch-compile
-
Updated
Jul 22, 2026 - Python
Evidence-gated torch.compile optimization lab with offline-verifiable bundles
-
Updated
Sep 6, 2026 - Python
Measurement-driven PyTorch training autotuner — tries precision × torch.compile × memory-format configs on your model, keeps the fastest, never returns one slower than your baseline. Honest, committed benchmarks.
performance deep-learning optimization pytorch gpu-acceleration auto-ml mixed-precision torch-compile training-speedup
-
Updated
Aug 25, 2026 - Python
Reproducible benchmark harness for neural TTS inference. Isolated Docker A/B arms, compiler vs eager measurement, request-path profiling, cross-process determinism controls, VRAM soak testing, and an ASR-scored regression suite that catches reference-audio leakage. Measured on OpenAudio S1-mini, RTX 5080.
-
Updated
Aug 18, 2026 - Python
Optimized PyTorch inference for NISQA speech quality and TTS naturalness models
deep-learning cuda inference pytorch speech-processing speech-quality speech-assessment nisqa torch-compile
-
Updated
Aug 10, 2026 - Python
One-command trust auditor for torch.compile: eager vs aot_eager vs inductor, five correctness checks, minimal repro, paste-ready report.
-
Updated
Sep 3, 2026 - Python
GPU systems benchmark comparing PyTorch eager execution and torch.compile across common deep learning operations. Measures execution time, throughput, TFLOPS, memory bandwidth, and kernel-level speedups through a reproducible benchmarking pipeline and publication-ready visualizations.
-
Updated
Jun 29, 2026 - Jupyter Notebook
Guards a real torch.compile(inductor) CPU precision bug: std/var_mean/std_mean accumulate in float32 instead of eager's double precision (pytorch/pytorch#197089)
-
Updated
Sep 19, 2026 - Python
Three measured notebooks on GPU inference optimization: roofline analysis, KV-cache decode optimization (4.21x), and CUDA-graph launch-overhead elimination (5.38x). Pure PyTorch.
gpu pytorch performance-optimization inference-optimization kv-cache llm-inference torch-compile roofline-analysis cuda-graphs
-
Updated
Jul 7, 2026 - Jupyter Notebook
PyTorch 2.9 + HIP/ROCm learning notes: dispatcher, accelerator backend, profiler, TorchInductor, runtime, GEMM, and C++ extensions.
-
Updated
Jul 20, 2026 - Python
Add this topic to your repo
To associate your repository with the torch-compile topic, visit your repo's landing page and select "manage topics."