🔥🔥 The extremely fast inference engine for decision model
-
Updated
Nov 12, 2026 - Python
🔥🔥 The extremely fast inference engine for decision model
Deterministic, no_std engine for verifiable, cross-vendor bit-identical LLM inference.
Blackwell 6000 + qwen 3.8 flash next: Swift 1.5 Qwen3.8 Flash-Next NVFP4 on one RTX PRO 6000 with vLLM, 4 users x 227k context, ~100 tok/s each, exact kernels + determinism fix
Deterministic LLM framework combining strict context governance (veritas/sim frameworks) with a high-fidelity somatic simulation engine tracking real-time biometric and status vectors.
Plugin-free, deterministic TensorRT-10 warp for LivePortrait & FasterLivePortrait. Fixes face deformation/melting and black-frame under GPU contention (the non-deterministic grid_sample_3d plugin), and unblocks TensorRT 10. Drop-in, ~25% faster.
Higgins Decomposition (Hs) — deterministic compositional inference instrument. 18 domains, 36 systems. CNT engine + CNQ quaternion verification (IEEE-floor on Backblaze, Planck CMB, SM Neutrino). Apache-2.0 (code) + CC BY 4.0 (docs).
Batch-invariant inference nodes for guaranteed reproducibility in ComfyUI. ThinkingMachines + ECHO 2.0 + Nemotron patterns.
Safety-first legal NLP system with chunk-based long-document processing, deterministic inference, clause extraction, and rule-based risk engine — built for traceability and deployment constraints.
Study bitwise reproducibility with experimental integer Llama inference, weight-free CPU checks and scoped device reports.
speedrunpapi, the universal engine: exact results with independent proofs. Output determinism for Circom and R1CS, bounded exact mathematics, bit-exact ONNX plans, audited arithmetic cores, circuit blocks with proofs, exact models
Bitwise batch-invariant matmul and attention kernels for MLX on Apple Silicon: identical logits at any batch size, with the harness that proves it.
Exact speculative decoding + lossless 12-bit weights for Qwen3 on a consumer AMD GPU (RX 9070 XT)
Integer-only LLM inference: Qwen2.5-0.5B within 0.75% of float32 perplexity, bit-identical on an NVIDIA GPU and a CPU. Its lab is merged into Harvard's ML Systems textbook.
To associate your repository with the deterministic-inference topic, visit your repo's landing page and select "manage topics."