#Speculative-decoding

Showing 12 of 12 repositories tagged #speculative-decoding, ranked by stars

intel
intel
intel-extension-for-transformers

⚡ Build your chatbot within minutes on your favorite device; offer SOTA compression techniques for LLMs; run LLMs efficiently on Intel Platforms⚡

Score
43
★ 2.2k ⑂ 216
Python
dphnAI
dphnAI
sonar

Large-scale LLM inference engine

Score
0
★ 1.8k ⑂ 202 +1/day
C++
Tencent
Tencent
AngelSlim

Model compression toolkit engineered for enhanced usability, comprehensiveness, and efficiency.

Score
100
★ 1.4k ⑂ 156 +29/day
Python
Avarok-Cybersecurity
Avarok-Cybersecurity
atlas

Pure Rust Inference Engine

Score
100
★ 585 ⑂ 83 +25/day
Rust
zengxiao-he
zengxiao-he
tessera

From teacher to tiles — a from-scratch LLM distillation & serving engine: custom Triton/CUDA kernels, FSDP distillation, paged-KV continuous batching, speculative decoding, a Rust gateway, a JAX oracle, and interpretability tooling.

Score
0
★ 512 ⑂ 5 +24/day
Python
chrisliu298
chrisliu298
awesome-on-policy-distillation

A curated collection of papers, technical reports, frameworks, and tools for on-policy distillation (OPD) of large language models

Score
86
★ 489 ⑂ 14 +29/day
AEON-7
AEON-7
Qwen3.6-27B-AEON-Ultimate-Uncensored-DFlash

Fully uncensored, capability-enhanced abliteration of Qwen3.6-27B. NVFP4 + z-lab DFlash speculative decoding (n=12) on the unified ghcr.io/aeon-7/aeon-vllm-ultimate:latest container, tuned for long-context draft acceptance on DGX Spark. 6 HF variants (BF16/NVFP4/MTP/MTP-XS), docker-compose, and QuickStart.

Score
71
★ 403 ⑂ 41 +3/day
Python
Infini-AI-Lab
Infini-AI-Lab
Sequoia

scalable and robust tree-based speculative decoding algorithm

Score
14
★ 376 ⑂ 37
Python
facebookresearch
facebookresearch
LayerSkip

Code for "LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding", ACL 2024

Score
57
★ 373 ⑂ 43
Python
raketenkater
raketenkater
llm-server

Auto-tuned launcher for GGUF models on llama.cpp / ik_llama.cpp — OpenAI-compatible server with multi-GPU tensor-split, MoE expert placement, measured flag tuning (AI Tune), hardware-matched HuggingFace downloads, and crash recovery. An Ollama alternative for multi-GPU rigs.

Score
0
★ 241 ⑂ 12 +1/day
Go
Sandermage
Sandermage
sndr_core_engine

SNDR Core Engine (Genesis) — vLLM runtime patch-overlay for Qwen3.6 + Gemma4 on consumer NVIDIA (Ampere sm_86, 2× A5000/3090). Qwen3.6-35B-A3B FP8 ~240 tok/s, 27B-int4 hybrid GDN+Mamba, Gemma4 26B/31B AWQ, 256K ctx. 321 patches: TurboQuant k8v4 KV, MTP/DFlash spec-decode, FULL cudagraph, hybrid GDN. vLLM pin dev424 + Control Center GUI.

Score
29
★ 119 ⑂ 6
Python
kssteven418
kssteven418
BigLittleDecoder

[NeurIPS'23] Speculative Decoding with Big Little Decoder

Score
0
★ 99 ⑂ 11
Python
Related Topics
#llm#cuda#quantization#llm-inference#qwen#rag#inference-engine#rocm#dflash#dgx-spark#nvfp4#openai-api

© 2026 GitRepoTrend · GitHub repositories by topic · Updated weekly