🌵 Speculative Speculative DecodingWhat if your draft model could speculate while the target model is still verifying? That's the idea behind Speculative Speculative Decoding (SSD). I've been… | Maxime Labonne
🌵 Speculative Speculative DecodingWhat if your draft model could speculate while the target model is still verifying? That's the idea behind Speculative Speculative Decoding (SSD). I've been… | Maxime Labonne
7 views2 months ago
linkedin.com
DFlash Boosts Speculative Decoding with Lightweight Block Diffusion | Kalyan KS posted on the topic | LinkedIn
DFlash Boosts Speculative Decoding with Lightweight Block Diffusion | Kalyan KS posted on the topic | LinkedIn
2 views4 months ago
linkedin.com
Speculative Decoding — Think Fast⚡, Then Think Right✅
Speculative Decoding — Think Fast⚡, Then Think Right✅
Apr 13, 2025
substack.com
How to Quadruple LLM Decoding Performance with Speculative Decoding (SpD) and Microscaling (MX) Formats on Qualcomm® Cloud AI 100
How to Quadruple LLM Decoding Performance with Speculative Decoding (SpD) and Microscaling (MX) Formats on Qualcomm® Cloud AI 100
Aug 1, 2024
qualcomm.com
Faster LLMs: Accelerate Inference with Speculative Decoding
Faster LLMs: Accelerate Inference with Speculative Decoding
11 months ago
ibm.com
T-pro 2.0: Efficient Russian Reasoning LLM
3:49
T-pro 2.0: Efficient Russian Reasoning LLM
4 months ago
YouTubeAI Research Roundup
Speculative Decoding: Make AI 2-3x Faster for Free | Tech Decoded
6:13
Speculative Decoding: Make AI 2-3x Faster for Free | Tech Decoded
1 month ago
YouTubeToc am
3:08
What is Speculative Decoding ?
5 views5 days ago
YouTubeDeepManim
20:27
Ep 121: Latency and Throughput — Making AI Fast | LLM Mastery Podcast
28 views2 weeks ago
YouTubecarlos Hernandez
7:09
Don't use speculative decoding until you watch this
7 views2 weeks ago
YouTubeDigitalOcean
0:49
Why ChatGPT streams faster than it used to: speculative decoding explained in 48 seconds
1.1K views2 weeks ago
YouTubeAdam Rosler
2:54
What is Speculative Decoding?
1 week ago
YouTubeStandarity
40:19
Speculation is all you need: Intro to Speculative Decoding for High Performance Inference
1 views2 months ago
YouTubeModal
5:04
Speculative Decoding: 2-3x Faster LLMs for Free
1 views1 month ago
YouTubeThe AI Century
1:25
Stop LLM Lag: The Secret to 1.4x Faster AI (ConfLayers) #Shorts
1 week ago
YouTubeCollapsedLatents
14:13
【論文解説】【爆速化】推測デコーディングをスパース計算で検証！驚きの成果公開！
1 views2 months ago
YouTube論文解説チャンネル
0:59
Speculative Decoding explained in Hindi #aiengineering #datascience #llm #mustdo Interview Question
24 views3 months ago
YouTubeLearn AI with RC
1:02:23
EP5: Speculative Decoding with Nadav Timor
116 views7 months ago
YouTubeThe Information Bottleneck
6:29
Inference Optimization: Making AI Faster & Cheaper (Latency, Throughput & GPUs)
56 views1 month ago
YouTubewecite
8:48
The Voynich Manuscript Decoded: A Reproducible Decoding Framework
159 views1 week ago
YouTubeCreation Unified
0:10
Qwen 推理性能最高提升8倍！这个 DFlash 前几天看到了，感觉就是以前的speculative decoding，结果今天看到有人加了 DDTree 技术。实际就是把 speculative decoding 的单链 draft 变成树状draft + Tree Attention 一次验证 + 最长匹配 prefix commit。目的是不让 draft 模型在 branching point 的预测被浪费， DFlash 每个 position 只留一个预测，DDTree 则保留多个，再让 target 模型一次看完挑哪个对。直接让 Qwen3-30B-MoE 在 HumanEvalT 实现了 8.22 倍速度提升！
38.4K views3 weeks ago
x.comnash_su - e/acc
0:41
Using LLMs on your Mac?dflash-mlx delivers a 4x increase in speed using block-diffusion speculative decoding.> Built from scratch on Metal for Apple Silicon.> Draft model proposes multiple tokens at once.> Target verifies the whole block in a single pass.> Accept the longest correct prefix.> Bit-identical output, up to 4x the throughput.> Every piece hand-rolled — hidden-state hooks, mixed-attention KV rollback, warm-path benchmarking. Macs just got a big upgrade for local AI!
6.1K views4 weeks ago
x.comMarkets & Mayhem
15:39
【AI论文解读】让 speculative decoding 更快更准！任务感知的 Draft 模型优化方案 | TAPS
1 month ago
bilibili熊二等兵
DFVG: A Heterogeneous Architecture for Speculative Decoding with Draft-on-FPGA and Verify-on-GPU | Proceedings of the 31st ACM International Conference on Architectural Support for Programming Languages and Operating Systems, Volume 2
2 months ago
acm.org
12:10
推测解码在Qwen3.6的llama.cpp上实现
758 views1 week ago
bilibili张曦偌
0:12
Algo Brief on Instagram: "Interesting fact: To make these massive models faster for real-time coding, engineers are now using Speculative Decoding. This involves a tiny, lightning-fast "draft" model predicting several tokens at once, while the massive "teacher" model (like Claude 4 or GPT-5) only steps in to verify or correct the draft, reducing latency by up to 40% without sacrificing the intelligence of the output."
3.6K views3 months ago
Instagramalgobrief
Multi-candidate Speculative Decoding | Natural Language Processing and Chinese Computing
2 months ago
acm.org
6:20
Encoder Decoder Network - Computerphile
157.1K viewsJun 13, 2018
YouTubeComputerphile
6:47
Transformer models: Encoder-Decoders
107K viewsJun 14, 2021
YouTubeHugging Face
12:27
Stuart Hall's Encoding/Decoding Model but it's easier to understand
66.6K viewsOct 22, 2020
YouTubeBarley Make Video
See more