Top Tweets for #Speculativedecoding
Need a fuckin research partner rn….!
#SpeculativeDecoding #ArtificialIntelligence
#MachineLearning
#DraftModels
#Research
Making the draft model and the target model align is the toughest optimisation I have made till now.
But loving it🌟.
#DraftModels #AI #LLM #Inference #ML #Optmisation #Speculativedecoding
When one user is chatting with an 8B model on a top GPU like NiVida H100, the GPU'S math unit is sit idle about 99% of the time. Each new work means reading all 16GB of weights from memory, which takes about 5ms.
(Continued)
#batching #speculativedecoding #nvidia #h100 #quantization
🎉 Four of our papers were accepted to the NeurIPS 2026 workshops last week, including one oral presentation.
Huge congrats and thanks to all my co-authors!
#NeurIPS2026 #OnDeviceAI #LLM #SpeculativeDecoding #DiffusionModels #TestTimeAdaptation #ContinualLearning
Can AI draft its own next moves? Gemma 4 makes speculative decoding a real local-AI choice.
#Gemma4 #LocalAI #SpeculativeDecoding https://t.co/FMCJSRDZ7C
3/
Huge thanks again to my brilliant co-authors @miles_wil, @ruiruiliii, @kouris_a, and @SIVenieris at Samsung AI Centre-Cambridge.
📄 https://t.co/R6rufbiBik
#EMNLP2026 #LLM #SpeculativeDecoding #EdgeAI #EfficientInference
Ever wished your AI could think ahead and speed up responses? Meet Laguna-S-2.1-DFlash-FP8, a speculative decoding marvel. It uses a draft model to guess tokens, making generation faster without losing quality. This is a game changer for real-time apps! #AI #SpeculativeDecoding
Grateful to my mentors @nrsuri and @soumyabrata_pal, and @AdobeResearch for supporting this work.
Paper link coming soon. Any queries? Drop a comment 👇
#ACL2026 #LLM #ReasoningAI #EfficientAI #SpeculativeDecoding #InferenceOptimization #AdobeResearch
🚀 Excited to share that our paper "Speculative Decoding with a Speculative Vocabulary" has been accepted at ACL 2026 Findings!
Congrats Miles, @ruiruiliii, @kouris_a, and @SIVenieris!
Paper: https://t.co/22wJVEOjlS
#ACL2026 #SpeculativeDecoding #EfficientInference #LLM
9/N. What speculative decoding results have you seen in your own deployments? We would love to hear what works (and what doesn't).
Paper: https://t.co/imokQeXDHj
Blog: https://t.co/IRNYLB97dF
Code: https://t.co/jMBqDt87aw
#LLM #MLSystems #SpeculativeDecoding
🚀 Is slow #LLM inference wasting your compute?
🤔 Ever wondered how compute resources for inference should be better utilized?
We've got a new game plan! 🎮
Presenting TETRIS, our work on speeding up batch #SpeculativeDecoding under limited compute resources! We dynamically select the best draft tokens across multiple requests, perfect for inference engines to slash waste and maximize throughput. 📈
💡 Discover how we're making #LLMs faster and more efficient!
Paper: https://t.co/LoEEOPh0gQ
Code: https://t.co/0vnCLZLhC2
Joint work with @WuZhaoxuan @BobbyZhouZijian @arun_v3rma Alok Prakash, Daniela Rus.
Check out our @aclmeeting #ACL2025 #ACL2025NLP poster on 28th July in Poster Session 1 Number 81📍Hall X5!
Assisted generation and its unprecedented charm is not only restricted to 10x LLM throughputs but also 10x possible quality improvements in output.
#llm #assistedgeneration #speculativedecoding #genai #ai
👏👏👏 @WuZhaoxuan @ZijianZhou524 @arun_v3rma @rus_daniel84837 @lululu0082 @JingtanW zitong @Dai_Zh et al.
https://t.co/Dkw3c6ky3C's first accepted papers @aclmeeting #ACL2025 #ACL2025NLP: #LLM #LLMs #AICopyright #DataCentricAI #Watermarking #SpeculativeDecoding #DataAttribution
How would the experience @aclmeeting #ACL2025 be different from that @emnlpmeeting #EMNLP2024?
Excited to share our Speculative Decoding support for the TensorRT-LLM Engine Builder! 🚀 Streamline workflows with pre-optimized configs and customize for your unique use cases. #AI #TensorRT #SpeculativeDecoding #genAI
🚀 We’re excited to introduce our Speculative Decoding integration for our TensorRT-LLM Engine Builder!
Our new integration allows engineers to:
• Leverage SpecDec as part of our Engine Builder flow
• Hit the ground running with pre-optimized configs
• Lift the hood and tune for any custom use case
👉 Check out the launch blog to learn more: https://t.co/LkhWhIKEKK
Huge shoutout to our model performance team for this feature, including @aqaderb, Justin Yi, Bryce Dubayah, and @defpan!
A novel strategy enhancing the efficiency of LLMs through an innovative recurrent drafter design. #LargeLanguageModels #SpeculativeDecoding #Efficiency
Most Popular Users
Elon Musk 
@elonmusk
240.8M followers
Barack Obama
@barackobama
118.7M followers
Cristiano Ronaldo
@cristiano
114.5M followers
Donald J. Trump
@realdonaldtrump
111.9M followers
Narendra Modi
@narendramodi
107M followers
Rihanna
@rihanna
98.7M followers
NASA
@nasa
92.2M followers
Justin Bieber
@justinbieber
91.8M followers
KATY PERRY
@katyperry
90M followers
Taylor Swift
@taylorswift13
84M followers
Lady Gaga
@ladygaga
75.5M followers
Virat Kohli
@imvkohli
73.4M followers
Kim Kardashian
@kimkardashian
70.8M followers
YouTube
@youtube
68.7M followers
Neymar Jr
@neymarjr
66.4M followers
Bill Gates
@billgates
65.2M followers
Selena Gomez
@selenagomez
63.1M followers
The Ellen Show
@theellenshow
62.2M followers
CNN
@cnn
61.7M followers
X
@x
60.6M followers