Alex Sherstinsky (@AlexSherstinsky) profile picture

Scientist, Engineer, Manager, Musician, Athlete, Thespian, Father, Husband, Son, Friend

Silicon Valley
Joined August 2008
Photo by Alex Sherstinsky (@AlexSherstinsky)Photo by Alex Sherstinsky (@AlexSherstinsky)Photo by Alex Sherstinsky (@AlexSherstinsky)Photo by Alex Sherstinsky (@AlexSherstinsky)Photo by Alex Sherstinsky (@AlexSherstinsky)Photo by Alex Sherstinsky (@AlexSherstinsky)Photo by Alex Sherstinsky (@AlexSherstinsky)Photo by Alex Sherstinsky (@AlexSherstinsky)Photo by Alex Sherstinsky (@AlexSherstinsky)Photo by Alex Sherstinsky (@AlexSherstinsky)
Alex Sherstinsky retweeted
Thanks to @GoogleDeepMind and @GraphonAI - we had a lot of fun hosting Gemini 3 build day @ AGI House last Saturday Cc. @DynamicWebPaige @vadiamit @bonniesjli @JessicaChenX
6
3
23
2,711
Alex Sherstinsky retweeted
Predibase + @Cometml: everything you need to build high performance #GenAI applications 💫 We're excited to be a launch partner for Comet's #Opik! Together we make it easy to customize best-of-breed small models (#SLMs), seamlessly track and #evaluate your experiments, and serve at blazing fast speeds in production! 📊 Track your Predibase fine-tuning jobs and see in-depth analytics in Comet: docs.predibase.com/user-guid… 📈 Log and evaluate Predibase LLM traces with Comet Opik: comet.com/docs/opik/tracing/…
1
2
5
388
Alex Sherstinsky retweeted
Want to supercharge your #LLM inference? 🔥 Traditional fine-tuning methods make it possible to achieve #GPT4 level accuracy with small open-source models (#SLMs). But, what about throughput? Introducing #TurboLoRA, a new parameter-efficient fine-tuning method that increases throughput by 2-3x while simultaneously achieving high model accuracy. Join our upcoming webinar to learn: ➡️ Key challenges GenAI teams face with #opensource models ➡️ How to increase #throughput and accuracy with fine-tuning ➡️What's under the hood w/ Turbo LoRA incl. spec decoding ➡️Performance #benchmarks pre / post fine-tuning By the time you leave, you'll know how to unlock lower latency, reduced inference costs, and higher accuracy — all in a single adapter. Sign-up here: go.predibase.com/3x-faster-i…
2
4
349
Alex Sherstinsky retweeted
Across 30 tasks, @UpstageAI's Solar Pro Preview fine-tuned outperforms every top fine-tuned open-source #LLM and #GPT4 and #GPT4o-mini base models! See more at our new Fine-Tuning Leaderboard: pbase.ai/4gnsoaI Try #SolarProPreview for free: pbase.ai/3Zk1N8j
2
1
7
578
Alex Sherstinsky retweeted
Replying to @upstageai
@upstageai fine-tuned Solar-Proofread w/ Predibase, achieving 79% accuracy vs. fine-tuned GPT-4o mini’s 71% for a major international media company. Discover the power of fine-tuned SLMs: pbase.ai/4e1DQaE #AI #LLM #Upstage #Predibase #SmallLanguageModels
1
3
95
Alex Sherstinsky retweeted
Is a lack of data holding you back from training your next model? Learn how you can generate synthetic data from as few as 10 seed examples for fine-tuning an LLM that outperforms GPT-4. Watch this 90-second or try for yourself with our free trial: pbase.ai/47iRWlq
1
7
384
Alex Sherstinsky retweeted
What should data + AI execs think about as they build their #GenAI stack? Here you go (courtesy of @schwentker): 🚀 Evaluate Your AI #Infra - hint: small fine-tuned models can improve efficiency and reduce costs. 🤝 Invest in Open-Source - #opensource provides greater flexibility, access to new innovations and you can own your models! 🛠 Prioritize #FineTuning Capabilities - get high performance without the overhead of massive models. 🔀 Adopt #Dynamic Model Adaptation: look for platforms that can serve many models on the same infra efficiently (see: loraexchange.ai) 🤸‍♂️ Stay Agile: the landscape is evolving rapidly, so agility is key if you want to out innovate the competition. Check out Robert's full blog for all his great insights incl. a summary of the key points from Piero Molino (@w4nderlus7) on how to build big with #small models just like Apple: linkedin.com/pulse/genai-arc…
3
9
403
Alex Sherstinsky retweeted
What if you could have your own highly-optimized #LLMs running in your #private cloud without any hassle? 🙋‍♂️ 🙋‍♀️ Well now you can. No more choosing between #performance and #security — have your LLM cake and eat it too! 🍰 😎 Want to learn how? 💡 Save a spot for our webinar to learn how to easily deploy LLMs in your cloud < 30 minutes with Predibase #VPC. We'll even show you how to get those models to outperform #GPT4! 💰 go.predibase.com/your-models…
1
5
7
369
Alex Sherstinsky retweeted
Replying to @upstageai
@upstageai's Solar Mini–already the best SLM for fine-tuning–is about to get faster, driving huge savings in terms of GPU-hrs A general purpose speculator makes the base model 1.35x faster, and fine-tuning with Turbo LoRA yields a 2.2x bump! Try it today: pbase.ai/3Ara7sx
4
5
243
Alex Sherstinsky retweeted
Want to build big with small models like #Apple 💡🍎 Then don't miss our #LLMOps Micro-Summit in SF (and streaming online) on Aug. 22nd. Leading AI researchers, engineers and founders will share what it takes to build a modern #GenAI stack that is fast, #efficient and built on #SLMs! We're covering everything from building blazing fast #inference to best practices in LLM #evaluation and #synthdata generation. Save your spot: lu.ma/LLMOpsMicroSummit?utm_…
3
3
209
Alex Sherstinsky retweeted
Guys fine-tuned Llama 3.1 8B is completely cracked. Just ran it through our fine-tuning test suite and blows GPT-4o mini out of the water on every task. There has never been an open model this small, this good.
60
204
50
2,002
348,058
Alex Sherstinsky retweeted
Fine tuning is going to become wildly popular again. Better grab the lambda labs compute now 😅
Yep I’d like to do many Llama 3.1 finetunes, coming up.
2
7
1
71
15,697
Alex Sherstinsky retweeted
Meta's new hotness #Llama3.1 is out and ready for #finetuning and #serving on Predibase 🔥 Try it for free! Here's $25 in credits to #customize Llama-3.1 or any of your favorite #opensource models for your use case: predibase.com/free-trial 🎉
3
1
12
624
Alex Sherstinsky retweeted
#Gemma2 and #Qwen2 now on Predibase! 🚀 Fine-tune and serve all of your favorite #SLMs with a fast, easy-to-use and cost efficient platform! Get started with $25 in free #credits: predibase.com/free-trial.
3
10
285
Alex Sherstinsky retweeted
Want to learn how to cut your #OpenAI spend by 50% or more with #SLMs, open-source and the latest techniques in LLM #finetuning? 😎 Great! Save your spot: pbase.ai/4eZFw54
3
3
275
Alex Sherstinsky retweeted
With Snowflake and @predibase you can now train small task-specific LLMs that outperform commercial models at a fraction of the cost—all within your Data Cloud. By combining Predibase’s world-class LLM fine-tuning and inference capabilities with the Snowflake Data Cloud, teams can train their own LLMs directly on their data and deploy using Snowflake’s GPU architecture. Learn more in this webinar and demo: youtube.com/watch?v=szmOLwZd…
9
1
19
3,447
Alex Sherstinsky retweeted
Faster #training = faster iteration = less time spent achieving high performance on your LLM #finetuning jobs 🏁 🏎️ Check out our deep dive blog to learn how we accelerated fine-tuning by 15x in less than 15 days: pbase.ai/3VK1zEa
4
6
286
Alex Sherstinsky retweeted
6/6 🏆 Predibase was featured in @CBinsights #EnterpriseAI Roadmap report pbase.ai/3WatgaQ
1
55
Alex Sherstinsky retweeted
5/6 💡 #LoRA adapters are the future of fine-tuning for efficiency and performance pbase.ai/3Vf4QuP
1
1
45
Alex Sherstinsky retweeted
4/6 📊 Fine-Tuning Index shows #OSS #LLMs outperforming GPT-4 on 85% of tested tasks pbase.ai/4cqxgtp
1
1
55