I can confirm from my M5 Max 128GB and my own tests that Qwen3.8‑Flash‑Next really does hit the output token rates shown in the graph with MTP On. Thanks @ddalcu for the update and for sharing the quantised weights.
MLX-Serve 26.9.2 is out. Fixes, Polish, Speed.
If you have 96-128GB Ram, 👑 Qwen Flash Next is the king of local models for now. Without KV8 Cache + M5Max, you might see it go above 150 Tok/s.
github.com/ddalcu/mlx-serve/…
Siddharth Jayaraman retweeted
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
How far can a ternary 27B model stretch 24 GB of VRAM? I tested @PrismML's Ternary Bonsai 27B on an RTX 3090... #PrismML #Bonsai27B #TernaryBonsai #Qwen36 #LocalLLM #LongContext #RTX3090 #LLMInference
medium.com/open-weights/how-…
@PrismML's Ternary Bonsai 27B benchmark results on an RTX 3090, showing VRAM allocation, prompt processing (pp512) and token generation (tg128) across context depths using F16 and Q4_0 KV caches.
Siddharth Jayaraman retweeted
New Anthropic research: Natural Language Autoencoders.
Models like Claude talk in words but think in numbers. The numbers—called activations—encode Claude’s thoughts, but not in a language we can read.
Here, we train Claude to translate its activations into human-readable text.
Siddharth Jayaraman retweeted
BovReg Final Conference has finished!🚀
We wish to thank all partners and participants who gave precious contribution to this interesting project and event!🙏
Siddharth Jayaraman retweeted
📣The 3rd BovReg project Annual Meeting has finished!
🙏🏻Thanks to all partners who participated, we had very fruitful discussions and presentations!
Looking forward to seeing you soon!💪😉
Siddharth Jayaraman retweeted
🕝The afternoon continues with dedicated sessions about work packages advancements and Democs card game!🃏
Siddharth Jayaraman retweeted
Andrea Talenti, giving a well timed intro to graph genomics - pangenomics #WCGALP2022
See the paper here:
nature.com/articles/s41467-0…
Siddharth Jayaraman retweeted
Submitted in 2020, accepted in 2022, so it's been a ride, but Andrea's paper on building a more representative cattle graph genome is finally out! nature.com/articles/s41467-0…
Siddharth Jayaraman retweeted
📣 Flash talks session in 20 minutes:
@Sidinformatics | @roslininstitute
Bala Kiran Manthri | @SLU_Official
@DrMeenu_bhati | @ETH_en
Praveen Krishna Chitneedi | Leibniz Institute for Farm Animal Biology
Stay tuned: youtube.com/channel/UC0qhdZc…
#nf_bovreg @CRGenomica
Siddharth Jayaraman retweeted
Application of long read sequencing to determine expressed antigen diversity in Trypanosoma brucei infections biorxiv.org/cgi/content/shor… #biorxiv_genomic
Siddharth Jayaraman retweeted
What a bunch of lovely people to work with. Team bonding over curry and beer is always good...
I'm going to "Symposium on African Livestock Trypanosomes". See you there? eventbrite.com/e/symposium-o… via @EventbriteUK
BBC News - Alarm as 'super malaria' spreads in South East Asia bbc.co.uk/news/health-413511…
Edinburgh Genomics Illumina Whole Genome Sequencing Symposium #IlluminaUGM eventregistration.illumina.c….
Siddharth Jayaraman retweeted
Going to be very exciting project - fun few years ahead @luisamfigueired @KeithRMatthews @rmc9z @MonicaMugnier et al. @wellcometrust xitter.cf/keithrmatthews/status/…























