The TechGuy 🐦

Props to every chain — $BTC, $ETH, $SOL, $SUI, $BNB — each moved crypto forward.🙌 But the next leap is @MultiversX | $EGLD ⚡ Ultra-scalable, sustainable, powered by @xPortalApp the best wallet in Web3 — multi-chain, on-chain 2FA. Soon stake with ~16% APR. We’re still early.🚀

@HuggingModels Yo someone make it 12B to fit on 12gb vram

Meet Kimi-K3-0.40B, a compact yet powerful feature extraction model that's taking the ML community by storm. With 17K downloads, it's built on the Kimi-K3 base and fine-tuned for precision. Perfect for turning raw text into rich embeddings. Let's dive in! #AI #NLP

@MSBIntel 👀👏🏻

BREAKING: Sui Foundation introduces a buyback program funded by network revenue from gas fees and stablecoin yield. Purchased $SUI reinvested into the ecosystem.

@Techguy0101010 yes sir! wait for the 27B release, it will fit 12gb :)

Today we are introducing Escha-W2 quantization. A 2-bit Qwen3.6-35B-A3B model built for fast, local inference. The complete model is 12.3GB on disk—small, enough to run on a single consumer GPU — while averaging ~100% of FP8 performance across 12 benchmarks, including: MMLU-Pro: 80.9 MATH-500: 93.8 GPQA-Diamond: 77.8 LiveCodeBench v6: 62.6 BFCL tool use: 88.9 RULER 8K–128K: 89.9 Commonsense-6: 76.1 On a single RTX 4090, the model runs: 225 tok/s single-stream generation on 12.3GB on-disk model size Compressing a 35B MoE model this far without collapsing its capabilities required more than a standard quantization pass. We built an end-to-end compression system combining state-of-the-art low-bit quantization with model-aware fine tuning and recovery to preserve capabilities most vulnerable to low-bit error. Quantizing the Qwen 3.6 35B model - from the base model to the final deployable checkpoint — took approximately 10 hours to complete. Escha-W2 runs through a custom Qwen3-MoE runtime, which includes the weight loader, low-bit decoding kernels and serving integration required to execute the Escha format efficiently. The runtime currently supports SGLang and ZML deployment. vLLM and llama.cpp will be supported in the next release. No retraining from scratch. No specialized accelerator. One consumer GPU. Model download: huggingface.co/EschaLabs/Qwen… Runtime download: huggingface.co/EschaLabs/esch… Apache-2.0 model and runtime.

@Eschalabs ❤️

@SuiNetwork 👀

Sui earns yield on stablecoin float, and that revenue funds $SUI buybacks. Purchased SUI is reinvested into the ecosystem, growing float and driving more yield to repurchase more SUI. Learn more 👇 sui.io/buybacks?utm_s…

Now Live : $Sui buybacks from stablecoin revenues. sui.io/buybacks?utm_s…

@EmanAbio 👀

@xBoard_app 👏🏻

A major xBoard upgrade is now live 🔥 Your wallets, chains and DeFi positions are now much easier to understand at a glance. The new Overview brings your net worth, assets, debt, rewards and allocation into one clear view, without turning your portfolio into an endless list. Wallet balances, NFTs, staking and DeFi positions now share one consistent Breakdown, while protocol and position details remain easy to explore whenever you need more depth. Navigation across wallets has been simplified, actions stay close at hand, and the mobile experience has been redesigned around the way most people actually use xBoard. This is a major step forward, but not the final one. We’ll keep improving the experience based on how people use it and on the feedback we receive. We’d genuinely love to know what feels clearer, what still feels confusing, and what you think should come next. Try the new xBoard now 🫡

@Revolut @OpenAI Why only 6 month?

From today, @openAI’s ChatGPT Go is a Revolut benefit. We’re delivering a subscription used by almost one billion people weekly, directly to our customers, at no extra cost.

@Eschalabs @AstraiaAI Benchmark?

@AstraiaAI We just released a 2bit 35b, do give it a try!

@Eschalabs 👏🏻👏🏻

@Techguy0101010 27B is almost ready, will release it soon!

@jorge_utd @nanbeige @UnslothAI 👀

@nanbeige Hey @UnslothAI any versions from you guys we can test out? Thank you.

We released Nanbeige4.2-3B, its Looped Transformer increases model capacity without adding parameters, delivering a capable 3B agent. Nanbeige4.5 is training with LoopSplit, mHC+depth attention & concatenated n-gram embeddings,already in the modeling code. huggingface.co/Nanbeige/Nanbe…

@OpenRouter @Alibaba_Qwen Benchmark?

Qwen3.7-Flash from @Alibaba_Qwen is live on OpenRouter. A fast vision capable reasoning model for multimodal agents, visual coding, search, and computer interaction, with tool use and a 1M context window. Try it now: openrouter.ai/qwen/qwen3.7-f…

@UnslothAI Someone should make it 12B 26B 34B ...

Kimi K3 can now be run locally! ✨ The 1-bit model retains ~78.9% accuracy after we shrunk it from 1.56TB to 594GB (-62% size). Run on a Mac Studio + 128GB RAM device. Kimi K3 is the strongest open model to date. Guide: unsloth.ai/docs/models/ki… GGUF: huggingface.co/unsloth/Kimi-K…

@TeksEdge Nice

🆕 llama.cpp just fixed an MTP performance bug! ~10% faster token generation with Qwen3.6-35B-A3B. When loading MTP GGUFs with --fit, llama.cpp was not counting the NextN/MTP block as part of n_gpu_layers. That could cause it to: 🧠 Offload one layer too few 🐌 Leave layer 0 running on the CPU ⚠️ Disable the fused Gated Delta Net GPU kernel 📉 Reduce token-generation performance The fix from PR #26177 is included in llama.cpp release b10152, published July 27, 2026. Any newer build should also contain it. The corrected build counts the MTP block properly, helping keep the model’s front layers on the GPU. 🚀 The contributor measured ~10% faster token generation with Qwen3.6-35B-A3B. This is not a universal 10% llama.cpp speedup but specifically addresses MTP models loaded using --fit. Source: llama.cpp PR #26177 / release b10152

@NanoGPTcom @Alibaba_Qwen Any benchmark?

Qwen3.7 Flash and Qwen3.7 Flash Thinking are now available on NanoGPT. @Alibaba_Qwen's multimodal models accept text, image, and video input, with near-1M context, tool calling, and structured outputs. Choose speed or deeper reasoning.

@Alibaba_Qwen Qwen 3.7 12B 26B 32B ??

Since launching Qwen3.8-Max-Preview, we've received valuable feedback from developers! To help users better explore Qwen3.8's agentic capabilities, we're officially launching the #QwenGrowthPlan today! 🚀 We invite you to: - Use Qwen3.8 to complete your real-world tasks - Submit your good or bad cases to us Every real task is nourishment for Qwen3.8's growth. We've prepared valuable rewards to recognize your active participation. For more detailed participation info and prizes, please check out the posters below👇. Besides emailing us, feel free to share your awesome cases or feedback directly on X and @Alibaba_Qwen Join us and let's grow together!

@Web3Twon @Alibaba_Qwen Any benchmarks

Oy what is Qwen 3.7 flash 👀 @Alibaba_Qwen 👀