Hugging Face
260 subscribers
1.34K photos
408 videos
2.19K links
Download Telegram
Hugging Face (Twitter)

RT @Zai_org: Introducing GLM-5: From Vibe Coding to Agentic Engineering

GLM-5 is built for complex systems engineering and long-horizon agentic tasks. Compared to GLM-4.5, it scales from 355B params (32B active) to 744B (40B active), with pre-training data growing from 23T to 28.5T tokens.

Try it now: chat.z.ai
Weights: huggingface.co/zai-org/GLM-5
Tech Blog: z.ai/blog/glm-5
OpenRouter (Previously Pony Alpha): openrouter.ai/z-ai/glm-5
Rolling out from Coding Plan Max users: z.ai/subscribe
Hugging Face (Twitter)

RT @mervenoyann: GLM-5 is out on @huggingface πŸ”₯

> A40B/744B, trained on more tokens (28.5T)
> outperforms/on par with closed sota
> allows commercial use (MIT licensed) πŸ’—

use with vLLM/SGLang locally or through HF Inference Providers thanks to @novita_labs and @Zai_org πŸ“¦
Hugging Face (Twitter)

RT @lvwerra: Public benchmarks lag behind what frontier labs are using internally to test and develop LLMs, yet they are the key driver of progress for LLMs.

This needs to change!

Excited to work with @SnorkelAI who are investing $3M do build out the evaluation ecosystem with the community.
Hugging Face (Twitter)

RT @Xianbao_QIAN: Pony Alpha is open sourced finally. Welcome GLM 5 by @Zai_org on @huggingface !

- 744B parameters (40B active)
- 28.5T pretrain tokens
- DeepSeek Sparse Attention (DSA)
- Trained by asynchronous RL infrastructure slime
- MIT license
- Native FP8 version available

huggingface.co/zai-org/GLM-5
This media is not supported in your browser
VIEW IN TELEGRAM
Hugging Face (Twitter)

RT @andimarafioti: 30x real-time speech-to-text in your browser.
No installs. No servers.
Just open the website.
Hugging Face (Twitter)

RT @vanstriendaniel: Data Designer from @nvidia now has full @huggingface Hub integration:
- load any dataset as a seed
- generate synthetic data,
- push results straight to the Hub
Hugging Face (Twitter)

RT @HuggingPapers: StepFun's Step 3.5 Flash

A sparse MoE model with 196B parameters, 11B active per token.

Achieves frontier-level reasoning comparable to GPT-5.2 xHigh and Gemini 3.0 Pro at 1/6th the decoding cost.

Ranks #1 on MathArena with 97.3% on AIME 2025.
Hugging Face (Twitter)

RT @turingcom: OpenEnv treats environments as first class infrastructure.

Through our collaboration with @AIatMeta and @huggingface, Turing is helping labs run tool-using agents against RL Environments that share:

-A standard step and reset API
-WebSocket sessions with per-client state
-MCP-style tool discovery and calling
-Observability hooks for rewards, errors, and drift

The payoff is simple. You can reuse the same evaluation pattern across domains and see where agents actually fail in long, tool-heavy workflows. Learn more below.
Hugging Face (Twitter)

RT @RisingSayak: An ambitious project today πŸ”₯

We got an agent to write custom kernels that actually work for a given model, hardware instruction set, and other relevant model-dependent constraints.

Benchmarks are our rewards here πŸ€ͺ

We got these kernels to work with Diffusers and `torch.compile` and they delivered ACTUAL SPEEDUP without messing up the quality πŸ†

Despite the competitive landscape, we don't like to keep things private. Read all of it in the blog post below:
https://huggingface.co/blog/custom-cuda-kernels-agent-skills
Hugging Face (Twitter)

RT @_lewtun: We trained a tiny 4B model to reason for millions of tokens through IMO-level problems.

Heaps excited to share our new blog post covering the full pipeline, from distilling the 🐳 to augmenting RL with a reasoning cache that unlocks extreme inference-time scaling for theorem proving.

https://huggingface.co/spaces/lm-provers/qed-nano-blogpost
πŸ‘1
Hugging Face (Twitter)

RT @j_dekoninck: Introducing QED-Nano: a 4B model for mathematical proof writing, competitive with larger models like GPT-OSS-120B.

We open-source our entire pipeline, including data, code, and a blog post, hoping that the community can build on these artifacts to create more specialized models.
Hugging Face (Twitter)

RT @RisingSayak: Will be there at the @OfficialINDIAai Impact Summit in Delhi from 17-19.

Will also present ReflectionFlow at the symposium on the 18th.

Kinda surprised there’s no real discussion about open science and open source, given the stellar speakers.

Anyway, looking forward to it!
⚑1πŸ‘1
Hugging Face (Twitter)

RT @Alibaba_Qwen: πŸš€ Qwen3.5-397B-A17B is here: The first open-weight model in the Qwen3.5 series.

πŸ–ΌοΈNative multimodal. Trained for real-world agents.
✨Powered by hybrid linear attention + sparse MoE and large-scale RL environment scaling.
⚑8.6x–19.0x decoding throughput vs Qwen3-Max
🌍201 languages & dialects
πŸ“œApache2.0 licensed

πŸ”—Dive in:
GitHub: github.com/QwenLM/Qwen3.5
Chat: chat.qwen.ai
API:https://modelstudio.console.alibabacloud.com/ap-southeast-1/?tab=doc#/doc/?type=model&url=2840914_2&modelId=group-qwen3.5-plus
Qwen Code: github.com/QwenLM/qwen-code
Hugging Face: https://huggingface.co/collections/Qwen/qwen35
ModelScope: https://modelscope.cn/collections/Qwen/Qwen35
blog: qwen.ai/blog?id=qwen3.5
πŸ‘€1
Hugging Face (Twitter)

RT @AdinaYakup: Happy Spring Festival🧧🐎
Here’s to another year of building and sharing!
ζ–°ηš„δΈ€εΉ΄οΌŒη»§η»­εΌ€ζΊεŒθ‘Œ πŸ€—
πŸ₯°1
Hugging Face (Twitter)

RT @Cohere_Labs: Very special to work with our @huggingface friends to bring Tiny Aya, the most capable multilingual open-weight model at its scale to the world! πŸš€

Big thanks to @ngxson for the huge help merging the changes into llama.cpp.
πŸ‘1
This media is not supported in your browser
VIEW IN TELEGRAM
Hugging Face (Twitter)

RT @Cohere_Labs: Introducing ✨Tiny Aya✨, a family of massively multilingual small language models built to run where people actually are.

Tiny Aya delivers strong multilingual performance in 70+ global languages in a 3.35B parameter model, efficient enough to run locally, even on a phone.
Hugging Face (Twitter)

RT @evijit: Today, @evaluatingevals is introducing Every Eval Ever, a unified, open data format and public dataset for AI evaluation results.
AI telegram bot.

β€” @aigram
😁1