This media is not supported in your browser
VIEW IN TELEGRAM
Hugging Face (Twitter)
RT @hugobowne: Training big models used to be reserved for OpenAI or DeepMind.
Now? Builders everywhere have access to clusters of 4090s, Modal credits, and open-weight models like LLaMA 3 and Qwen. 🛠️
In this episode of @VanishingData, @TheZachMueller (@huggingface ), joins me to break down what scaling actually looks like in 2025 for individual devs and small teams:
• When to leave Colab and how not to drown in infra the moment you do
• How Accelerate simplifies training and inference across multiple GPUs
• Why “data parallelism” is just the start and where things break
• Lessons from helping everyone from solo devs to research labs scale up
• What people still get wrong about distributed training and inference
Links in 🧵
1/
RT @hugobowne: Training big models used to be reserved for OpenAI or DeepMind.
Now? Builders everywhere have access to clusters of 4090s, Modal credits, and open-weight models like LLaMA 3 and Qwen. 🛠️
In this episode of @VanishingData, @TheZachMueller (@huggingface ), joins me to break down what scaling actually looks like in 2025 for individual devs and small teams:
• When to leave Colab and how not to drown in infra the moment you do
• How Accelerate simplifies training and inference across multiple GPUs
• Why “data parallelism” is just the start and where things break
• Lessons from helping everyone from solo devs to research labs scale up
• What people still get wrong about distributed training and inference
Links in 🧵
1/
This media is not supported in your browser
VIEW IN TELEGRAM
Hugging Face (Twitter)
RT @NVIDIAAIDev: 🎶 Meet Audio-Flamingo 3 – a fully open LALM trained on sound, speech, and music datasets. 🎶
Handles 10-min audio, long-form text, and voice conversations. Perfect for audio QA, dialog, and reasoning.
On @huggingface ➡️ https://huggingface.co/nvidia/audio-flamingo-3
From #NVIDIAResearch.
RT @NVIDIAAIDev: 🎶 Meet Audio-Flamingo 3 – a fully open LALM trained on sound, speech, and music datasets. 🎶
Handles 10-min audio, long-form text, and voice conversations. Perfect for audio QA, dialog, and reasoning.
On @huggingface ➡️ https://huggingface.co/nvidia/audio-flamingo-3
From #NVIDIAResearch.
Hugging Face (Twitter)
RT @reach_vb: Qwen COOKED - beats Kimi K2 and competitive to Claude Opus 4 at 25% total parameters 🤯
RT @reach_vb: Qwen COOKED - beats Kimi K2 and competitive to Claude Opus 4 at 25% total parameters 🤯
Hugging Face (Twitter)
RT @reach_vb: missed this, @NVIDIAAIDev silently dropped Open Reasoning Nemotron models (1.5-32B), SoTA on LiveCodeBench, CC-BY 4.0 licensed 🔥
> 32B competing with Qwen3 235B and DeepSeek R1
> Available across 1.5B, 7B, 14B and 32B size
> Supports upto 64K output tokens
> Utilises GenSelect (combines multiple parallel generations)
> Built on top of Qwen 2.5 series
> Allows commercial usage
Works out of the box in transformers, vllm, mlx, llama.cpp and more!
RT @reach_vb: missed this, @NVIDIAAIDev silently dropped Open Reasoning Nemotron models (1.5-32B), SoTA on LiveCodeBench, CC-BY 4.0 licensed 🔥
> 32B competing with Qwen3 235B and DeepSeek R1
> Available across 1.5B, 7B, 14B and 32B size
> Supports upto 64K output tokens
> Utilises GenSelect (combines multiple parallel generations)
> Built on top of Qwen 2.5 series
> Allows commercial usage
Works out of the box in transformers, vllm, mlx, llama.cpp and more!
Hugging Face (Twitter)
RT @lhoestq: A new Pandas feature landed 3 days ago and no one noticed.
Upload ONLY THE NEW DATA to dedupe-based storage like @huggingface (Xet). Data that already exist in other files don't need to be uploaded.
Possible thanks to the recent addition of Content Defined Chunking for Parquet.
RT @lhoestq: A new Pandas feature landed 3 days ago and no one noticed.
Upload ONLY THE NEW DATA to dedupe-based storage like @huggingface (Xet). Data that already exist in other files don't need to be uploaded.
Possible thanks to the recent addition of Content Defined Chunking for Parquet.
Hugging Face (Twitter)
RT @casper_hansen_: This is not a SMALL update. This is huge! Give us this for every model please Qwen team🙏
RT @casper_hansen_: This is not a SMALL update. This is huge! Give us this for every model please Qwen team🙏
Hugging Face (Twitter)
RT @nic_o_martin: Beyond happy to announce that I'm joining 🤗 @huggingface as a #MachineLearningEngineer focused on #WebML!
RT @nic_o_martin: Beyond happy to announce that I'm joining 🤗 @huggingface as a #MachineLearningEngineer focused on #WebML!
Hugging Face (Twitter)
RT @ClementDelangue: Now number one trending dataset on @huggingface, out of almost half a million! huggingface.co/datasets https://twitter.com/NousResearch/status/1945181587600982450#m
RT @ClementDelangue: Now number one trending dataset on @huggingface, out of almost half a million! huggingface.co/datasets https://twitter.com/NousResearch/status/1945181587600982450#m
Hugging Face (Twitter)
RT @MaziyarPanahi: Perfect Sunday: I just used Kimi-K2 by @Kimi_Moonshot to vibe code a @Gradio app! 🔥
You can use "Anycoder" Space by @_akhaliq hosted on @huggingface for free. It was super quick! 🤗
PS: I am aware of using Gradio to vibe code another Gradio! Pun very much intended here! 😂
RT @MaziyarPanahi: Perfect Sunday: I just used Kimi-K2 by @Kimi_Moonshot to vibe code a @Gradio app! 🔥
You can use "Anycoder" Space by @_akhaliq hosted on @huggingface for free. It was super quick! 🤗
PS: I am aware of using Gradio to vibe code another Gradio! Pun very much intended here! 😂
This media is not supported in your browser
VIEW IN TELEGRAM
Hugging Face (Twitter)
RT @AdinaYakup: From paper to project page in one click🚀
AnyCoder 🔥 turns research PDFs into structured, shareable project pages in seconds!
https://huggingface.co/spaces/akhaliq/anycoder
Powered by 8 SoTA open models on @huggingface
RT @AdinaYakup: From paper to project page in one click🚀
AnyCoder 🔥 turns research PDFs into structured, shareable project pages in seconds!
https://huggingface.co/spaces/akhaliq/anycoder
Powered by 8 SoTA open models on @huggingface
This media is not supported in your browser
VIEW IN TELEGRAM
Hugging Face (Twitter)
RT @vitrupo: Jack Dorsey says AI must be permissionless because constraint kills innovation.
Five CEOs shouldn't dictate what brings humanity forward.
Open source is the answer.
To protect ourselves, we have to race ahead. Eliminating single points of failure before they become civilization's choke points.
RT @vitrupo: Jack Dorsey says AI must be permissionless because constraint kills innovation.
Five CEOs shouldn't dictate what brings humanity forward.
Open source is the answer.
To protect ourselves, we have to race ahead. Eliminating single points of failure before they become civilization's choke points.
Hugging Face (Twitter)
RT @yagilb: I'm not sure how HF is paying for all those TBs going in and out, but at least now we're chipping in a little bit. Thanks @huggingface for being the great library of AI models for us all 🙏
RT @yagilb: I'm not sure how HF is paying for all those TBs going in and out, but at least now we're chipping in a little bit. Thanks @huggingface for being the great library of AI models for us all 🙏
Hugging Face (Twitter)
RT @HaihaoShen: 🥳INT4 model for updated Qwen3-235B-A22B:
vLLM MoE seems not working well; yet HF transformers can run pretty well.
RT @HaihaoShen: 🥳INT4 model for updated Qwen3-235B-A22B:
vLLM MoE seems not working well; yet HF transformers can run pretty well.
huggingface.co
Intel/Qwen3-235B-A22B-Instruct-2507-int4-mixed-rtn-AutoRound-inc · Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Hugging Face (Twitter)
RT @intrstllrninja: today i'm releasing 50k rows of tool-use reasoning dataset compilation on huggingface
includes following BFCL scenarios:
- single turn tool-use
- multiturn tool-use
- multistep tool-use
- relevance reasoning
https://huggingface.co/datasets/interstellarninja/hermes_reasoning_tool_use
RT @intrstllrninja: today i'm releasing 50k rows of tool-use reasoning dataset compilation on huggingface
includes following BFCL scenarios:
- single turn tool-use
- multiturn tool-use
- multistep tool-use
- relevance reasoning
https://huggingface.co/datasets/interstellarninja/hermes_reasoning_tool_use
huggingface.co
interstellarninja/hermes_reasoning_tool_use · Datasets at Hugging Face
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
Hugging Face (Twitter)
RT @vllm_project: The @huggingface Transformers ↔️ @vllm_project integration just leveled up: Vision-Language Models are now supported out of the box!
If the model is integrated into Transformers, you can now run it directly with vLLM.
https://github.com/vllm-project/vllm/pull/20543
Great work @RTurganbay 👏
RT @vllm_project: The @huggingface Transformers ↔️ @vllm_project integration just leveled up: Vision-Language Models are now supported out of the box!
If the model is integrated into Transformers, you can now run it directly with vLLM.
https://github.com/vllm-project/vllm/pull/20543
Great work @RTurganbay 👏
Hugging Face (Twitter)
RT @ClementDelangue: It's out! and you can already run inference on the HF model page thanks to @hyperbolic_labs! https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct https://twitter.com/ClementDelangue/status/1947753298879975771#m
RT @ClementDelangue: It's out! and you can already run inference on the HF model page thanks to @hyperbolic_labs! https://huggingface.co/Qwen/Qwen3-Coder-480B-A35B-Instruct https://twitter.com/ClementDelangue/status/1947753298879975771#m
Hugging Face (Twitter)
RT @itsPaulAi: Wait so Alibaba Qwen has just released ANOTHER model??
Qwen3-Coder is simply one of the best coding model we've ever seen.
→ Still 100% open source
→ Up to 1M context window 🔥
→ 35B active parameters
→ Same performance as Sonnet 4
They're releasing a CLI tool as well ↓
RT @itsPaulAi: Wait so Alibaba Qwen has just released ANOTHER model??
Qwen3-Coder is simply one of the best coding model we've ever seen.
→ Still 100% open source
→ Up to 1M context window 🔥
→ 35B active parameters
→ Same performance as Sonnet 4
They're releasing a CLI tool as well ↓
Hugging Face (Twitter)
RT @AdinaYakup: Qwen3-Coder 💻 agentic code model by @Alibaba_Qwen
https://huggingface.co/collections/Qwen/qwen3-coder-687fc861e53c939e52d52d10
✨ 480B total, 35B activated MoE
✨ Agentic Coding + Browser Use → Top code model performance
✨ 256K context (up to 1M via Yarn) for repo-scale understanding
RT @AdinaYakup: Qwen3-Coder 💻 agentic code model by @Alibaba_Qwen
https://huggingface.co/collections/Qwen/qwen3-coder-687fc861e53c939e52d52d10
✨ 480B total, 35B activated MoE
✨ Agentic Coding + Browser Use → Top code model performance
✨ 256K context (up to 1M via Yarn) for repo-scale understanding