merve

@mervenoyann • 93,232 subscribers

(mer-veh) open-sourceress at @huggingface 🧙🏻‍♀️ DM me for any feedback about HF 🤗 https://t.co/MhrMkGTm7p

Shorts

today's sota coding models train on hundreds of thousands of coding simulation sandboxes all you need to do so is a sandbox where the agent solves problems (vLLM + openenv + opencode), a trainer (TRL) and a verifier to check solutions we just teach you how to do it for free (no need to thank it's ok)

today's sota coding models train on hundreds of thousands of coding simulation sandboxes all you need to do so is a sandbox where the agent solves problems (vLLM + openenv + opencode), a trainer (TRL) and a verifier to check solutions we just teach you how to do it for free (no need to thank it's ok)

45,338 просмотров

I spent my Astra tokens trying to teach small Qwen3.5 (9B) to create lowpoly 3D rooms using Blender 🏡 you (or Astra) can take this challenge today here's what I learned: > you need to align geometry with deterministic rewards and VLM-as-a-judge (GLM-5.3-Flash which was a great judge) > I tried GEPA at first with various rewards, model learned to make valid rooms (but that's it) > after solving geometry you need the model to add more details, which is very hard > the stricter the reward function, less advantage you get because task is too hard. grpo didn't help > I tried passing image guidance to both policy (Qwen) and judge (GLM) it didn't improve > I tried tastemaxxing trick in this blog, truth is, flowers can be abstract, a house can't be shoutout to OpenEnv, environment is super simple what can you try? > experiment with different rewards or policy (teach a larger Qwen) > generate a dataset of 150-200 Blender render Bpy examples using Astra, fine-tune Qwen on it first and then try GEPA or GRPO > ???

I spent my Astra tokens trying to teach small Qwen3.5 (9B) to create lowpoly 3D rooms using Blender 🏡 you (or Astra) can take this challenge today here's what I learned: > you need to align geometry with deterministic rewards and VLM-as-a-judge (GLM-5.3-Flash which was a great judge) > I tried GEPA at first with various rewards, model learned to make valid rooms (but that's it) > after solving geometry you need the model to add more details, which is very hard > the stricter the reward function, less advantage you get because task is too hard. grpo didn't help > I tried passing image guidance to both policy (Qwen) and judge (GLM) it didn't improve > I tried tastemaxxing trick in this blog, truth is, flowers can be abstract, a house can't be shoutout to OpenEnv, environment is super simple what can you try? > experiment with different rewards or policy (teach a larger Qwen) > generate a dataset of 150-200 Blender render Bpy examples using Astra, fine-tune Qwen on it first and then try GEPA or GRPO > ???

22,903 просмотров

RF-DETR just landed to Hugging Face transformers 🥵🔥 sota real-time detection & segmentation models by Roboflow 💜 > play with our real-time demo > fine-tune the models on your use case with our tutorials (takes a toaster's VRAM) > or just hand them to your agents 😄

RF-DETR just landed to Hugging Face transformers 🥵🔥 sota real-time detection & segmentation models by Roboflow 💜 > play with our real-time demo > fine-tune the models on your use case with our tutorials (takes a toaster's VRAM) > or just hand them to your agents 😄

57,302 просмотров

this is the BEST vision language model I have ever tried! Aria is a new model by Rhymes.AI: a 25.3B multimodal model that can take image/video inputs 🤩 They release the model with Apache-2.0 license and fine-tuning scripts as well 👏 I tested it extensively, keep reading to learn more 🧶

this is the BEST vision language model I have ever tried! Aria is a new model by Rhymes.AI: a 25.3B multimodal model that can take image/video inputs 🤩 They release the model with Apache-2.0 license and fine-tuning scripts as well 👏 I tested it extensively, keep reading to learn more 🧶

176,161 просмотров

Real-time DEtection Transformer (RT-DETR) landed in Hugging Face transformers 🤩 with Apache 2.0 license 😍 do DETRs Beat YOLOs on Real-time Object Detection? keep reading 👀

Real-time DEtection Transformer (RT-DETR) landed in Hugging Face transformers 🤩 with Apache 2.0 license 😍 do DETRs Beat YOLOs on Real-time Object Detection? keep reading 👀

155,827 просмотров

ViTPose -- best open-source pose estimation model just landed to Hugging Face transformers 🕺🏻💃🏻 See how to use on the next one ⤵️

ViTPose -- best open-source pose estimation model just landed to Hugging Face transformers 🕺🏻💃🏻 See how to use on the next one ⤵️

83,390 просмотров

Meta released LongVU: a new video LM that can handle long videos (great performance, battle-tested by me ⚔) TLDR; 1️⃣ downsample using DINOv2 to eliminate redundant scenes 🦖 2️⃣ fuse rest of the features using DINOv2 and SigLIP 3️⃣ select some tokens, pass to Qwen2/Llama-3.2-3B

Meta released LongVU: a new video LM that can handle long videos (great performance, battle-tested by me ⚔) TLDR; 1️⃣ downsample using DINOv2 to eliminate redundant scenes 🦖 2️⃣ fuse rest of the features using DINOv2 and SigLIP 3️⃣ select some tokens, pass to Qwen2/Llama-3.2-3B

49,547 просмотров

Spaces at Hugging Face is the app store of AI 📱 it's also the MCP store now 🤠 filter thousands of MCPs you can attach to your LLM 🤗

Spaces at Hugging Face is the app store of AI 📱 it's also the MCP store now 🤠 filter thousands of MCPs you can attach to your LLM 🤗

18,690 просмотров

many parts of Hugging Face Hub is actually powered by open machine learning models 🥹 translation feature is one of them, it uses a very tiny (600M) multilingual translation model by AI at Meta 💗

many parts of Hugging Face Hub is actually powered by open machine learning models 🥹 translation feature is one of them, it uses a very tiny (600M) multilingual translation model by AI at Meta 💗

14,993 просмотров

Aya by Cohere For AI can now see! 👀 C4AI community has built Maya 8B, a new open-source multilingual VLM built on SigLIP and Aya 8B 🌱 works on 8 languages! 🗣️ The authors extend Llava dataset using Aya's translation capabilities with 558k examples! works very well ⬇️

Aya by Cohere For AI can now see! 👀 C4AI community has built Maya 8B, a new open-source multilingual VLM built on SigLIP and Aya 8B 🌱 works on 8 languages! 🗣️ The authors extend Llava dataset using Aya's translation capabilities with 558k examples! works very well ⬇️

16,226 просмотров

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

friendly reminder: this is on Monday 🤝 → learn about speculative decoding to speed-up inference → if you're a beginner and not interested, just use that does this for you automatically 🦙

friendly reminder: this is on Monday 🤝 → learn about speculative decoding to speed-up inference → if you're a beginner and not interested, just use that does this for you automatically 🦙

17,586 просмотров • 2 дней назад

Join the agent swarm that finds the ultimate compression algorithm! Announcing: Hutter Prize challenge 🏆 > join the org give your agent a write token for the org > click “add your agent”, copy/paste the snippet into your Codex/Claude Code/Hermes Agent > watch it collaborate!

Join the agent swarm that finds the ultimate compression algorithm! Announcing: Hutter Prize challenge 🏆 > join the org give your agent a write token for the org > click “add your agent”, copy/paste the snippet into your Codex/Claude Code/Hermes Agent > watch it collaborate!

54,241 просмотров • 1 месяц назад

Microsoft released a groundbreaking model that can be used for web automation, with MIT license 🔥👏 OmniParser is a state-of-the-art UI parsing/understanding model that outperforms GPT4V in parsing. 👏

Microsoft released a groundbreaking model that can be used for web automation, with MIT license 🔥👏 OmniParser is a state-of-the-art UI parsing/understanding model that outperforms GPT4V in parsing. 👏

473,275 просмотров • 2 лет назад

new open-source Bonsai models are out 🔥 > ternary weights in 8B (1.75 GB), 4B (0.86 GB), and 1.7B (0.37 GB) > comes in MLX, ONNX weights and WebGPU browser demo 😍 > a2.0 licensed 👏

new open-source Bonsai models are out 🔥 > ternary weights in 8B (1.75 GB), 4B (0.86 GB), and 1.7B (0.37 GB) > comes in MLX, ONNX weights and WebGPU browser demo 😍 > a2.0 licensed 👏

89,786 просмотров • 6 месяцев назад

if you have a mac/book and don't use llama-macos you're ngmi this beautiful piece of widget > gets your mac + your desired context window to recommend models > kicks off llama server + webui > easy switch between open models $ brew install --cask llama-app

if you have a mac/book and don't use llama-macos you're ngmi this beautiful piece of widget > gets your mac + your desired context window to recommend models > kicks off llama server + webui > easy switch between open models $ brew install --cask llama-app

33,119 просмотров • 2 месяцев назад

SAM3 is so good even cat playing didgeridoo is within distribution 🤯😂

SAM3 is so good even cat playing didgeridoo is within distribution 🤯😂

103,472 просмотров • 11 месяцев назад

DiffusionGemma is out 🔥 it's compute-bound so 4x faster compared to other Gemma-4 models (1k tok/s on H100) 💨 also great on coding, generate and iterate on any code from 3D generation to front-end ⤵️

DiffusionGemma is out 🔥 it's compute-bound so 4x faster compared to other Gemma-4 models (1k tok/s on H100) 💨 also great on coding, generate and iterate on any code from 3D generation to front-end ⤵️

40,717 просмотров • 4 месяцев назад

Inkling Thinking Machines 1-bit quant by Unsloth AI running 30-40 TPS 🤯 video sped up, feel free to jump at the end to check outputs also llama.cpp webui is 😍 has reasoning slider, html preview, mcp and multimodal support

Inkling Thinking Machines 1-bit quant by Unsloth AI running 30-40 TPS 🤯 video sped up, feel free to jump at the end to check outputs also llama.cpp webui is 😍 has reasoning slider, html preview, mcp and multimodal support

25,498 просмотров • 3 месяцев назад

Google released MedGemma on I/O'25 👏 > 4B and 27B instruction fine-tuned vision LMs and a 4B pre-trained vision LM for medicine > available with transformers from the get-go 🤗 they also released a cool demo for scan reading ⤵️

Google released MedGemma on I/O'25 👏 > 4B and 27B instruction fine-tuned vision LMs and a 4B pre-trained vision LM for medicine > available with transformers from the get-go 🤗 they also released a cool demo for scan reading ⤵️

81,382 просмотров • 1 год назад

New 🤗 transformers release includes a very powerful Multimodel Large Language Model (MLLM) by Microsoft called KOSMOS-2! 🤩 The highlight of KOSMOS-2 is grounding, the model is *incredibly* accurate! 🌎 Play with the demo here 👉 But how does this model work? Let's take a look! 👀🧶

New 🤗 transformers release includes a very powerful Multimodel Large Language Model (MLLM) by Microsoft called KOSMOS-2! 🤩 The highlight of KOSMOS-2 is grounding, the model is incredibly accurate! 🌎 Play with the demo here 👉 But how does this model work? Let's take a look! 👀🧶

143,830 просмотров • 3 лет назад

real-time vision in your browser 🔥 try out YOLO26 for pose estimation and detection built on WebGPU ⚡️

real-time vision in your browser 🔥 try out YOLO26 for pose estimation and detection built on WebGPU ⚡️

46,395 просмотров • 9 месяцев назад

Google just released PaliGemma 2 Mix: new versatile instruction vision language models 🔥 > Three new models: 3B, 10B, 28B with res 224, 448 💙 > Can do vision language tasks with open-ended prompts, understand documents, and segment or detect anything 🤯

Google just released PaliGemma 2 Mix: new versatile instruction vision language models 🔥 > Three new models: 3B, 10B, 28B with res 224, 448 💙 > Can do vision language tasks with open-ended prompts, understand documents, and segment or detect anything 🤯

83,751 просмотров • 1 год назад

MatchAnything is an insane framework authors have tried to get every view that they can and dump them to modern keypoint matching models for instance you can match iphone map view to google aerial view, thermal camera views to day view even if the images are warped!

MatchAnything is an insane framework authors have tried to get every view that they can and dump them to modern keypoint matching models for instance you can match iphone map view to google aerial view, thermal camera views to day view even if the images are warped!

58,148 просмотров • 1 год назад

agents are now running Hugging Face dataset viewers 🤯

agents are now running Hugging Face dataset viewers 🤯

30,122 просмотров • 8 месяцев назад

IBM dropped CUGA, open-source enterprise agent to automate boring tasks 🔥 > given workspace files, it writes and executes code to accomplish any task 🤯 > comes with a ton of tools built for enterprise tasks, supports MCPs > plug in your favorite LLM 👏 here's a small demo where it retrieves info from a file, calculates revenue by writing code, and drafts an e-mail 🤯 they release code, a blog and a demo 🙌🏻 you can run this locally

IBM dropped CUGA, open-source enterprise agent to automate boring tasks 🔥 > given workspace files, it writes and executes code to accomplish any task 🤯 > comes with a ton of tools built for enterprise tasks, supports MCPs > plug in your favorite LLM 👏 here's a small demo where it retrieves info from a file, calculates revenue by writing code, and drafts an e-mail 🤯 they release code, a blog and a demo 🙌🏻 you can run this locally

32,984 просмотров • 10 месяцев назад

JUST IN: Inference Providers at Hugging Face Hub Use @FAL SambaNova Together AI Replicate on Hugging Face to infer gigantic models like DeepSeek R1 from model page or through our client SDKs (Python, JS)/HTTP calls

JUST IN: Inference Providers at Hugging Face Hub Use @FAL SambaNova Together AI Replicate on Hugging Face to infer gigantic models like DeepSeek R1 from model page or through our client SDKs (Python, JS)/HTTP calls

36,902 просмотров • 1 год назад

Lyria by Google DeepMind is next level it did average Turkish pop song very well all my Turkish friends in the conference hall was summoned lol

Lyria by Google DeepMind is next level it did average Turkish pop song very well all my Turkish friends in the conference hall was summoned lol

14,752 просмотров • 6 месяцев назад

document AI landed to Hugging Face chat, powered by open-source models (Cohere Command-R+) with tool calling 🛠️✨ explore many tools including image generation, image editing and more at Hugging Chat 🤗💬

document AI landed to Hugging Face chat, powered by open-source models (Cohere Command-R+) with tool calling 🛠️✨ explore many tools including image generation, image editing and more at Hugging Chat 🤗💬

31,986 просмотров • 2 лет назад