ModelScope

894 posts

ModelScope

@ModelScope2022

Driving innovations with open communities. 💬 Join our Discord: https://t.co/4M9AHVh5qa

HangZhou, China Katılım Nisan 2024

151 Takip Edilen10.6K Takipçiler

ModelScope@ModelScope2022·9h

Robot videos are messy. WGO-Bench asks a simple question: can VLMs tell what is happening, when it starts, and when it ends? 🤖 modelscope.ai/datasets/macro… 🎥 100 robot and egocentric manipulation episodes with 743 human-labeled subtask segments across HomER, RoboInter DROID, and RoboCOIN Galaxea R1 Lite sources. ⏱️ Two core tasks: boundary detection for meaningful manipulation steps, and subtask labeling for completed events like picking, placing, opening, closing, pouring, wiping, or transferring objects. 🧩 Built for downstream robot learning: each segment includes start/end timestamps and a concise visible-action label, useful for policy training, reward modeling, and automatic annotation pipelines. ✒️ License: CC BY-NC-SA 4.0.

English

1.5K

ModelScope@ModelScope2022·14h

OvisOCR2 just released on ModelScope, an exclusive 0.8B end-to-end OCR model for page-level document parsing. 🚀 License: Apache 2.0. 🤖 modelscope.ai/models/ATH-Maa… 🏆 OmniDocBench v1.6 SOTA: 96.58 overall, the first end-to-end model to rank #1 on a leaderboard previously led by pipeline systems. 📄 PureDocBench leader: highest Avg3 score at 75.06, showing strong page-level document parsing beyond a single benchmark. 🧩 Structured output: generates natural-reading-order Markdown, formats formulas as LaTeX, tables as HTML, and visual regions as image tags with bbox coordinates. ⚙️ Compact deployment: post-trained from Qwen3.5-0.8B with SFT, RL, and OPD, with vLLM inference support.

English

2.6K

ModelScope@ModelScope2022·1d

@DJLougen @elonmusk you are the best😂

English

157

Daniel Lougen@DJLougen·1d

sorry @elonmusk

ModelScope@ModelScope2022

Wan-Dancer is here! A music-driven portrait dance video generation model from Wan. 14B params, now open source. 🚀 🤖modelscope.ai/models/Wan-AI/… 🎮 Demo time👉modelscope.ai/studios/Wan-AI… One photo + one music track → up to 3 minutes of high-quality dance video at 720p/30fps. Breaks the 20-second barrier that limits most existing methods. 🎵 5 dance styles: Chinese classical, K-Pop, street dance, tap, Latin 🧠 Two-stage DiT pipeline: Global DiT plans keyframes from full music context, Local DiT refines motion details and frame transitions ⚡ RoPE-based temporal alignment keeps motion precisely synced to beat across long sequences ✅ Keyframe editing supported. LoRA fine-tuning available. DiffSynth-Studio compatible.

English

745

ModelScope@ModelScope2022·1d

English

289

14.6K

ModelScope@ModelScope2022·1d

🌟 Build your first AI app with Gradio on ModelScope Studio. In this step-by-step tutorial, we show how to upload a simple Python project, choose Gradio as the framework, deploy with free CPU resources, and turn it into an online interactive demo. Watch the full tutorial: youtube.com/watch?v=SFlsNh… 🔗modelscope.ai/studios

YouTube

English

1.3K

ModelScope@ModelScope2022·3d

@aiseomastery sry! it's actually a 5B model. we wrongly put the info above. just corrected in the comment.

English

105

AI Mastery Guide@aiseomastery·3d

@ModelScope2022 A 0.5B model doing real time interactive world simulation on a single desktop setup is impressive 🌆

English

164

ModelScope@ModelScope2022·4d

A real-time world simulator, now in 0.5B. ABot-World-0.5B-LF brings action-conditioned interactive rollout to ModelScope. 🎮License: Apache 2.0. 🤖 modelscope.ai/models/amap_cv… ⚡ Desktop-ready rollout: reported at 720p, 16 FPS, 1.2s latency, and 19GB GPU memory on a single-desktop setup. 🌍 Infinite interaction: responds to user actions continuously and keeps generating beyond fixed video-length limits. 🎬 LongForcing training helps the model expand new scenes and dynamics during rollout, without prompt switching or scene lock-in. 🛠️ Built on Wan2.2-TI2V-5B as a causal student model for lower-latency interactive world rollout.

English

238

18.8K

ModelScope@ModelScope2022·3d

Correction: ABot-World-0-5B-LF is the 5B LongForcing variant of ABot-World-0, not a 0.5B model. Apologies for the mistake.🙇‍♂️

English

843

ModelScope@ModelScope2022·4d

Huge congratulations to @MiniMax_AI on the $2B financing milestone! 🚀 MiniMax’s CEO has committed 1% of the company’s total equity to supporting the open-source community over the long term. We look forward to our continued collaboration and to bringing more great models to the open-source community together!🤖

RyanLee@RyanLeeMiniMax

I’m incredibly excited to share this: MiniMax has just closed a new $2B funding round. 🚀 At the same time, our CEO, IO, shared three long-term commitments with the team: • No salary until we achieve AGI. • Over the next four years, he will dedicate shares equivalent to 4% of the company’s total equity from his personal holdings to reward employees who are building MiniMax for the long term. • Another 1% will be committed to supporting the open-source community. The funding is exciting. But what excites me even more is what it represents: a long-term commitment to AGI, to our people, and to the open-source ecosystem. We’re living through one of the most exciting moments in the history of AI, and we’re just getting started. If you’re passionate about frontier AI, open source, and building the future, we’d love to build with you. Intelligence with Everyone. 🚀

English

11K

ModelScope@ModelScope2022·5d

MolParser-Mobile is here for ultrafast chemical structure recognition: a 9.98M model that turns molecular images into E-SMILES at 1,520 molecules/sec. ⚗️ 🤖 modelscope.ai/models/UniPars… 🏆 SOTA recognition accuracy on BioVista: MolParser-Mobile reaches 0.801, +0.098 over MolParser 1.0. 📊 Speed + size: 95.4% smaller than MolParser 1.0 and 38x faster, moving from 39.8 to 1,520 molecules/sec on the reported setup. 🔬 Built for scale: designed for web-scale chemical literature mining and industrial OCSR workflows. 🛠️ Use it through the MolParser library or load directly with Transformers. License: cc-by-nc-sa-4.0

English

1.7K

ModelScope@ModelScope2022·5d

@jun_song It’s very nice talking to you! Hope you love our swag!

English

173

Jun Song@jun_song·5d

Just finished Qwen meetup Seoul. Got some modelscope merch too. Once I saw the picture from today’s speech realized I gained so much weight after quitting from the job. Need to do some fat cutting now.

JEFF l AI 🦞@0xjeffai

Spoke at @Alibaba_Qwen Korea meetup today, sharing an AI agent automation perspective as a non-developer. Got to meet Qwen ambassador legend @jun_song and had some really meaningful conversations with the Qwen team 🙏

English

3.3K

ModelScope@ModelScope2022·5d

SenseNova-Vision-7B-MoT is live on ModelScope: a unified multimodal generation model for computer vision. 🚀 🤖 modelscope.ai/models/SenseNo… 📄 modelscope.ai/papers/2607.18… 📊 Strong against generalist vision models: compared with Youtu-VL, it reports higher COCO detection mAP, Cityscapes mIoU, RefCOCO / RefCOCOg grounding, and NYUv2 depth δ1. 🧩 One model, many output types: structured text for boxes, points, OCR, GUI grounding, keypoints, and camera records; image-like outputs for masks, depth, normals, and point maps. 🧠 No task-specific heads: visual tasks are expressed through text, image, or mixed text-image generation, with outputs decoded back into benchmark formats. 📚 Trained on SenseNova-Vision-Corpus-50M, covering structured visual understanding, dense geometry, segmentation, and multi-view visual geometry. License: CC BY-NC 4.0.

English

8.9K

ModelScope@ModelScope2022·5d

@YukiVerse_HK @0xKingsKuan 啊哈～怎么没有来 say hi？我今天就戴了 Qwen Glasses

中文

YukiVerse@YukiVerse_HK·5d

@0xKingsKuan 模型厂商肯定赞助呀，还有不少社区业赞助，今天台上有一个叫 @ModelScope2022 杭州的社区

中文

YukiVerse@YukiVerse_HK·5d

Qwen Meetup Seoul 🇰🇷 千问和DeepSeek在韩国积极布局中国制造的大模型出海亚洲还是易如反掌的~~~ 听不懂韩语只能蹭个饭了🙂 此刻需要翻译眼镜

Gangnam-gu, Republic of Korea 🇰🇷 中文

2.2K

ModelScope@ModelScope2022·5d

MOSS-Transcribe-Diarize 0.9B is here: one pass from long-form audio to speaker-aware transcripts with timestamps, diarization, and acoustic event tags. 🎧 License: Apache 2.0. 🤖 modelscope.ai/models/openmos… 📄 modelscope.ai/papers/2601.01… 📊 Benchmark: the 0.9B model beats Doubao, ElevenLabs, Gemini 3 Pro, and VIBEVOICE ASR on AISHELL-4 and Alimeeting, while also leading available baselines on Podcast CER/cpCER. 🎙️ Timestamped + long-form: handles hour-long audio and outputs structured segments with start/end timestamps and stable speaker IDs. 🌐 Multilingual + event-aware: auto-detects language per segment, preserves speaker tags across languages, and can mark non-speech events, overlap, and interruptions. ⚙️ Built with a Qwen3-0.6B style text backbone, Whisper-Medium audio encoder, and an audio-text adaptor for joint transcription + diarization.

English

161

9.6K

ModelScope@ModelScope2022·5d

LingBot-World-V2 brings real-time, playable world modeling to ModelScope. 🌍 🤖 modelscope.ai/models/Robbyan… 🎮 More ways to act: attacking, archery, spell-casting, shooting, and text-triggered scene events. ⚡ Real-time feel: causal-fast inference with KV caching, built for rapid response and 720p, 60fps interactive streams. 🧠 Agentic scenes: a pilot agent controls character behavior, while a director agent keeps adding new environment elements as the world evolves. License: CC BY-NC-SA 4.0 for non-commercial use.

English

4.2K

ModelScope@ModelScope2022·5d

LingBot-Video lands on ModelScope as a three-model open-source MoE video foundation model series for embodied intelligence. 🚀 🤖 modelscope.ai/collections/Ro… 📊 Benchmarks cover T2V, TI2V, and RBench across video quality, embodied domains, and robotic tasks, with LingBot-Video reaching 0.620 average on RBench. ⚡ Sparse MoE scaling keeps long-video generation efficient: the 30B-A3B setup reaches about 3.18x speedup over Dense-30B at 1M tokens. 🧠 Built with a Single-Stream DiT, sparse expert routing, 70,000+ hours of embodiment-oriented data, and reward-aligned post-training. What it unlocks: • Text/image/structured prompts to open-world video • Action-to-Video simulation for robot future trajectories • Use cases across data synthesis, policy evaluation, and action planning

English

3.6K

ModelScope@ModelScope2022·6d

R3-Embedding-0.6B is now on ModelScope, a compact bi-encoder for query-conditional agent skill retrieval. 🧭 🤖 modelscope.ai/models/Tencent… 📄 modelscope.ai/papers/2606.03… 🧩 Built for skill routing: retrieves the right tool or agent skill for a user request, instead of treating skills like generic text documents. 🔎 Recall stage: embeds user queries and agent skills independently, then ranks candidates by cosine similarity. ⚙️ Two-stage setup: pairs with R3-Rerank-0.6B for reranking after fast embedding-based retrieval. Base: Qwen3-Embedding-0.6B. License: Apache 2.0.