English
English Indonesia Türkçe Deutsch Español Português Pусский Français Italiano Nederlands Polski العربية ไทย Tiếng Việt 繁體中文 简体中文 日本語 한국어
TheInclusionAI's profile picture. InclusionAI (IAI) envisions AGI as humanity's shared milestone. See @AntLingAGI model series, and OSS projects like AReaL & AWorld https://discord.gg/2X4zBSz9c6

InclusionAI

@TheInclusionAI

InclusionAI (IAI) envisions AGI as humanity's shared milestone. See @AntLingAGI model series, and OSS projects like AReaL & AWorld https://discord.gg/2X4zBSz9c6

66Posts 390Followers 5Following
TheInclusionAI's profile picture.
InclusionAI
@TheInclusionAI
Sep 29

A remarkable moment on scaling！

AntLingAGI's profile picture. MoE model series with foundation (Ling), reasoning (Ring) and any-to-any (Ming) from Ant Group’s AGI initiative, @TheInclusionAI. https://t.co/jtwwGFLA1j
Ant Ling
@AntLingAGI
Sep 29

🚀 Ring-1T-preview: Deep Thinking, No Waiting The first 1 trillion open-source thinking model -> Early results in natural language: AIME25/92.6, HMMT25/84.5, ARC-AGI-1/50.8, LCB/78.3, CF/94.7 -> Solved IMO25 Q3 in one shot, with partial solutions for Q1/Q2/Q4/Q5 Still evolving!

AntLingAGI's tweet image. 🚀 Ring-1T-preview: Deep Thinking, No Waiting The first 1 trillion open-source thinking model -> Early results in natural language: AIME25/92.6, HMMT25/84.5, ARC-AGI-1/50.8, LCB/78.3, CF/94.7 -> Solved IMO25 Q3 in one shot, with partial solutions for Q1/Q2/Q4/Q5 Still evolving!
75
232
2K
486
501K
0
1
11
0
710
TheInclusionAI's profile picture.
InclusionAI
@TheInclusionAI
Sep 26

Ring-flash-linear-2.0 ：cost effective ，as fast as flashlight ⚡️

AntLingAGI's profile picture. MoE model series with foundation (Ling), reasoning (Ring) and any-to-any (Ming) from Ant Group’s AGI initiative, @TheInclusionAI. https://t.co/jtwwGFLA1j
Ant Ling
@AntLingAGI
Sep 26

🚀Meet Ring-flash-linear-2.0 & Ring-mini-linear-2.0 --> ultra-fast, SOTA reasoning LLMs with hybrid linear attentions --> 2x faster than same-size MoE & 10x faster than 32B models --> Enhanced with advanced RL methods Try the future of reasoning!

17
76
788
213
219K
0
1
14
2
2K
InclusionAI reposted
zzqsmall's profile picture.
zzqsmall
@zzqsmall
Sep 23

amazing try with our baby moe model, good reason to buy a new iphone 17

awnihannun's profile picture. AI @apple
Awni Hannun
@awnihannun
Sep 22

Managed to get Ling Mini 16B (1.4B active) running on my iPhone Air. It runs very fast with MLX. It's a DWQ of Ling Mini quantized to 3 bits-per-weight. A 16B model running on an Air at this speed is pretty awesome:

19
34
527
204
92K
0
1
5
0
544
TheInclusionAI's profile picture.
InclusionAI
@TheInclusionAI
Sep 23

Nice work！We released Ling-flash-2.0, Ring -flash-2.0, you can try more and talk to us.😇

adrgrondin's profile picture. iOS Developer @Match @MeeticGroup. AI & ML enthusiast. Building @LocallyAIApp for iPhone, a private and local AI chat.
Adrien Grondin
@adrgrondin
Sep 22

Another demo of the iPhone 17 Pro’s on-device LLM performance This time with Ling mini 2.0 by @TheInclusionAI, a 16B MoE model with 1.4B active parameters running at ~120tk/s Thanks to @awnihannun for the MLX DWQ 2-bit quants

16
31
393
140
46K
0
0
7
0
792
InclusionAI reposted
gm8xx8's profile picture.
𝚐𝔪𝟾𝚡𝚡𝟾
@gm8xx8
Sep 19

𝘐𝘕𝘊𝘓𝘜𝘚𝘐𝘖𝘕 𝘐𝘚 𝘖𝘕 𝘈 𝘚𝘏𝘐𝘗𝘗𝘐𝘕𝘎 𝘚𝘗𝘙𝘌𝘌 Another addition to the Ring family, Ring-flash-2.0 (100B total, 6.1B active) is a high-performance thinking model built on Ling-flash-2.0-base, tuned with Long-CoT SFT, RLVR, and RLHF, and designed to tackle a core…

gm8xx8's tweet image. 𝘐𝘕𝘊𝘓𝘜𝘚𝘐𝘖𝘕 𝘐𝘚 𝘖𝘕 𝘈 𝘚𝘏𝘐𝘗𝘗𝘐𝘕𝘎 𝘚𝘗𝘙𝘌𝘌 Another addition to the Ring family, Ring-flash-2.0 (100B total, 6.1B active) is a high-performance thinking model built on Ling-flash-2.0-base, tuned with Long-CoT SFT, RLVR, and RLHF, and designed to tackle a core…
gm8xx8's tweet image. 𝘐𝘕𝘊𝘓𝘜𝘚𝘐𝘖𝘕 𝘐𝘚 𝘖𝘕 𝘈 𝘚𝘏𝘐𝘗𝘗𝘐𝘕𝘎 𝘚𝘗𝘙𝘌𝘌 Another addition to the Ring family, Ring-flash-2.0 (100B total, 6.1B active) is a high-performance thinking model built on Ling-flash-2.0-base, tuned with Long-CoT SFT, RLVR, and RLHF, and designed to tackle a core…
gm8xx8's tweet image. 𝘐𝘕𝘊𝘓𝘜𝘚𝘐𝘖𝘕 𝘐𝘚 𝘖𝘕 𝘈 𝘚𝘏𝘐𝘗𝘗𝘐𝘕𝘎 𝘚𝘗𝘙𝘌𝘌 Another addition to the Ring family, Ring-flash-2.0 (100B total, 6.1B active) is a high-performance thinking model built on Ling-flash-2.0-base, tuned with Long-CoT SFT, RLVR, and RLHF, and designed to tackle a core…
gm8xx8's tweet image. 𝘐𝘕𝘊𝘓𝘜𝘚𝘐𝘖𝘕 𝘐𝘚 𝘖𝘕 𝘈 𝘚𝘏𝘐𝘗𝘗𝘐𝘕𝘎 𝘚𝘗𝘙𝘌𝘌 Another addition to the Ring family, Ring-flash-2.0 (100B total, 6.1B active) is a high-performance thinking model built on Ling-flash-2.0-base, tuned with Long-CoT SFT, RLVR, and RLHF, and designed to tackle a core…
gm8xx8's profile picture. 𝐇𝐈𝐆𝐇-𝐒𝐈𝐆𝐍𝐀𝐋, 𝐇𝐈𝐆𝐇-𝐓𝐀𝐒𝐓𝐄 𝐀𝐈 | 𝐎𝐏𝐄𝐍 𝐌𝐎𝐃𝐄𝐋𝐒 & 𝐒𝐘𝐒𝐓𝐄𝐌𝐒
𝚐𝔪𝟾𝚡𝚡𝟾
@gm8xx8
Sep 17

Ling-flash-2.0 (100B total, 6.1B active) is the third MoE model in the Ling 2.0 family, now open-sourced. Trained on 20T+ tokens with supervised fine-tuning and multi-stage reinforcement learning, it adopts a 1/32 activation-ratio architecture with sigmoid routing, MTP, QK-Norm,…

gm8xx8's tweet image. Ling-flash-2.0 (100B total, 6.1B active) is the third MoE model in the Ling 2.0 family, now open-sourced. Trained on 20T+ tokens with supervised fine-tuning and multi-stage reinforcement learning, it adopts a 1/32 activation-ratio architecture with sigmoid routing, MTP, QK-Norm,…
gm8xx8's tweet image. Ling-flash-2.0 (100B total, 6.1B active) is the third MoE model in the Ling 2.0 family, now open-sourced. Trained on 20T+ tokens with supervised fine-tuning and multi-stage reinforcement learning, it adopts a 1/32 activation-ratio architecture with sigmoid routing, MTP, QK-Norm,…
gm8xx8's tweet image. Ling-flash-2.0 (100B total, 6.1B active) is the third MoE model in the Ling 2.0 family, now open-sourced. Trained on 20T+ tokens with supervised fine-tuning and multi-stage reinforcement learning, it adopts a 1/32 activation-ratio architecture with sigmoid routing, MTP, QK-Norm,…
gm8xx8's tweet image. Ling-flash-2.0 (100B total, 6.1B active) is the third MoE model in the Ling 2.0 family, now open-sourced. Trained on 20T+ tokens with supervised fine-tuning and multi-stage reinforcement learning, it adopts a 1/32 activation-ratio architecture with sigmoid routing, MTP, QK-Norm,…
2
12
98
33
27K
1
3
25
7
5K
TheInclusionAI's profile picture.
InclusionAI
@TheInclusionAI
Sep 19

🚀🚀🚀 Ring-flash-2.0 shows a new breakthrough about Long-CoT RL traning on MoE models.

AntLingAGI's profile picture. MoE model series with foundation (Ling), reasoning (Ring) and any-to-any (Ming) from Ant Group’s AGI initiative, @TheInclusionAI. https://t.co/jtwwGFLA1j
Ant Ling
@AntLingAGI
Sep 19

We open-source Ring-flash-2.0 — the thinking version of Ling-flash-2.0. --> SOTA reasoning in math, code, logic & beyond. --> 100B-A6B, 200+ tok/s on 4×H20 GPUs. --> Powered by "icepop"🧊, solving RL instability in MoE LLMs.

AntLingAGI's tweet image. We open-source Ring-flash-2.0 — the thinking version of Ling-flash-2.0. --> SOTA reasoning in math, code, logic & beyond. --> 100B-A6B, 200+ tok/s on 4×H20 GPUs. --> Powered by "icepop"🧊, solving RL instability in MoE LLMs.
AntLingAGI's tweet image. We open-source Ring-flash-2.0 — the thinking version of Ling-flash-2.0. --> SOTA reasoning in math, code, logic & beyond. --> 100B-A6B, 200+ tok/s on 4×H20 GPUs. --> Powered by "icepop"🧊, solving RL instability in MoE LLMs.
21
121
640
158
259K
1
0
8
2
1K
TheInclusionAI's profile picture.
InclusionAI
@TheInclusionAI
Sep 17

Small activation,big performance, significant milestone of MoE LLM.🚀🚀🚀

AntLingAGI's profile picture. MoE model series with foundation (Ling), reasoning (Ring) and any-to-any (Ming) from Ant Group’s AGI initiative, @TheInclusionAI. https://t.co/jtwwGFLA1j
Ant Ling
@AntLingAGI
Sep 17

⚡️Ling-flash-2.0⚡️ is now open source. 100B MoE LLM • only 6.1B active params --> 3x faster than 36B dense (200+ tok/s on H20) --> Beats ~40B dense LLM on complex reasoning --> Powerful coding and frontend development Small activation. Big performance.

AntLingAGI's tweet image. ⚡️Ling-flash-2.0⚡️ is now open source. 100B MoE LLM • only 6.1B active params --> 3x faster than 36B dense (200+ tok/s on H20) --> Beats ~40B dense LLM on complex reasoning --> Powerful coding and frontend development Small activation. Big performance.
AntLingAGI's tweet image. ⚡️Ling-flash-2.0⚡️ is now open source. 100B MoE LLM • only 6.1B active params --> 3x faster than 36B dense (200+ tok/s on H20) --> Beats ~40B dense LLM on complex reasoning --> Powerful coding and frontend development Small activation. Big performance.
AntLingAGI's tweet image. ⚡️Ling-flash-2.0⚡️ is now open source. 100B MoE LLM • only 6.1B active params --> 3x faster than 36B dense (200+ tok/s on H20) --> Beats ~40B dense LLM on complex reasoning --> Powerful coding and frontend development Small activation. Big performance.
8
84
471
240
32K
0
1
5
1
662
InclusionAI reposted
AQ_MedAI's profile picture.
AQ-MedAI
@AQ_MedAI
Sep 17

AQ Med AI team says hi to everyone！👋🏻 We’re on a mission to bring more MedAI breakthroughs to the world. 🦾 We invite all researchers, developers, and Med AI geeks to join us on this journey, transforming cutting-edge research into real-world impact. 🚀💥 🔗 GitHub, HuggingFace,…

AQ_MedAI's tweet image. AQ Med AI team says hi to everyone！👋🏻 We’re on a mission to bring more MedAI breakthroughs to the world. 🦾 We invite all researchers, developers, and Med AI geeks to join us on this journey, transforming cutting-edge research into real-world impact. 🚀💥 🔗 GitHub, HuggingFace,…
2
4
20
6
6K
TheInclusionAI's profile picture.
InclusionAI
@TheInclusionAI
Sep 14

cool

bigeagle_xd's profile picture. 🐰 @mianmoe 和旺崽的🐻爸爸 | opinions are my own
🐻熊狸
@bigeagle_xd
Sep 13

congs to the release! my wife has contributions in the post-training part 🥳

6
8
172
34
19K
0
0
7
0
818
TheInclusionAI's profile picture.
InclusionAI
@TheInclusionAI
Sep 14

More reasoning work will coming soon🚀

gm8xx8's profile picture. 𝐇𝐈𝐆𝐇-𝐒𝐈𝐆𝐍𝐀𝐋, 𝐇𝐈𝐆𝐇-𝐓𝐀𝐒𝐓𝐄 𝐀𝐈 | 𝐎𝐏𝐄𝐍 𝐌𝐎𝐃𝐄𝐋𝐒 & 𝐒𝐘𝐒𝐓𝐄𝐌𝐒
𝚐𝔪𝟾𝚡𝚡𝟾
@gm8xx8
Sep 12

Inclusion AI follows up on Ling-2.0 with a reasoning-oriented release: Ring-mini-2.0 (16.8B total / 1.4B active) - Built on Ling-mini-2.0 - Long-CoT SFT, stable RLVR, RLHF - 128K context, 500 tok/s on a single H20 (dual streaming, routed and shared experts) - Excels in logic,…

gm8xx8's tweet image. Inclusion AI follows up on Ling-2.0 with a reasoning-oriented release: Ring-mini-2.0 (16.8B total / 1.4B active) - Built on Ling-mini-2.0 - Long-CoT SFT, stable RLVR, RLHF - 128K context, 500 tok/s on a single H20 (dual streaming, routed and shared experts) - Excels in logic,…
gm8xx8's tweet image. Inclusion AI follows up on Ling-2.0 with a reasoning-oriented release: Ring-mini-2.0 (16.8B total / 1.4B active) - Built on Ling-mini-2.0 - Long-CoT SFT, stable RLVR, RLHF - 128K context, 500 tok/s on a single H20 (dual streaming, routed and shared experts) - Excels in logic,…
1
9
36
15
14K
0
0
5
0
489
InclusionAI reposted
gm8xx8's profile picture.
𝚐𝔪𝟾𝚡𝚡𝟾
@gm8xx8
Sep 12

Inclusion AI follows up on Ling-2.0 with a reasoning-oriented release: Ring-mini-2.0 (16.8B total / 1.4B active) - Built on Ling-mini-2.0 - Long-CoT SFT, stable RLVR, RLHF - 128K context, 500 tok/s on a single H20 (dual streaming, routed and shared experts) - Excels in logic,…

gm8xx8's tweet image. Inclusion AI follows up on Ling-2.0 with a reasoning-oriented release: Ring-mini-2.0 (16.8B total / 1.4B active) - Built on Ling-mini-2.0 - Long-CoT SFT, stable RLVR, RLHF - 128K context, 500 tok/s on a single H20 (dual streaming, routed and shared experts) - Excels in logic,…
gm8xx8's tweet image. Inclusion AI follows up on Ling-2.0 with a reasoning-oriented release: Ring-mini-2.0 (16.8B total / 1.4B active) - Built on Ling-mini-2.0 - Long-CoT SFT, stable RLVR, RLHF - 128K context, 500 tok/s on a single H20 (dual streaming, routed and shared experts) - Excels in logic,…
gm8xx8's profile picture. 𝐇𝐈𝐆𝐇-𝐒𝐈𝐆𝐍𝐀𝐋, 𝐇𝐈𝐆𝐇-𝐓𝐀𝐒𝐓𝐄 𝐀𝐈 | 𝐎𝐏𝐄𝐍 𝐌𝐎𝐃𝐄𝐋𝐒 & 𝐒𝐘𝐒𝐓𝐄𝐌𝐒
𝚐𝔪𝟾𝚡𝚡𝟾
@gm8xx8
Sep 10

𝐋𝐈𝐍𝐆 𝟐.𝟎 𝐈𝐒 𝐇𝐄𝐑𝐄 Inclusion AI has open-sourced a new family of MoE-based language models that push both performance and efficiency. The first release, Ling-mini-2.0, has 16B total parameters but activates just 1.4B per token (789M non-embedding). Trained on 20T+…

gm8xx8's tweet image. 𝐋𝐈𝐍𝐆 𝟐.𝟎 𝐈𝐒 𝐇𝐄𝐑𝐄 Inclusion AI has open-sourced a new family of MoE-based language models that push both performance and efficiency. The first release, Ling-mini-2.0, has 16B total parameters but activates just 1.4B per token (789M non-embedding). Trained on 20T+…
gm8xx8's tweet image. 𝐋𝐈𝐍𝐆 𝟐.𝟎 𝐈𝐒 𝐇𝐄𝐑𝐄 Inclusion AI has open-sourced a new family of MoE-based language models that push both performance and efficiency. The first release, Ling-mini-2.0, has 16B total parameters but activates just 1.4B per token (789M non-embedding). Trained on 20T+…
gm8xx8's tweet image. 𝐋𝐈𝐍𝐆 𝟐.𝟎 𝐈𝐒 𝐇𝐄𝐑𝐄 Inclusion AI has open-sourced a new family of MoE-based language models that push both performance and efficiency. The first release, Ling-mini-2.0, has 16B total parameters but activates just 1.4B per token (789M non-embedding). Trained on 20T+…
1
0
24
10
6K
1
9
36
15
14K
TheInclusionAI's profile picture.
InclusionAI
@TheInclusionAI
Sep 12

Yes ！

AdinaYakup's profile picture. @huggingface 🤗 | AI Research 🔍 Chinese ML community. opinions are my own.
Adina Yakup
@AdinaYakup
Sep 12

Ring-mini-2.0 🔥 Latest reasoning model by @InclusionAI666 @AntLing20041208 huggingface.co/inclusionAI/Ri… ✨ 16B/1.4B active - MIT license ✨ Trained on 20T tokens of high-quality data ✨ 128K context length ✨ Reasoning with CoT + RL

AdinaYakup's tweet card. inclusionAI/Ring-mini-2.0 · Hugging Face
inclusionAI/Ring-mini-2.0 · Hugging Face
Source: huggingface.co
5
25
182
54
8K
0
0
4
0
368
TheInclusionAI's profile picture.
InclusionAI
@TheInclusionAI
Sep 12

🚀🚀🚀🧠🧠🧠

AntLingAGI's profile picture. MoE model series with foundation (Ling), reasoning (Ring) and any-to-any (Ming) from Ant Group’s AGI initiative, @TheInclusionAI. https://t.co/jtwwGFLA1j
Ant Ling
@AntLingAGI
Sep 11

🔥 Exciting release! We’re open-sourcing **Ring-mini-2.0**, a powerful yet lightweight 16B-A1B thinking model! 💡 Trained with a novel stable RLVR + RLHF strategy to achieve balanced and robust performance across tasks. 🧠 Outperforms similar-scale dense models in logical…

AntLingAGI's tweet image. 🔥 Exciting release! We’re open-sourcing **Ring-mini-2.0**, a powerful yet lightweight 16B-A1B thinking model! 💡 Trained with a novel stable RLVR + RLHF strategy to achieve balanced and robust performance across tasks. 🧠 Outperforms similar-scale dense models in logical…
17
79
538
235
68K
0
0
5
0
271
TheInclusionAI's profile picture.
InclusionAI
@TheInclusionAI
Sep 12

🚀🚀🚀

AntLingAGI's profile picture. MoE model series with foundation (Ling), reasoning (Ring) and any-to-any (Ming) from Ant Group’s AGI initiative, @TheInclusionAI. https://t.co/jtwwGFLA1j
Ant Ling
@AntLingAGI
Sep 9

🚀 Open-sourcing Ling-mini-2.0 — 16B-A1B MoE LLM.💻Trained on 20T+ tokens w/ SFT + RLVR + RLHF. ⚡ 300+ tok/s (7× faster vs dense). 📦 Open source FP8 training + 4 pretrain CKPTs 👉 Ideal starting point for small-size MoE LLM research & application. 🤗huggingface.co/inclusionAI/Li…

AntLingAGI's tweet image. 🚀 Open-sourcing Ling-mini-2.0 — 16B-A1B MoE LLM.💻Trained on 20T+ tokens w/ SFT + RLVR + RLHF. ⚡ 300+ tok/s (7× faster vs dense). 📦 Open source FP8 training + 4 pretrain CKPTs 👉 Ideal starting point for small-size MoE LLM research & application. 🤗huggingface.co/inclusionAI/Li…
18
26
90
30
47K
0
0
3
1
463
TheInclusionAI's profile picture.
InclusionAI
@TheInclusionAI
Sep 1

Inclusion AI's AWorld introduces a comprehensive end-to-end agentic learning method, talk to authors.@gujinjie @zhuangchenyi

_akhaliq's profile picture. AI research paper tweets, ML @Gradio (acq. by @HuggingFace 🤗) dm for promo ,submit papers here: https://t.co/UzmYN5YmrQ
AK
@_akhaliq
Aug 29

AWorld Orchestrating the Training Recipe for Agentic AI

_akhaliq's tweet image. AWorld Orchestrating the Training Recipe for Agentic AI
1
6
35
19
12K
0
0
3
0
717

United States Trends

Something went wrong.


Something went wrong.