#kimina
LightNews LightNews
shokames.bsky.social
老顧客驚喜包～為Kimina Studio的兩位看版娘畫了服設! 感謝kimina studio的委託!🥰
#kiminastudio #commission
December 15, 2024 at 6:44 AM
jalonso.eurosky.social
Kimina: Interactive mathematical proof assistant. demo.projectnumina.ai #AI #ITP #LeanProver #Math
Kimina Prover Demo – AI-Powered Proof Validation in Action
Try the live demo of Kimina Prover – a new AI-powered assistant for interactive proof validation. Test it now!
demo.projectnumina.ai
May 29, 2025 at 9:46 AM
sungkim.bsky.social
Moonshot AI's Kimina-Prover, a new Lean theorem proving model.

Using an RL pipeline for proof exploration, Kimina-Prover Preview achieved 80.7% on the miniF2F — currently SOTA on this benchmark.
April 14, 2025 at 3:51 PM
yspmina.bsky.social
@𝓴𝓲𝓶𝓲𝓷𝓪 update 📸 #yspgram

Com o vestido de uma pessoa e a maquiagem feita por outra, mas o carisma é todo meu. 🖤

Não percam a oportunidade de virar shots comigo pela primeira vez na vida, não é sempre que a bartender pode beber e ao invés de embebedar os outros. ❄️
January 25, 2026 at 12:07 AM
yspmina.bsky.social
@𝓴𝓲𝓶𝓲𝓷𝓪 update 📸 #yspgram

Quem trabalha e não vai para as festas também pode biscoitar a outfit da noite?

Quando cansarem da concorrência e quiserem um drink feito pelas mãos dessa que voz fala deem uma passadinha no Corner Pub, o primeiro shot é por conta do bolso do meu chefe. 😉🍸🍻
January 17, 2026 at 1:40 AM
yspmina.bsky.social
@𝓴𝓲𝓶𝓲𝓷𝓪 update 📸 #yspgram

Todos estão animados para o baile de sábado, mas e o quintou onde vai ser? Se ainda não tem ideias que tal passar no Corner Pub deixar uma gorjeta para a sua bartender favorita comprar um sapato novo para a festa? 🖤🍸

PS: shots de tequila por conta da casa até às 23h.
January 23, 2026 at 12:09 AM
jalonso.eurosky.social
Kimina-prover preview: Towards large formal reasoning models with reinforcement learning. ~ Numina & Kimi Team. github.com/MoonshotAI/K... #LLMs #LeanProver
GitHub - MoonshotAI/Kimina-Prover-Preview: Technical report of Kimina-Prover Preview.
Technical report of Kimina-Prover Preview. Contribute to MoonshotAI/Kimina-Prover-Preview development by creating an account on GitHub.
github.com
April 15, 2025 at 11:39 AM
yspmina.bsky.social
@𝓴𝓲𝓶𝓲𝓷𝓪 update 📸 #yspgram

É sexta feira, isso significa que estou aqui para convidar vocês para dar uma passadinha no Corner e ter um esquenta antes de perderem a linha na noitada.

Ghost do crush? Vodka. Casa sem luz? Tequila. Síndico te destratou? Gasolina.

Bebidas para todas as situações. 😘🍺
January 31, 2026 at 12:05 AM
sungkim.bsky.social
StepFun-Prover-Preview-7B & 32B

LLMs for formal theorem proving via tool-integrated reasoning! Using RL + environment feedback, they generate Lean 4 proofs efficiently.

- 7B matches DeepSeek-Prover-V2-671B & Kimina-Prover-72B on miniF2F-test (pass@1)!
August 13, 2025 at 1:27 PM
aime-hq.bsky.social
Another open source model from China: Kimina-Prover Preview is "the first large formal reasoning model that can reason in a human-like way and prove mathematical theorems rigorously in the Lean 4 language."

github.com/MoonshotAI/K...
GitHub - MoonshotAI/Kimina-Prover-Preview: Technical report of Kimina-Prover Preview.
Technical report of Kimina-Prover Preview. Contribute to MoonshotAI/Kimina-Prover-Preview development by creating an account on GitHub.
github.com
April 15, 2025 at 5:19 PM
adinayakup.bsky.social
Another powerful reasoning model just dropped on the Hub🔥

🚀 Kimina-Prover Preview: A large formal reasoning model for Lean 4, by Moonshot AI & Numina.

huggingface.co/collections/...
Kimina Prover Preview - a AI-MO Collection
State-of-the-Art Models for Formal Mathematical Reasoning
huggingface.co
April 14, 2025 at 2:09 PM
petsi.bsky.social
Janoon. Ihan Kiminä. 🍻
July 16, 2025 at 1:55 PM
sungkim.bsky.social
Moonshot AI has partnered with Numina to release Kimina-Prover 72B

On miniF2F benchmark, Kimina-Prover achieves a state-of-the-art pass rate of 92.2%.

Blog: huggingface.co/blog/AI-MO/k...
Demo: demo.projectnumina.ai
Models: huggingface.co/collections/...
July 10, 2025 at 11:42 PM
monz-tw.bsky.social
📌
我叫絨絨鼠 öㅅö 也可以稱呼我MZ或是MonZ
是居住在次元裂縫裡的上古絨鼠精， 每天為生活在二次元的客戶們設計服裝、繪製肖像而忙碌著。

畫圖、接委託、開實況、閒聊雜談

人設造型 By 絨絨鼠　3D製作 By KIMINA STUDIO

【Twitch】 twitch.tv/mz_fuwafuwa
【Youtube】 reurl.cc/xEKxyz
October 17, 2024 at 8:20 AM
sungkim.bsky.social
Moonshot AI also releases Kimina Lean Server, which serves the Lean REPL using FastAPI. It supports massive parallelization to verify Lean 4 proofs at scale.

github.com/project-numi...
April 25, 2025 at 4:15 PM
dorialexander.bsky.social
huggingface.co/blog/AI-MO/k...
Kimina-Prover-RL
A Blog post by Project-Numina on Hugging Face
huggingface.co
September 22, 2025 at 11:24 PM
tomiwalamies.bsky.social
Eroaako tämä Sauli Niinistön mediakohtelu Pohjois-Korean meiningistä mitenkään?

Ihan Juche-Kiminä kulkee Toveri Journalistin ja Toveri Kuvaajan kanssa.

www.hs.fi/urheilu/art-...
Urheilu | Sauli Niinistö varmisti kuntosalilla: ”Kuin kuka tahansa”
Sauli Niinistö toimi Topi Juvosen varmistajana penkkipunnerrussarjoissa.
www.hs.fi
March 13, 2025 at 9:51 PM
jalonso.eurosky.social
Kimina-Prover: Applying test-time RL search on large formal reasoning models. ~ Haiming Wang et als. huggingface.co/blog/AI-MO/k... #LLMs #ITP #LeanProver
Kimina-Prover: Applying Test-time RL Search on Large Formal Reasoning Models
A Blog post by Project-Numina on Hugging Face
huggingface.co
July 11, 2025 at 6:14 AM
timkellogg.me
Goedel Prover V2: an 8B that matches a 671B

These are language models that drive a formal proof assistant. It turns out, 671B of world knowledge doesn’t help a whole lot for math

blog.goedel-prover.com
A line chart titled “Performance vs Inference Compute” plots model accuracy against inference passes (log scale on the x-axis: 32 to 8K). The y-axis shows Performance on miniF2F (%), ranging from 65% to 95%. There are six models compared: • Goedel-Prover-V2-32B (Self-Correction Mode) – purple line, consistently highest performance (~90–93.5%), plateaus around 1K passes. • Goedel-Prover-V2-32B – blue line, slightly lower than self-correction mode, around 89–91.5%. • Kimina-Prover-72B – brown line, starts near 83%, climbs to about 87%. • Goedel-Prover-V2-8B – green line, starts ~83.5%, ends ~86.5%. • Deepseek-Prover-V2-67B – orange line, slightly lower than Kimina, ends near 88%. • Deepseek-Prover-V2-7B – red line, lowest throughout, begins at ~75%, ends near 79%. All models show performance gains with increased passes, with diminishing returns past 1K. The legend is located in the bottom left. The chart is cleanly styled with a light purple grid background.
July 16, 2025 at 10:42 PM
ngangou.bsky.social
Mr and Mrs KIMINA-MAKUMBU 👑💖 #truelove #congoleselove
July 16, 2025 at 7:48 PM
bbc6musicbot.bsky.social

🔊 #NowPlaying on #BBC6Music's #AmbientFocus

Kimina:
🎵 Aliamka

#Kimina

▶️ 🪄 Automagic 🔊 show 📻 playlist on Spotify
▶️ Song on #Bandcamp:
Kimina - Aliamka, by Various Artists
from the album INSHA
naug.bandcamp.com
September 21, 2024 at 3:57 AM
ngangou.bsky.social
Mr and Mrs KIMINA-MAKUMBU 👑👑
My lifetime soulmate #reallove #meandyou #forever #together
April 18, 2025 at 1:47 AM
timkellogg.me
Kamina-Prover: a 7B fine-tune of Moonshot Kimi trained to use the Lean theorem prover

This is the first LLM I’ve seen to use theorem provers at inference-time

Look at that performance! Hard to beat that pass@8192 score

github.com/MoonshotAI/K...
A white slide‑style chart titled “Performance Comparison of Large Reasoning Models by Evaluation Type.” A vertical blue axis at the far left is labeled “Performance (%)” with tick marks at 0, 20, 40, 60, 80, and 100. Along the bottom, the x‑axis reads “Evaluation Type.” Three evenly spaced category labels appear: “miniF2F,” “IMO (formalized),” and “AIME(formalized).” For each category there are four narrow bars, grouped side‑by‑side: • Dark‑blue bar (legend: “Kimina‑Prover (pass@8192)”): • miniF2F — 80.7 % • IMO — 40.0 % • AIME — 86.7 % • Light‑blue bar (legend: “Kimina‑Prover (pass@32)”): • miniF2F — 68.8 % • IMO — 20.0 % • AIME — 46.7 % • Beige‑tan bar (legend: “Gemini 2.5 Pro (pass@32)”): • miniF2F — 37.7 % • IMO — 5.0 % • AIME — 13.3 % • Charcoal‑gray bar (legend: “OpenAI o3‑mini (pass@32)”): • miniF2F — 24.6 % • IMO — 0.0 % • AIME — 6.7 % Percentage values are printed in bold above each bar. The legend box, outlined in blue, sits in the upper right quadrant of the plot area. Centered below the x‑axis, a footnote in italic gray text reads: “Note: While Gemini 2.5 and OpenAI o3‑mini can solve all 15 AIME problems using informal reasoning, both models struggle to formalize these solutions, highlighting a gap between informal reasoning capacity and formal reasoning capabilities.”
April 15, 2025 at 10:27 PM
jalonso.eurosky.social
Kimina Lean server: A high-performance Lean server for large-scale verification. ~ Marco Dos Santos et als. openreview.net/forum?id=orP... #ITP #LeanProver #LLMs
Kimina Lean Server: A High-Performance Lean Server for Large-Scale...
Recent progress in neural theorem proving has been driven by the training of large language models on Lean 4 problems via reinforcement learning, a process that requires fast and scalable...
openreview.net
November 29, 2025 at 11:17 AM
daruma.es
«A sus 24 años, Kikuko Kimina está exhausta por la explotación laboral y por la sensación de haber perdido el rumbo de su vida. Una noche, tras enfrentarse a su jefe en el restaurante, conoce a Rita Tojima, la dueña de una pequeña floristería de barrio, [...]
April 17, 2026 at 10:30 AM