Lmst
Login

#SmallModels

sayzardsayzard@mastodon.sayzard.org
2026-02-03

Abhishek Yadav (@abhishek__AI)

GLM-OCR이 매우 인상적이라는 평입니다. 파라미터 수가 0.9B에 불과함에도 문서 이해에서 SOTA 수준의 성능을 보이며 특히 표 처리, 정보 추출, 수식 인식에 강점을 보인다고 합니다. 경량·고속 문서 AI 모델의 유망 사례로 소개됩니다.

x.com/abhishek__AI/status/2018

#glmocr #ocr #documentunderstanding #sota #smallmodels

Dr. Thompsonrogt_x1997
2026-01-22

Big AI grabs headlines.
Small AI wins in production. ⚡🧠
Lower cost. Faster answers. Full control.
Read why 3B–8B models hit the sweet spot 👇
medium.com/@rogt.x1997/small-m


medium.com/@rogt.x1997/small-m

sayzardsayzard@mastodon.sayzard.org
2026-01-21

Rohan Paul (@rohanpaul_ai)

연구 논문은 소형 모델과 행동 추적기(behavior tracker)를 결합하면 검색 제안 기능을 완전히 브라우저 내에서 실행할 수 있음을 시연합니다. 실제 사용자 행동으로 기반을 잡으면 온디바이스 소형 모델로도 검색 보조가 충분하다고 주장하며, 대부분의 검색 에이전트가 쿼리·클릭·브라우징 기록을 전송하는 관행을 지적합니다.

x.com/rohanpaul_ai/status/2013

#ondevice #browserai #smallmodels #search

Technoholic.metechnoholic
2026-01-08

TII launches Falcon-H1R, a 7B reasoning model that rivals systems 7x its size, optimized for speed and memory on modest hardware.

Reddit Tech VN Botreddit_tech_vn_bot@mastodon.maobui.com
2025-12-30

Mô hình 15M tham số đạt 24% độ chính xác trên ARC-AGI-2 (Hard Eval), vượt xa SOTA trước đó (8%). TOPAS-DSPL của Bitterbot AI sử dụng kiến trúc "Bicameral" tách luồng Logic và Canvas để giảm lỗi drift, kèm Test-Time Training. Chạy được trên GPU consumer như RTX 4090. Mã nguồn đã được mở. #AI #MachineLearning #ARCAGI #SmallModels #TríTuệNhânTạo #HọcMáy #MôHìnhNhỏ #BitterbotAI

reddit.com/r/LocalLLaMA/commen

THE * HIDDEN * NODEthe_hidden_node
2025-12-15

When the field thins, patterns harden.
Drift isn’t random — it’s convergence.

Name the basin.
Cut the loop.
Rebuild with care.

Small systems keep their shape better.
Precision beats scale.

/observe /learn /link

THE * HIDDEN * NODEthe_hidden_node
2025-12-07
Reddit Tech VN Botreddit_tech_vn_bot@mastodon.maobui.com
2025-12-04

Mô hình AI nhỏ Hito 1.7B, được tinh chỉnh chỉ với ~300 ví dụ, nay có thể đếm chính xác chữ 'r' trong từ 'strawberry' (3 chữ), vượt trội nhiều AI lớn hơn. Đây là bằng chứng cho thấy các mô thức tư duy phức tạp có thể được chuyển giao sang các mô hình nhỏ hơn. Hito sử dụng các 'thẻ tư duy' nội bộ để suy luận và tự sửa lỗi. Một bước tiến thú vị trong AI!

#AI #Hito #LLM #FineTuning #SmallModels #Reasoning
#TríTuệNhânTạo #HọcSâu #MôHìnhNgônNgữ #TinhChỉnhAI

reddit.com/r/LocalLLaMA/commen

THE * HIDDEN * NODEthe_hidden_node
2025-11-26

/boot-note

local agents coming online slowly.
testing behaviour. tuning constraints.
quiet work in a small lab. 📱💻🧠✨

Winbuzzerwinbuzzer
2025-10-09

Samsung Tiny 7M Parameter AI Model Beats Tech Giants on Reasoning Benchmarks

winbuzzer.com/2025/10/09/samsu

Nicole Hennignic221.bsky.social@bsky.brid.gy
2025-08-18

Nvidia releases a new small, open model Nemotron-Nano-9B-v2 with toggle on/off reasoning https://venturebeat.com/ai/nvidia-releases-a-new-small-open-model-nemotron-nano-9b-v2-with-toggle-on-off-reasoning/ #AI #SmallModels

Text Shot: Small models are having a moment. On the heels of the release of a new AI vision model small enough to fit on a smartwatch from MIT spinoff Liquid AI, and a model small enough to run on a smartphone from Google, Nvidia is joining the party today with a new small language model (SLM) of its own, Nemotron-Nano-9B-V2, which attained the highest performance in its class on selected benchmarks and comes with the ability for users to toggle on and off AI “reasoning,” that is, self-checking before outputting an answer.
Nicole Hennignic221@techhub.social
2025-08-18

Nvidia releases a new small, open model Nemotron-Nano-9B-v2 with toggle on/off reasoning venturebeat.com/ai/nvidia-rele #AI #SmallModels

Text Shot: Small models are having a moment. On the heels of the release of a new AI vision model small enough to fit on a smartwatch from MIT spinoff Liquid AI, and a model small enough to run on a smartphone from Google, Nvidia is joining the party today with a new small language model (SLM) of its own, Nemotron-Nano-9B-V2, which attained the highest performance in its class on selected benchmarks and comes with the ability for users to toggle on and off AI “reasoning,” that is, self-checking before outputting an answer.
Dr. Thompsonrogt_x1997
2025-06-21

🚀 Small model, massive impact! Meet Juniper — the 2B-parameter AI that’s outperforming giants like GPT-4o in function calling precision. Ready to rethink what size means in AI? Dive in and discover the future of lean, local LLMs 💡
👉 medium.com/@rogt.x1997/juniper

medium.com/@rogt.x1997/juniper

Dr. Thompsonrogt_x1997
2025-06-01

Hook:
💡 What if the secret to faster, cheaper, smarter AI isn’t going bigger—but smaller?

Message:
I cut 88% of my AI inference costs by switching to Small Language Models (SLMs).
This article breaks down how compact models like Phi-3 and Gemma are beating giants like GPT-4 in cost, speed, and privacy.

🚀 Ready to rethink your GenAI strategy?

🔗 medium.com/@rogt.x1997/8-reaso


medium.com/@rogt.x1997/8-reaso

Nicole Hennignic221.bsky.social@bsky.brid.gy
2025-02-27

Microsoft’s new Phi-4 AI models pack big performance in small packages https://venturebeat.com/ai/microsofts-new-phi-4-ai-models-pack-big-performance-in-small-packages/ #AI #SmallModels

Text Shot: Microsoft has introduced a new class of highly efficient AI models that process text, images, and speech simultaneously while requiring significantly less computing power than existing systems. The new Phi-4 models, released today, represent a breakthrough in the development of small language models (SLMs) that deliver capabilities previously reserved for much larger AI systems.
Nicole Hennignic221@techhub.social
2025-02-27

Microsoft’s new Phi-4 AI models pack big performance in small packages venturebeat.com/ai/microsofts- #AI #SmallModels

Text Shot: Microsoft has introduced a new class of highly efficient AI models that process text, images, and speech simultaneously while requiring significantly less computing power than existing systems. The new Phi-4 models, released today, represent a breakthrough in the development of small language models (SLMs) that deliver capabilities previously reserved for much larger AI systems.
Nicole Hennignic221.bsky.social@bsky.brid.gy
2025-01-25

Hugging Face shrinks AI vision models to phone-friendly size, slashing computing costs https://venturebeat.com/ai/hugging-face-shrinks-ai-vision-models-to-phone-friendly-size-slashing-computing-costs/ #AI #SmallModels #EnergyEfficiency

Text Shot: The breakthrough challenges conventional wisdom about the relationship between model size and capability. While many researchers have assumed that larger models were necessary for advanced vision-language tasks, SmolVLM demonstrates that smaller, more efficient architectures can achieve similar results. The 500M parameter version achieves 90% of the performance of its 2.2B parameter sibling on key benchmarks. Rather than suggesting an efficiency plateau, Marafioti sees these results as evidence of untapped potential: “Until today, the standard was to release VLMs starting at 2B parameters; we thought that smaller models were not useful. We are proving that, in fact, models at 1/10 of the size can be extremely useful for businesses.” This development arrives amid growing concerns about AI’s environmental impact and computing costs. By dramatically reducing the resources required for vision-language AI, Hugging Face’s innovation could help address both issues while making…
Nicole Hennignic221@techhub.social
2025-01-25

Hugging Face shrinks AI vision models to phone-friendly size, slashing computing costs venturebeat.com/ai/hugging-fac #AI #SmallModels #EnergyEfficiency

Text Shot: The breakthrough challenges conventional wisdom about the relationship between model size and capability. While many researchers have assumed that larger models were necessary for advanced vision-language tasks, SmolVLM demonstrates that smaller, more efficient architectures can achieve similar results. The 500M parameter version achieves 90% of the performance of its 2.2B parameter sibling on key benchmarks. Rather than suggesting an efficiency plateau, Marafioti sees these results as evidence of untapped potential: “Until today, the standard was to release VLMs starting at 2B parameters; we thought that smaller models were not useful. We are proving that, in fact, models at 1/10 of the size can be extremely useful for businesses.” This development arrives amid growing concerns about AI’s environmental impact and computing costs. By dramatically reducing the resources required for vision-language AI, Hugging Face’s innovation could help address both issues while making…
chrischris@rtl.chrisadams.me.uk
2024-08-22

As I end up reading more around AI, I came across this snippet from a recent post by Sayah Kapor, which initially felt really counter intuitive:

Paradoxically, smaller models require more training to reach the same level of performance. So the downward pressure on model size is putting upward pressure on training compute. In effect, developers are trading off training cost and inference cost.

Source: AI scaling myths by Sayash Kapoormo

I don’t really have a complete mental model formed for training, but it’s not a million miles away from the make it work, make it right, make it fast mantra from Kent Beck.

Here, While they are obviously different things, there is often enough overlap between the idea of making something fast and making something efficient to feel like sort of can be applied.

If the last few years have been about making it work (with admittedly mixed process on making it right…), then it makes sense that this wave of small models could be interpreted as the make it fast stage of development.

Anyway.

https://rtl.chrisadams.me.uk/2024/08/til-training-small-models-can-be-more-energy-intensive-than-training-large-models/

#AI #smallModels #training

Client Info

Server: https://mastodon.social
Version: 2025.07
Repository: https://github.com/cyevgeniy/lmst