@knowledgator

Open-source ML research company focused on information extraction #ExplainableAI #AI #opensource #InformationExtraction #UnstructuredData #NLP

knowledgator.com/
Joined October 2023
  • Tweets718
  • Following54
  • Followers740
106Media
We are releasing the most general and efficient encoder model ⚡ GLiFormer can extract virtually any structured JSON from text or PDFs up to 100× faster than generative LLMs for multi-task information extraction workloads. It also stays fast on CPU. 🧵
9
19
7
167
26,249
We are excited to see biomedical NER fine-tunes of our GLiFormer models. This is one of the most important applications for such models.
OpenMed
@OpenMed_AI15h
Problem lists are full of abbreviations: COPD, CHF, T2DM, OSA. GLiFormer BioNER Disease Base finds all four in one line of a past medical history, and COPD again later in the note, with exact offsets. A few lines of OpenMed code, on CPU.
2
10
355
Small enough to run locally. Strong enough to make useful decisions. GLiClass v4 Edge ranks #1 under 1B parameters on the Decision Index’s full-score leaderboard, ahead of most 2B-class models and some 4B models. Check this 400M-parameter model @huggingface: huggingface.co/knowledgator/…
4
3
1
24
900
If you missed it, GLiClass is now in OpenClaw 🦞 Local decisions on CPU. Fully offline. GLiClass v4 Edge: 435M parameters, competitive with 4B models on public Decision Index 0.3. Route models. Rerank results. Triage messages. Model: huggingface.co/knowledgator/… Docs: docs.openclaw.ai/plugins/onn…
3
2
20
615
🚀 Introducing GLiClass v4: our most general and parameter-efficient model family. The edge model reaches 4B-class accuracy on Decision Index with just 435M parameters, while the 27B version performs better than Jev on many of our benchmarks. The release presents 3 ways to scale decision models: model size, latent reasoning, and graph reasoning.
4
13
3
129
5,326
more replies
Parameter efficiency shows up in the results. Edge leads our published sub-1B Decision Index comparison: • GLiClass v4 Edge: 47.49 • vLLM-SR d3-lite: 36.93 • Crivo-150M: 36.39 A 10.56-point lead over the next model with just 435M parameters.
1
3
245
🛠️ Easily fine-tune GLiClass on your own data, labels, and decision rules. 🤗 Model family: huggingface.co/collections/k… 💻 Code and training scripts: github.com/Knowledgator/GLiC… 📚 Documentation: docs.knowledgator.com/docs/f… 📖 Full blog: blog.knowledgator.com/glicla…

GLiClass-v4 - a knowledgator Collection

The next generation of zero-shot classifiers.

huggingface.co
3
228
Knowledgator retweeted
David Keane
@DavidKeane0808Sep 26
With @knowledgator we can do this at @SCXAICloud - whilst introducing new capabilities ontop of the classifier that can be truly tailored..
TypeSafe AI
@typesafeaiSep 25
You've never routed like this before. @OpenRouter is bringing Jev to all of your LLM calls, so your agentic workflows never have to waste a token again. As always, faster, cheaper, more intelligent. Go build the future.
1
2
4
239
GLiClass (open-source Jev, before Jev existed) has landed in OpenClaw 🔥 Run it locally via ONNX for lightweight agent decisions instead of spending a full LLM call on every small choice. Learn more about GLiClass here: github.com/Knowledgator/GLiC…

GitHub - Knowledgator/GLiClass: Generalist and Lightweight Model for Text Classification

Generalist and Lightweight Model for Text Classification - Knowledgator/GLiClass

github.com
Peter Steinberger 🦞
@steipeteSep 23
Next version of OpenClaw uses a decision model to automatically decide between steer or queue. (Lab feature, we support Jef* and API-compat (e.g. local models like Kef...) and ONNX variants)
4
8
2
57
16,722
Knowledgator retweeted
Alex
@Alex_tra_memorySep 23
we ported @knowledgator’s GLiClass Edge Apps v2 to Core ML on an M5 Pro. 32.7M params, ~1.4 ms per decision on CPU. it played 2048 for 2 minutes and reached the 4096 tile. code: github.com/FluidInference/Fl… models: huggingface.co/FluidInferenc…
Knowledgator
@knowledgatorSep 20
The Jeff model, based on our GLiFormer, should be the most parameter-efficient model (a 500M-parameter model) for its performance score. And we created it right before the Jev release. Right now, imagine if we scale it 😉
4
11
1
102
7,910
Knowledgator retweeted
Srimon@srimon_12Sep 21
who need Jev when a 32m model with 5 min free kaggle training can do the same? GLiClass family from @knowledgator are open source.. Codes and Write up coming soon
1
2
6
616
Knowledgator retweeted
Omar Espejel
@omarespejelSep 21
Everyone is talking about Jev, no one is talking about @knowledgator gliformer Jev is an api, gliformer runs in my GPU Thanks @knowledgator
1
6
426
The Jeff model, based on our GLiFormer, should be the most parameter-efficient model (a 500M-parameter model) for its performance score. And we created it right before the Jev release. Right now, imagine if we scale it 😉
Logan Markewich
@LoganMarkewichSep 20
Jeff has been added to JevBench! Ranked at #9 currently 🚀 github.com/logan-markewich/j… Jeff is a drop-in replacement for jev based on GliFormer from @knowledgator. The only models ranking higher right now are LLM/decoder based models. Jevbench is proving that classifiers and encoders are fairly capable. Encoders are extremely easy to fine-tune and host, which means you could easily improve performance and keep your data private. Shoutout to @airesearch12 for creating JevBench!
6
8
1
66
12,276
Jeff has been added to JevBench! Ranked at #9 currently 🚀 github.com/logan-markewich/j… Jeff is a drop-in replacement for jev based on GliFormer from @knowledgator. The only models ranking higher right now are LLM/decoder based models. Jevbench is proving that classifiers and encoders are fairly capable. Encoders are extremely easy to fine-tune and host, which means you could easily improve performance and keep your data private. Shoutout to @airesearch12 for creating JevBench!
11
33
8
220
69,095
Load more