EmbeddingGemma 2: Unified and Efficient Multimodal Representations
Paper โข 2610.37364 โข Published โข 37
AI & ML interests
Reproducible Open-Source Machine Learning ๐๐ป
Recent Activity
View all activity
osansevieroย
authored a
paper 9 days ago
lucifertrjย
posted an update about 1 month ago
Post
147
Two ideas from Andrej Karpathyโs tips on making LLM output easier to digest went into my last video:
> I put ASD-STE100 in the script guidelines so the narration sounds less AI-written
> then turned my project into a launch video with the Brag skill in Claude Code. (with voice)
+ ASD-STE100 also works well for video scripts.
๐ Watch on how to build these launch videos with single prompt: https://www.youtube.com/watch?v=OBX3b71n5EE
> I put ASD-STE100 in the script guidelines so the narration sounds less AI-written
> then turned my project into a launch video with the Brag skill in Claude Code. (with voice)
+ ASD-STE100 also works well for video scripts.
๐ Watch on how to build these launch videos with single prompt: https://www.youtube.com/watch?v=OBX3b71n5EE
lucifertrjย
posted an update about 2 months ago
Post
1,415
You can now automate EDD (eval-driven development) with Coding Harness Agents
> build a baseline LLM-based application
> score every change with judge evals
> keep what improves, reject what regresses
I made a tutorial on what EDD is, how it works, and how to use eval scores across experiments to improve an LLM app. It builds on Jeffrey's (Confident AI) article on EDD and Eugene Yan's write-up on product evals.
> Setup: a baseline RAG app using Qdrant and Gemini that every experiment starts from
> Step 1: a binary-labelled dataset with critiques
> Step 2: aligning the LLM-as-a-judge evaluator with Opik evals
> Step 3: a harness loop that runs each experiment and scores it against the baseline.
Tracing and experiment comparison then show what improved, what regressed, and what to tweak next.
Source code is open source.
Full guide (source code linked in the description): https://www.youtube.com/watch?v=e6akw_fKWPk
> build a baseline LLM-based application
> score every change with judge evals
> keep what improves, reject what regresses
I made a tutorial on what EDD is, how it works, and how to use eval scores across experiments to improve an LLM app. It builds on Jeffrey's (Confident AI) article on EDD and Eugene Yan's write-up on product evals.
> Setup: a baseline RAG app using Qdrant and Gemini that every experiment starts from
> Step 1: a binary-labelled dataset with critiques
> Step 2: aligning the LLM-as-a-judge evaluator with Opik evals
> Step 3: a harness loop that runs each experiment and scores it against the baseline.
Tracing and experiment comparison then show what improved, what regressed, and what to tweak next.
Source code is open source.
Full guide (source code linked in the description): https://www.youtube.com/watch?v=e6akw_fKWPk
- 4 replies
lucifertrjย
posted an update 2 months ago
Post
2,967
Published a guide to TurboQuant quantization: how the algorithm works and what Qdrant adds on top of it.
It also includes a benchmark comparing float32, scalar, binary and TurboQuant across BEIR's SciFact, ArguAna and NFCorpus, measured with recall@10, precision@10 and nDCG@10.
๐ HF article: https://huggingface.co/blog/lucifertrj/turboquant-quantization-explained
It also includes a benchmark comparing float32, scalar, binary and TurboQuant across BEIR's SciFact, ArguAna and NFCorpus, measured with recall@10, precision@10 and nDCG@10.
๐ HF article: https://huggingface.co/blog/lucifertrj/turboquant-quantization-explained
satpalsrย
submitted a
paper to Daily Papers 3 months ago
Post
190
The capture stack for collecting high-fidelity egocentric RGB-D data on commodity mobile devices is now open-source.
https://x.com/fpv_labs/status/2082098617704436058
https://x.com/fpv_labs/status/2082098617704436058
Post
1,950
We just released an end-to-end system that sets a new state of the art in egocentric video understanding, generating fine-grained action labels from raw robot and human videos while outperforming Gemini, GPT, Claude, and other leading models.
https://x.com/fpv_labs/status/2079600323883331880
https://x.com/fpv_labs/status/2079600323883331880
osansevieroย
authored a
paper 4 months ago
osansevieroย
submitted a
paper to Daily Papers 4 months ago
Post
275
One prompt, three answers - which model is from where?
johko/llm-blind-date
I built a little demo where you give three models (Apertus, Llama, Qwen3) the same prompt and in the end you have to guess which is which just based on their answers.
GIve it a try! ;)
johko/llm-blind-date
I built a little demo where you give three models (Apertus, Llama, Qwen3) the same prompt and in the end you have to guess which is which just based on their answers.
GIve it a try! ;)
satpalsrย
submitted a
paper to Daily Papers 6 months ago
satpalsrย
authored a
paper 6 months ago
Post
220
We're open-sourcing our infra with 10M+ frames of dataset!
We're releasing Stera, an open-source infra that turns an off-the-shelf device in your pocket into a high-fidelity multimodal data pipeline. It's built around four layers. Capture โ Process โ Evaluate โ Export.
Stera Capture removes the need for bespoke/gated hardware and runs on an off-the-shelf iPhone. It fuses together synchronized RGB, IMU, Lidar-guided depth, and 6-DoF pose out of the box from ARKit and exports them to a raw MCAP file.
Dataset: fpvlabs/stera-10m
Launch Details: https://x.com/fpv_labs/status/2055262652033908832
We're releasing Stera, an open-source infra that turns an off-the-shelf device in your pocket into a high-fidelity multimodal data pipeline. It's built around four layers. Capture โ Process โ Evaluate โ Export.
Stera Capture removes the need for bespoke/gated hardware and runs on an off-the-shelf iPhone. It fuses together synchronized RGB, IMU, Lidar-guided depth, and 6-DoF pose out of the box from ARKit and exports them to a raw MCAP file.
Dataset: fpvlabs/stera-10m
Launch Details: https://x.com/fpv_labs/status/2055262652033908832
Post
197
OpenAI is hiring for SLAM Engineers!
And open-source shouldn't lag behind.
It's pretty hard and necessary problem required to be solved for bringing generalisable robots in real-world.
We are pushing out first deep down & will be open-sourcing stuff in the next releases. Hope everyone is ready! Cheers to HF & more hugs.
Find us at https://x.com/fpv_labs/status/2042585804162371713
And open-source shouldn't lag behind.
It's pretty hard and necessary problem required to be solved for bringing generalisable robots in real-world.
We are pushing out first deep down & will be open-sourcing stuff in the next releases. Hope everyone is ready! Cheers to HF & more hugs.
Find us at https://x.com/fpv_labs/status/2042585804162371713
awsaf49ย
authored a
paper 11 months ago
awsaf49ย
submitted a
paper to Daily Papers 11 months ago
nouamanetaziย
posted an update about 1 year ago
Post
5,467
After training ๐๐ฆ๐จ๐ฅ๐๐๐ on ๐๐๐ ๐๐๐๐๐ฌ for nearly a month, I've come to realize something most people overlook: ๐ข๐ง๐๐ซ๐๐ฌ๐ญ๐ซ๐ฎ๐๐ญ๐ฎ๐ซ๐ ๐ข๐ฌ ๐ญ๐ก๐ ๐ฆ๐๐ค๐-๐จ๐ซ-๐๐ซ๐๐๐ค ๐๐๐๐ญ๐จ๐ซ ๐ข๐ง ๐๐๐ ๐ญ๐ซ๐๐ข๐ง๐ข๐ง๐ . ๐ฅ
Everyone talks about model architecture and data quality. And yes, those matter immensely. But here's what nobody tells you: when your training run fails at 2 AM because of mysterious ๐๐๐๐ ๐๐ซ๐ซ๐จ๐ซ๐ฌ, or when your expensive GPU cluster is running at ๐๐% ๐๐๐๐ข๐๐ข๐๐ง๐๐ฒ, the problem isn't your model. It's most probably a ๐ฆ๐ข๐ฌ๐ฎ๐ฌ๐ ๐จ๐ ๐ญ๐ก๐ ๐ก๐๐ซ๐๐ฐ๐๐ซ๐. ๐ ๏ธ
Questions that seemed simple but had no clear answers: Why is ๐๐จ๐ ๐ญ๐ซ๐๐ข๐ง๐ข๐ง๐ ๐ฌ๐ฅ๐จ๐ฐ๐๐ซ ๐ญ๐ก๐๐ง ๐๐๐ง๐ฌ๐ ๐ฆ๐จ๐๐๐ฅ๐ฌ? Which ๐๐๐๐ ๐๐ฅ๐๐ ๐ฌ should we actually set? How often should we checkpoint without killing throughput?
That's why we built ๐๐ก๐ ๐๐ฆ๐จ๐ฅ ๐๐ซ๐๐ข๐ง๐ข๐ง๐ ๐๐ฅ๐๐ฒ๐๐จ๐จ๐ค ๐: a complete guide covering everything from model architecture and data curation to the SmolLM3 training marathon, post-training techniques, and crucially, the ๐ข๐ง๐๐ซ๐๐ฌ๐ญ๐ซ๐ฎ๐๐ญ๐ฎ๐ซ๐ ๐ฅ๐๐ฒ๐๐ซ that most teams get wrong.
We validated real vs theoretical bandwidth across the entire stack: ๐๐๐๐ ๐ก๐ข๐ญ๐ญ๐ข๐ง๐ ๐ ๐๐/๐ฌ, ๐๐๐๐ข๐ง๐ค ๐.๐ ๐ซ๐๐๐๐ก๐ข๐ง๐ ๐๐๐ ๐๐/๐ฌ, ๐๐๐๐ ๐๐๐ง๐ ๐๐ญ ๐๐.๐ ๐๐/๐ฌ. Then we ran collective operations across ๐๐๐ ๐๐๐๐ฌ (16 nodes, 8xH100s each) and measured how performance degrades at scale: all-reduce drops from ๐๐๐ ๐๐/๐ฌ on a single node to ๐๐๐-๐๐๐ ๐๐/๐ฌ across 16 nodes.
If you've ever wondered why your training runs are slower than they should be, or you're planning to scale up and want to avoid expensive mistakes, this guide might save you weeks of debugging.
๐๐ก๐ ๐๐ฆ๐จ๐ฅ ๐๐ซ๐๐ข๐ง๐ข๐ง๐ ๐๐ฅ๐๐ฒ๐๐จ๐จ๐ค: https://lnkd.in/e5MKXUHS
Shared with โค๏ธ by the HuggingFace team
Everyone talks about model architecture and data quality. And yes, those matter immensely. But here's what nobody tells you: when your training run fails at 2 AM because of mysterious ๐๐๐๐ ๐๐ซ๐ซ๐จ๐ซ๐ฌ, or when your expensive GPU cluster is running at ๐๐% ๐๐๐๐ข๐๐ข๐๐ง๐๐ฒ, the problem isn't your model. It's most probably a ๐ฆ๐ข๐ฌ๐ฎ๐ฌ๐ ๐จ๐ ๐ญ๐ก๐ ๐ก๐๐ซ๐๐ฐ๐๐ซ๐. ๐ ๏ธ
Questions that seemed simple but had no clear answers: Why is ๐๐จ๐ ๐ญ๐ซ๐๐ข๐ง๐ข๐ง๐ ๐ฌ๐ฅ๐จ๐ฐ๐๐ซ ๐ญ๐ก๐๐ง ๐๐๐ง๐ฌ๐ ๐ฆ๐จ๐๐๐ฅ๐ฌ? Which ๐๐๐๐ ๐๐ฅ๐๐ ๐ฌ should we actually set? How often should we checkpoint without killing throughput?
That's why we built ๐๐ก๐ ๐๐ฆ๐จ๐ฅ ๐๐ซ๐๐ข๐ง๐ข๐ง๐ ๐๐ฅ๐๐ฒ๐๐จ๐จ๐ค ๐: a complete guide covering everything from model architecture and data curation to the SmolLM3 training marathon, post-training techniques, and crucially, the ๐ข๐ง๐๐ซ๐๐ฌ๐ญ๐ซ๐ฎ๐๐ญ๐ฎ๐ซ๐ ๐ฅ๐๐ฒ๐๐ซ that most teams get wrong.
We validated real vs theoretical bandwidth across the entire stack: ๐๐๐๐ ๐ก๐ข๐ญ๐ญ๐ข๐ง๐ ๐ ๐๐/๐ฌ, ๐๐๐๐ข๐ง๐ค ๐.๐ ๐ซ๐๐๐๐ก๐ข๐ง๐ ๐๐๐ ๐๐/๐ฌ, ๐๐๐๐ ๐๐๐ง๐ ๐๐ญ ๐๐.๐ ๐๐/๐ฌ. Then we ran collective operations across ๐๐๐ ๐๐๐๐ฌ (16 nodes, 8xH100s each) and measured how performance degrades at scale: all-reduce drops from ๐๐๐ ๐๐/๐ฌ on a single node to ๐๐๐-๐๐๐ ๐๐/๐ฌ across 16 nodes.
If you've ever wondered why your training runs are slower than they should be, or you're planning to scale up and want to avoid expensive mistakes, this guide might save you weeks of debugging.
๐๐ก๐ ๐๐ฆ๐จ๐ฅ ๐๐ซ๐๐ข๐ง๐ข๐ง๐ ๐๐ฅ๐๐ฒ๐๐จ๐จ๐ค: https://lnkd.in/e5MKXUHS
Shared with โค๏ธ by the HuggingFace team
ariG23498ย
authored a
paper about 1 year ago
Post
16,943
deepseek-ai/DeepSeek-OCR is out! ๐ฅ my take โคต๏ธ
> pretty insane it can parse and re-render charts in HTML
> it uses CLIP and SAM features concatenated, so better grounding
> very efficient per vision tokens/performance ratio
> covers 100 languages
> pretty insane it can parse and re-render charts in HTML
> it uses CLIP and SAM features concatenated, so better grounding
> very efficient per vision tokens/performance ratio
> covers 100 languages
- 5 replies
vumichienย
authored a
paper about 1 year ago