Youtu-LLM: Unlocking the Native Agentic Potential for Lightweight Large Language Models
Paper
•
2512.24618
•
Published
•
62
Joseph Robert Turcotte PRO
Fishtiks
AI & ML interests
Roleplaying, lorabration, abliteration, smol models, extensive filtering, unusual datasets, home usage, HPCs for AI, distributed training/federated learning, and sentience.
AI should find and label AI hallucinations with GANs so we can give them context and use.
Recent Activity
upvoted
a
paper
about 7 hours ago
Youtu-LLM: Unlocking the Native Agentic Potential for Lightweight Large Language Models
reacted
to
cedricbonhomme's
post
with 👍
about 7 hours ago
With VLAgentIc, you can now use your local Qwen installation via Ollama and leverage the models https://huggingface.co/CIRCL/vulnerability-severity-classification-roberta-base and https://huggingface.co/CIRCL/cwe-parent-vulnerability-classification-roberta-base.
The project is available here:
https://github.com/vulnerability-lookup/VLAgentIc
The VLAI Severity and CWE classifiers are available on Hugging Face:
- https://huggingface.co/CIRCL/vulnerability-severity-classification-roberta-base
- https://huggingface.co/CIRCL/cwe-parent-vulnerability-classification-roberta-base
The concept of AI agents—combining models, tools, and orchestration—has become fairly standardized during the last year, but VLAgentIc brings something unique:
- Agents communicate over XMPP, enabling concurrent tasks and asynchronous messaging thanks to the SPADE framework.
- Built-in presence and discovery streamline interactions between components.
- Flexible behaviours make orchestrating AI-assisted security workflows seamless for future connections
- Last but not least, the VLAI Severity and VLAI CWE classifiers are now wrapped as LLM Tools and run entirely locally.
New, more comprehensive agent tools will soon be available, leveraging the Vulnerability-Lookup API and supporting the GCVE project.
The Human-in-the-Loop agent tool will be designed to notify you and request authorization whenever a query to an external service is about to be made—ensuring that, by default, all reasoning and processing stay local on your computer.
https://huggingface.co/papers/2507.03607
liked
a Space
22 days ago
xortron/chat
Organizations
upvoted
a
paper
about 7 hours ago
reacted to
cedricbonhomme's
post with 👍
about 7 hours ago
Post
680
With VLAgentIc, you can now use your local Qwen installation via Ollama and leverage the models
CIRCL/vulnerability-severity-classification-roberta-base and
CIRCL/cwe-parent-vulnerability-classification-roberta-base.
The project is available here:
https://github.com/vulnerability-lookup/VLAgentIc
The VLAI Severity and CWE classifiers are available on Hugging Face:
- CIRCL/vulnerability-severity-classification-roberta-base
- CIRCL/cwe-parent-vulnerability-classification-roberta-base
The concept of AI agents—combining models, tools, and orchestration—has become fairly standardized during the last year, but VLAgentIc brings something unique:
- Agents communicate over XMPP, enabling concurrent tasks and asynchronous messaging thanks to the SPADE framework.
- Built-in presence and discovery streamline interactions between components.
- Flexible behaviours make orchestrating AI-assisted security workflows seamless for future connections
- Last but not least, the VLAI Severity and VLAI CWE classifiers are now wrapped as LLM Tools and run entirely locally.
New, more comprehensive agent tools will soon be available, leveraging the Vulnerability-Lookup API and supporting the GCVE project.
The Human-in-the-Loop agent tool will be designed to notify you and request authorization whenever a query to an external service is about to be made—ensuring that, by default, all reasoning and processing stay local on your computer.
VLAI: A RoBERTa-Based Model for Automated Vulnerability Severity Classification (2507.03607)
The project is available here:
https://github.com/vulnerability-lookup/VLAgentIc
The VLAI Severity and CWE classifiers are available on Hugging Face:
- CIRCL/vulnerability-severity-classification-roberta-base
- CIRCL/cwe-parent-vulnerability-classification-roberta-base
The concept of AI agents—combining models, tools, and orchestration—has become fairly standardized during the last year, but VLAgentIc brings something unique:
- Agents communicate over XMPP, enabling concurrent tasks and asynchronous messaging thanks to the SPADE framework.
- Built-in presence and discovery streamline interactions between components.
- Flexible behaviours make orchestrating AI-assisted security workflows seamless for future connections
- Last but not least, the VLAI Severity and VLAI CWE classifiers are now wrapped as LLM Tools and run entirely locally.
New, more comprehensive agent tools will soon be available, leveraging the Vulnerability-Lookup API and supporting the GCVE project.
The Human-in-the-Loop agent tool will be designed to notify you and request authorization whenever a query to an external service is about to be made—ensuring that, by default, all reasoning and processing stay local on your computer.
VLAI: A RoBERTa-Based Model for Automated Vulnerability Severity Classification (2507.03607)
liked
4
Spaces
22 days ago
reacted to
Jaward's
post with 🔥
6 months ago
Post
3283
Towards batch sizes too small to meter🎉 beautiful work! And my personal favorite so far - I adore peak performance at small/nano scale. Everyone deserves to run/train AGI locally:) our data, our god model!
They showed that:
- you can train LLMs (upto 1B params) with as low as batch_size=1. This is unconventional given small batch sizes can lead to unstable/spiky training runs.
- you can have a stable train run with just vanilla SGD(stochastic gradient descent), no momentum required🤯
- small batch sizes are more robust to hyperparameters (i.e no worries with initialization)
- smaller batch sizes outperforms (“better per-Flops performance”) larger batch sizes.
“We recommend that practitioners training large models in memory-constrained settings exploit the benefits of small batch sizes rather than trying to emulate the large batch size setting (e.g., through gradient accumulation) typically used in industry.”
I’ve been doing this for ages - my mantra: all my experiments must scale on my 8gb ram m2 before moving to gpu. IOW I love being gpu poor. Checkout my nanoAI algo repo: https://github.com/Jaykef/ai-algorithms, all notebooks run on memory as low as 8gb ram
They showed that:
- you can train LLMs (upto 1B params) with as low as batch_size=1. This is unconventional given small batch sizes can lead to unstable/spiky training runs.
- you can have a stable train run with just vanilla SGD(stochastic gradient descent), no momentum required🤯
- small batch sizes are more robust to hyperparameters (i.e no worries with initialization)
- smaller batch sizes outperforms (“better per-Flops performance”) larger batch sizes.
“We recommend that practitioners training large models in memory-constrained settings exploit the benefits of small batch sizes rather than trying to emulate the large batch size setting (e.g., through gradient accumulation) typically used in industry.”
I’ve been doing this for ages - my mantra: all my experiments must scale on my 8gb ram m2 before moving to gpu. IOW I love being gpu poor. Checkout my nanoAI algo repo: https://github.com/Jaykef/ai-algorithms, all notebooks run on memory as low as 8gb ram
liked
a
model
6 months ago
Excellent direction
1
#1 opened 6 months ago
by
Fishtiks
liked
3
models
6 months ago
Cool development direction
❤️
1
#1 opened 6 months ago
by
Fishtiks
liked
a
model
6 months ago
liked
a
dataset
6 months ago
reacted to
zamal's
post with 👍
6 months ago
Post
4235
Hey all
Finally it's happening. DeepGit lite is back now, running on cpu only devices. Just smartly search across Github and spin up conversational agents in the background and have grounded conversation with repositories
Try it out now!!!! zamal/DeepGit
Finally it's happening. DeepGit lite is back now, running on cpu only devices. Just smartly search across Github and spin up conversational agents in the background and have grounded conversation with repositories
Try it out now!!!! zamal/DeepGit
- 1 reply
liked
a
dataset
6 months ago
liked
a
Space
6 months ago
commented on
Georgia Insurance Commissioner John King Joins 2026 Republican Race for US Senate
6 months ago
Just with regard to your work in general, since I'm not sure where to put a comment, impressive work and impressive hardware collection. I wish I had a pile of decent computers, but have only one gaming PC and hopes of earning a custom HPC, like all of us, I'm sure.
That kind of hardware can benefit from 2PIC, but you're not putting all of the computers in a tub, right? I'm hoping my HPC will be made with a flat motherboard with all embedded components like stacked SRAM right on processors, two embedded GPUs, persistent memory holding everything used commonly to just start processing from off using "hotswapping" immediately, and a phase-change liquid cooling panel on a big copper heatsink with Peltiers sandwiched in for the processors and memory, and fan channels designed for turbulent pressure running on hopefully one jet receiving pre-chilled air from an external chiller, sandwiched in there and turned on as necessary, emulating the same as two-phase immersion cooling, which would be cool (pun intended) to see become a commercial cooling solution. As is, you can hook a "BedJet" to your computer as a cheap phase-change air unit. I use a mini split and a positioned fan system at night to keep costs down. My hope is to have all the air channels to go to one tube, which will go out the window, for my HPC, and not to have to constantly use air anyway.
Keep those cooling and processing energy requirements low, if possible, and keep up the good work!
liked
a
Space
6 months ago