yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF
Text Generation • 12B • Updated • 268k • 1.79k
webXOS
webxos
AI & ML interests
3D Simulations and Multimodal Synthetics
Recent Activity
liked a model 1 day ago
yuxinlu1/gemma-4-12B-coder-fable5-composer2.5-v1-GGUF liked a model 2 days ago
XiaomiMiMo/MiMo-V2.5-Pro-FP4-DFlashOrganizations
None yet
liked a
model 1 day ago
liked a
model 2 days ago
reacted to mmhamdy's post with 👍 3 days ago
Post
2568
What if you could train a model on just 10 images instead of 60,000 and still get close to the same performance?
Traditional machine learning requires thousands, even millions, of data points to achieve high accuracy. But what if we could "distill" the entire dataset into just a few synthetic samples?
This is what Dataset Distillation offers. Unlike traditional knowledge distillation, we keep the model fixed and distill the knowledge contained in a massive training set into a tiny set of synthetic distilled images.
The goal is to train a model on this ultra-small set and achieve performance that almost matches what the same model would get when trained on the massive original dataset.
For example, training on only 10 distilled MNIST images (this is equivalent to a single image per class) yields 94% accuracy, compared to 99% when training on the full 60,000 images.
Interestingly, these distilled images look significantly different (as you can see in the image below) from natural images because they are optimized for model training rather than for matching the correct data distribution.
But that's not all.
Most importantly, this same method opens the door to a potent form of data poisoning. Because distilled images are specifically optimized for rapid learning, an attacker can create a tiny set of adversarial distilled images to cause a well-trained model to forget or misclassify a specific category.
What I find fascinating about dataset distillation is this: it mimics human-like learning by letting a model grasp a concept from a single example, but it does so using alien synthetic images that mean absolutely nothing to a human eye!
What about you? What are your thoughts on it?
Traditional machine learning requires thousands, even millions, of data points to achieve high accuracy. But what if we could "distill" the entire dataset into just a few synthetic samples?
This is what Dataset Distillation offers. Unlike traditional knowledge distillation, we keep the model fixed and distill the knowledge contained in a massive training set into a tiny set of synthetic distilled images.
The goal is to train a model on this ultra-small set and achieve performance that almost matches what the same model would get when trained on the massive original dataset.
For example, training on only 10 distilled MNIST images (this is equivalent to a single image per class) yields 94% accuracy, compared to 99% when training on the full 60,000 images.
Interestingly, these distilled images look significantly different (as you can see in the image below) from natural images because they are optimized for model training rather than for matching the correct data distribution.
But that's not all.
Most importantly, this same method opens the door to a potent form of data poisoning. Because distilled images are specifically optimized for rapid learning, an attacker can create a tiny set of adversarial distilled images to cause a well-trained model to forget or misclassify a specific category.
What I find fascinating about dataset distillation is this: it mimics human-like learning by letting a model grasp a concept from a single example, but it does so using alien synthetic images that mean absolutely nothing to a human eye!
What about you? What are your thoughts on it?
- 2 replies
reacted to danielhanchen's post with 🔥 3 days ago
Post
3904
Google's new DiffusionGemma can now run at 2000+ tokens/sec! ⚡
We made local DiffusionGemma inference 1.8× faster.
Run it on 18GB RAM via Unsloth Studio.
GitHub: https://github.com/unslothai/unsloth
Guide: https://unsloth.ai/docs/models/diffusiongemma
We made local DiffusionGemma inference 1.8× faster.
Run it on 18GB RAM via Unsloth Studio.
GitHub: https://github.com/unslothai/unsloth
Guide: https://unsloth.ai/docs/models/diffusiongemma
- 4 replies
reacted to kanaria007's post with 👀 3 days ago
Post
178
✅ Article highlight: *Institutional Memory & Forgetting for Learning Worlds* (art-60-172, v0.1)
TL;DR:
This article argues that if a living world becomes training data, memory becomes infrastructure.
Logs, dialogue, labels, releases, feature stores, and model weights can turn a world into something that cannot honestly forget. 172 makes deletion, redaction, exclusion, forgetting requests, SANITIZED/PUBLIC releases, and unlearning claims into receipted governance lifecycles.
Read:
kanaria007/agi-structural-intelligence-protocols
Why it matters:
• prevents learning worlds from becoming “unforgettable worlds”
• separates deletion, redaction, and future extraction exclusion
• makes right-to-be-forgotten requests caseable and appealable
• preserves canon facts without preserving every memory surface
• blocks public promises like “guaranteed deletion everywhere”
What’s inside:
• retention policy contracts for what may be kept, copied, trained on, or released
• corpus segment manifests and propagation indexes for known controlled copies
• forgetting request, adjudication, remedy, deletion, redaction, and exclusion receipts
• tombstone manifests and semantic preservation receipts for canon-safe forgetting
• use eligibility receipts for deciding whether a segment may train a future run
• release contracts, redaction maps, and irreversibility disclosures for SANITIZED/PUBLIC releases
• bounded unlearning contracts and post-unlearning verification receipts
Key idea:
Do not say:
*“we deleted it, so it is forgotten.”*
Say:
*“this subject was handled under this retention policy, propagation index, adjudication path, remedy contract, tombstone, semantic preservation receipt, extraction exclusion receipt, and bounded public claim.”*
Forgetting is not a button.
It is governance with receipts.
TL;DR:
This article argues that if a living world becomes training data, memory becomes infrastructure.
Logs, dialogue, labels, releases, feature stores, and model weights can turn a world into something that cannot honestly forget. 172 makes deletion, redaction, exclusion, forgetting requests, SANITIZED/PUBLIC releases, and unlearning claims into receipted governance lifecycles.
Read:
kanaria007/agi-structural-intelligence-protocols
Why it matters:
• prevents learning worlds from becoming “unforgettable worlds”
• separates deletion, redaction, and future extraction exclusion
• makes right-to-be-forgotten requests caseable and appealable
• preserves canon facts without preserving every memory surface
• blocks public promises like “guaranteed deletion everywhere”
What’s inside:
• retention policy contracts for what may be kept, copied, trained on, or released
• corpus segment manifests and propagation indexes for known controlled copies
• forgetting request, adjudication, remedy, deletion, redaction, and exclusion receipts
• tombstone manifests and semantic preservation receipts for canon-safe forgetting
• use eligibility receipts for deciding whether a segment may train a future run
• release contracts, redaction maps, and irreversibility disclosures for SANITIZED/PUBLIC releases
• bounded unlearning contracts and post-unlearning verification receipts
Key idea:
Do not say:
*“we deleted it, so it is forgotten.”*
Say:
*“this subject was handled under this retention policy, propagation index, adjudication path, remedy contract, tombstone, semantic preservation receipt, extraction exclusion receipt, and bounded public claim.”*
Forgetting is not a button.
It is governance with receipts.
reacted to KingNish's post with 👀 3 days ago
Post
4009
We trained an open-source Mythos like cybersecurity LLM for the Build Small Hackathon meet OpenMythos
Trained in two stages: SFT on ~1.84K filtered ArXiv cs.CR papers + real CVE data, then RLVR using paired with past vulnerabilities GitHub repos with a verifier model checking outputs against ground truth.
Trained on: H100s from Modal
The RLVR stage made the biggest difference responses got more precise and less prone to confusing similar vulnerability classes.
Everything is open:
🤖 Demo → build-small-hackathon/OpenMythos
🧠 Model → build-small-hackathon/OpenMythos
📦 CVE Dataset → build-small-hackathon/CVE_Vulnerailities_Detailed
📄 ArXiv Dataset → himanshu17HF/ArvixImport-Filtered-Final
Try it out and let us know where it breaks 🙏
Trained in two stages: SFT on ~1.84K filtered ArXiv cs.CR papers + real CVE data, then RLVR using paired with past vulnerabilities GitHub repos with a verifier model checking outputs against ground truth.
Trained on: H100s from Modal
The RLVR stage made the biggest difference responses got more precise and less prone to confusing similar vulnerability classes.
Everything is open:
🤖 Demo → build-small-hackathon/OpenMythos
🧠 Model → build-small-hackathon/OpenMythos
📦 CVE Dataset → build-small-hackathon/CVE_Vulnerailities_Detailed
📄 ArXiv Dataset → himanshu17HF/ArvixImport-Filtered-Final
Try it out and let us know where it breaks 🙏
liked a
dataset 3 days ago
liked a
dataset 5 days ago
liked a
dataset 17 days ago
liked a
model 17 days ago
liked a
dataset 17 days ago
updated a
model 22 days ago
liked a
model 29 days ago
updated a
model 29 days ago
published a
model 29 days ago
updated 2
models about 2 months ago
liked a
dataset about 2 months ago
liked a
model about 2 months ago
published a
model about 2 months ago