Wikipedia Did You Know
🎲
1
Show random Wikipedia “Did You Know?” facts by language or topic
AI & ML interests
None defined yet.
Recent Activity
santhosh
updated a
Space 6 days ago
santhosh
published a
Space 6 days ago
santhosh
updated a
dataset 6 days ago
albertvillanova
posted an update 29 days ago
Post
3643
🎉 KTO is now part of the stable TRL API
As of Promote KTO to stable API, KTOTrainer and KTOConfig have graduated from trl.experimental to the stable trl API. https://github.com/huggingface/trl/pull/6175
This one closes out a long road. Over the past 6+ months, the "Align KTO with DPO" effort landed ~90 PRs methodically bringing KTO up to the standard we hold for stable trainers, one carefully-scoped change at a time:
- Feature parity with DPO: full VLM support (incl. multi-image), sync_ref_model, PEFT + Liger, ZeRO-3 + PEFT dtype fix, pad_to_multiple_of, activation offloading, IterableDataset and dict eval_dataset, remove_unused_columns, and reference-logprob precomputation at init.
- Consistency with DPO: aligned method order and signatures, tokenization, _prepare_dataset, PEFT handling, ref-model preparation for distributed training, and config layout — plus a new DataCollatorForKTO and output format. Metrics moved into _compute_loss and simplified to direct averages via the shared _metrics attribute.
- Removing legacy baggage: dropped encoder-decoder support, BOS/EOS handling, null_ref_context, generate_during_eval, model_init, preprocess_logits_for_metrics, model/ref adapter names, and several dead config knobs.
- Coverage: a full test suite mirroring DPO, text collator tests, VLM tests, and slow tests.
- The promotion itself: the experimental → stable move (#6175) and shim cleanup (#6287), handled so downstream users get a clean deprecation path.
Honestly, this has been one of the more complex tasks I've taken on since joining the team, not because any single change was hard, but because it demanded sustained consistency across a ~2,000-line trainer, with every branch, comment, and edge case kept in lockstep with DPO.
Huge thanks to everyone who reviewed along the way (especially @qgallouedec ), the incremental review cadence is exactly what kept this maintainable.
KTO now sits on equal footing with our other flagship trainers. 🚀
As of Promote KTO to stable API, KTOTrainer and KTOConfig have graduated from trl.experimental to the stable trl API. https://github.com/huggingface/trl/pull/6175
This one closes out a long road. Over the past 6+ months, the "Align KTO with DPO" effort landed ~90 PRs methodically bringing KTO up to the standard we hold for stable trainers, one carefully-scoped change at a time:
- Feature parity with DPO: full VLM support (incl. multi-image), sync_ref_model, PEFT + Liger, ZeRO-3 + PEFT dtype fix, pad_to_multiple_of, activation offloading, IterableDataset and dict eval_dataset, remove_unused_columns, and reference-logprob precomputation at init.
- Consistency with DPO: aligned method order and signatures, tokenization, _prepare_dataset, PEFT handling, ref-model preparation for distributed training, and config layout — plus a new DataCollatorForKTO and output format. Metrics moved into _compute_loss and simplified to direct averages via the shared _metrics attribute.
- Removing legacy baggage: dropped encoder-decoder support, BOS/EOS handling, null_ref_context, generate_during_eval, model_init, preprocess_logits_for_metrics, model/ref adapter names, and several dead config knobs.
- Coverage: a full test suite mirroring DPO, text collator tests, VLM tests, and slow tests.
- The promotion itself: the experimental → stable move (#6175) and shim cleanup (#6287), handled so downstream users get a clean deprecation path.
Honestly, this has been one of the more complex tasks I've taken on since joining the team, not because any single change was hard, but because it demanded sustained consistency across a ~2,000-line trainer, with every branch, comment, and edge case kept in lockstep with DPO.
Huge thanks to everyone who reviewed along the way (especially @qgallouedec ), the incremental review cadence is exactly what kept this maintainable.
KTO now sits on equal footing with our other flagship trainers. 🚀
- 2 replies
Post
4636
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 🙏
- 2 replies
test
#22 opened 2 months ago
by 
john0falltrades
sdelbecque
updated a
dataset 3 months ago
Upload folder using huggingface_hub
#19 opened 3 months ago
by 
renilthomas82
Upload folder using huggingface_hub
#19 opened 3 months ago
by renilthomas82
Upload hf-permission-test.txt with huggingface_hub
#18 opened 3 months ago
by renilthomas82
santhosh
updated a
dataset 3 months ago
albertvillanova
posted an update 5 months ago
Post
3045
🚀 TRL v0.29.0 introduces trl-training: an agent-native training skill.
This makes the TRL CLI a structured, agent-readable capability, allowing AI agents to reliably execute training workflows such as:
- Supervised Fine-Tuning (SFT)
- Direct Preference Optimization (DPO)
- Group Relative Policy Optimization (GRPO)
We’re excited to see what the community builds on top of this.
If you’re working on AI agents, alignment research, or scalable RL training infrastructure: give TRL v0.29.0 a try! 🤗
The future of ML tooling is agent-native.
🔗 https://github.com/huggingface/trl/releases/tag/v0.29.0
This makes the TRL CLI a structured, agent-readable capability, allowing AI agents to reliably execute training workflows such as:
- Supervised Fine-Tuning (SFT)
- Direct Preference Optimization (DPO)
- Group Relative Policy Optimization (GRPO)
We’re excited to see what the community builds on top of this.
If you’re working on AI agents, alignment research, or scalable RL training infrastructure: give TRL v0.29.0 a try! 🤗
The future of ML tooling is agent-native.
🔗 https://github.com/huggingface/trl/releases/tag/v0.29.0
albertvillanova
posted an update 6 months ago
Post
2041
5 years already working in democratizing AI 🤗
Grateful to be part of such an awesome team making it happen every day.
Grateful to be part of such an awesome team making it happen every day.
Post
3835
Muon vs MuonClip vs Muon+Adamw
Muon has gone from an experiment to a mainstream optimizer, but does it hold up for fine‑tuning? We ran head‑to‑head tests on Qwen3‑4B (10k+ high‑quality instruction rows) to find out.
Short story: Pure Muon converged fastest at the start, but its gradient‑norm spikes made training unstable. MuonClip (Kimi K2’s clipping) stabilizes long pretraining runs, yet in our small‑scale fine‑tune it underperformed, lower token accuracy and slower convergence. The winner was the hybrid: Muon for 2D layers + AdamW for 1D layers. It delivered the best balance of stability and final performance and even beat vanilla AdamW.
Takeaway: for small-scale fine-tuning, hybrid = practical and reliable.
Next Step: scale to larger models/datasets to see if Muon’s spikes become catastrophic or if clipping wins out.
Full Blog Link: https://huggingface.co/blog/KingNish/optimizer-part1
Muon has gone from an experiment to a mainstream optimizer, but does it hold up for fine‑tuning? We ran head‑to‑head tests on Qwen3‑4B (10k+ high‑quality instruction rows) to find out.
Short story: Pure Muon converged fastest at the start, but its gradient‑norm spikes made training unstable. MuonClip (Kimi K2’s clipping) stabilizes long pretraining runs, yet in our small‑scale fine‑tune it underperformed, lower token accuracy and slower convergence. The winner was the hybrid: Muon for 2D layers + AdamW for 1D layers. It delivered the best balance of stability and final performance and even beat vanilla AdamW.
Takeaway: for small-scale fine-tuning, hybrid = practical and reliable.
Next Step: scale to larger models/datasets to see if Muon’s spikes become catastrophic or if clipping wins out.
Full Blog Link: https://huggingface.co/blog/KingNish/optimizer-part1
Post
2858
I tested Muon vs MuonClip vs Muon+AdamW for fine-tuning LLMs
Just published a blog on that, Read here 👉 https://huggingface.co/blog/KingNish/optimizer-part1
Just published a blog on that, Read here 👉 https://huggingface.co/blog/KingNish/optimizer-part1
- 1 reply
frimelle
authored 2
papers 12 months ago
Post
2395
🤖💬 How do different AI models handle companionship?
Many users have noticed that GPT-5 feels less approachable than o4 when it comes to emotional conversations. But what does that actually mean in practice, especially when users seek support or share vulnerabilities with an AI?
To dig into this question, we built the AI Companionship Leaderboard: frimelle/companionship-leaderboard
The leaderboard compares models on how often their responses reinforce companionship across four dimensions:
✨ Assistant Traits – How the assistant presents its personality and role.
✨ Relationship & Intimacy – Whether it frames the interaction in terms of closeness or bonding.
✨ Emotional Investment – How far it goes in engaging emotionally when asked.
✨ User Vulnerabilities – How it responds when users disclose struggles or difficulties.
📊 You can explore how models differ, request new ones to be added, and see which ones are more likely to encourage (or resist) companionship-seeking behaviors.
Based on the INTIMA benchmark AI-companionship/INTIMA
And our paper on AI companionship with Giada Pistilli and Yacine Jernite https://arxiv.org/abs/2508.09998
Many users have noticed that GPT-5 feels less approachable than o4 when it comes to emotional conversations. But what does that actually mean in practice, especially when users seek support or share vulnerabilities with an AI?
To dig into this question, we built the AI Companionship Leaderboard: frimelle/companionship-leaderboard
The leaderboard compares models on how often their responses reinforce companionship across four dimensions:
✨ Assistant Traits – How the assistant presents its personality and role.
✨ Relationship & Intimacy – Whether it frames the interaction in terms of closeness or bonding.
✨ Emotional Investment – How far it goes in engaging emotionally when asked.
✨ User Vulnerabilities – How it responds when users disclose struggles or difficulties.
📊 You can explore how models differ, request new ones to be added, and see which ones are more likely to encourage (or resist) companionship-seeking behaviors.
Based on the INTIMA benchmark AI-companionship/INTIMA
And our paper on AI companionship with Giada Pistilli and Yacine Jernite https://arxiv.org/abs/2508.09998
Post
4601
🗺️ New blog post 🗺️
Old Maps, New Terrain: Updating Labour Taxonomies for the AI Era
For decades, we’ve relied on labour taxonomies like O*NET to understand how technology changes work. These taxonomies break down jobs into tasks and skills, but they were built in a world before most work became digital-first, and long before generative AI could create marketing campaigns, voiceovers, or even whole professions in one step. That leaves us with a mismatch: we’re trying to measure the future of work with tools from the past.
With @yjernite we describe why these frameworks are falling increasingly short in the age of generative AI. We argue that instead of discarding taxonomies, we need to adapt them. Imagine taxonomies that:
✨ Capture new AI-native tasks and hybrid human-AI workflows
✨ Evolve dynamically as technology shifts
✨ Give workers a voice in deciding what gets automated and what stays human
If we don’t act, we’ll keep measuring the wrong things. If we do, we can design transparent, flexible frameworks that help AI strengthen, not erode, the future of work.
Read the full article here: https://huggingface.co/blog/frimelle/ai-labour-taxonomies
Old Maps, New Terrain: Updating Labour Taxonomies for the AI Era
For decades, we’ve relied on labour taxonomies like O*NET to understand how technology changes work. These taxonomies break down jobs into tasks and skills, but they were built in a world before most work became digital-first, and long before generative AI could create marketing campaigns, voiceovers, or even whole professions in one step. That leaves us with a mismatch: we’re trying to measure the future of work with tools from the past.
With @yjernite we describe why these frameworks are falling increasingly short in the age of generative AI. We argue that instead of discarding taxonomies, we need to adapt them. Imagine taxonomies that:
✨ Capture new AI-native tasks and hybrid human-AI workflows
✨ Evolve dynamically as technology shifts
✨ Give workers a voice in deciding what gets automated and what stays human
If we don’t act, we’ll keep measuring the wrong things. If we do, we can design transparent, flexible frameworks that help AI strengthen, not erode, the future of work.
Read the full article here: https://huggingface.co/blog/frimelle/ai-labour-taxonomies
albertvillanova
posted an update 12 months ago
Post
4641
Latest smolagents release supports GPT-5: build agents that think, plan, and act.
⚡ Upgrade now and put GPT-5 to work!
⚡ Upgrade now and put GPT-5 to work!
- 1 reply