Gemma 4 Technical Report
Paper • 2607.02770 • Published • 67
AI & ML interests
None defined yet.
Recent Activity
View all activity
osanseviero
authored a
paper 8 days ago
osanseviero
submitted a
paper to Daily Papers 10 days ago
albertvillanova
posted an update 12 days ago
Post
3475
🎉 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
Shrijanagain
posted an update 18 days ago
Post
178
Welcome Researcher and Developers!
SKT AI Labs, we are pushing the boundaries of AI architecture and research—and today, we are thrilled to open our doors to the global research community!
We warmly welcome researchers, developers, and AI enthusiasts to join us and contribute to our R&D efforts.
🧪 What You Can Explore:
We invite you to experiment with our WMF (Weight Manifold Fusion) technology. You can test this high-dimensional fusion technique on smaller models to gain a deeper understanding of its behavior and token convergence.
---------- CHECK OUT:
SPACE : SKT-NRS/RD
EXPERIMENT : sKT-Ai-Labs/SKT-SURYA-H
DIRECT TO MAIN DISCUSSION : SKT-NRS/RD#1
🤝 Your Feedback Shapes the Future :
If it works: Fantastic! Share your results with us and contribute directly to the core vision of SKT AI Labs.
If it doesn't work: No problem at all! Your critical feedback is just as valuable to us. Every experiment and anomaly helps us refine this architecture to make it more stable and robust.
We firmly believe that true innovation stems from community collaboration and transparent testing. Let's build the future of advanced AI together. Your ideas, test results, and feedback are always welcome!
You Can Still Research and Development On WMF Only SKT-SURYA-H Model is Dismissed.
Let's innovate and build together! 💡
SKT AI Labs, we are pushing the boundaries of AI architecture and research—and today, we are thrilled to open our doors to the global research community!
We warmly welcome researchers, developers, and AI enthusiasts to join us and contribute to our R&D efforts.
🧪 What You Can Explore:
We invite you to experiment with our WMF (Weight Manifold Fusion) technology. You can test this high-dimensional fusion technique on smaller models to gain a deeper understanding of its behavior and token convergence.
---------- CHECK OUT:
SPACE : SKT-NRS/RD
EXPERIMENT : sKT-Ai-Labs/SKT-SURYA-H
DIRECT TO MAIN DISCUSSION : SKT-NRS/RD#1
🤝 Your Feedback Shapes the Future :
If it works: Fantastic! Share your results with us and contribute directly to the core vision of SKT AI Labs.
If it doesn't work: No problem at all! Your critical feedback is just as valuable to us. Every experiment and anomaly helps us refine this architecture to make it more stable and robust.
We firmly believe that true innovation stems from community collaboration and transparent testing. Let's build the future of advanced AI together. Your ideas, test results, and feedback are always welcome!
You Can Still Research and Development On WMF Only SKT-SURYA-H Model is Dismissed.
Let's innovate and build together! 💡
bofenghuang
authored 2
papers 19 days ago
Shrijanagain
posted an update 21 days ago
Post
198
🚀 Big News for the AI Community! 🔥
We’re excited to release NRS_QWEN_MYTHOS_1M — a powerful reasoning model built on Qwen 3.5 9B!
At SKT AI LABS, we’ve supercharged this 9B model with our proprietary Neural Reasoning System (NRS) to deliver next-level performance.
🔥 Why This Model is a Game-Changer:
✅ 100x Reasoning Capacity — Exceptional deep logical thinking and complex problem-solving
✅ 1 Million Token Context — Perfect for massive codebases, long documents, and multi-turn agentic workflows
✅ Advanced Thinking Mode — Native <think> tags for true step-by-step Chain-of-Thought reasoning
✅ Tool-Use Ready — Optimized for Python execution, Web Search, and self-correction
✅ Blazing Fast — Runs smoothly on consumer GPUs like RTX 3090/4090
Technical Highlights:
Base: Qwen 3.5 9B
Tuning: NRS-specific high-quality reasoning data
Context: 1M Tokens (YaRN Scaling)
License: NRS DOCS
Whether you’re a developer building coding agents, a researcher working with long-context data, or someone who loves powerful reasoning — this model is built for you.
👉 Try it now on Hugging Face:
SKT-NRS/NRS_QWEN_MYTHOS_1M
Drop a comment: What will you build with it first? 👇
#AI #OpenSource #LLM #Qwen #ReasoningModel #HuggingFace #NewModel #AICommunity
We’re excited to release NRS_QWEN_MYTHOS_1M — a powerful reasoning model built on Qwen 3.5 9B!
At SKT AI LABS, we’ve supercharged this 9B model with our proprietary Neural Reasoning System (NRS) to deliver next-level performance.
🔥 Why This Model is a Game-Changer:
✅ 100x Reasoning Capacity — Exceptional deep logical thinking and complex problem-solving
✅ 1 Million Token Context — Perfect for massive codebases, long documents, and multi-turn agentic workflows
✅ Advanced Thinking Mode — Native <think> tags for true step-by-step Chain-of-Thought reasoning
✅ Tool-Use Ready — Optimized for Python execution, Web Search, and self-correction
✅ Blazing Fast — Runs smoothly on consumer GPUs like RTX 3090/4090
Technical Highlights:
Base: Qwen 3.5 9B
Tuning: NRS-specific high-quality reasoning data
Context: 1M Tokens (YaRN Scaling)
License: NRS DOCS
Whether you’re a developer building coding agents, a researcher working with long-context data, or someone who loves powerful reasoning — this model is built for you.
👉 Try it now on Hugging Face:
SKT-NRS/NRS_QWEN_MYTHOS_1M
Drop a comment: What will you build with it first? 👇
#AI #OpenSource #LLM #Qwen #ReasoningModel #HuggingFace #NewModel #AICommunity
Mallard74
authored a
paper about 2 months ago
misovalko
authored a
paper about 2 months ago
Shrijanagain
posted an update about 2 months ago
Post
2624
We are pleased to announce that the W-IMG Vision Dataset infrastructure is officially live.
The complete asset infrastructure is now accessible on Hugging Face for internal validation and architecture scaling targets.
Dataset Endpoint - sKT-Ai-Labs/W-IMG
#SovereignAI #ComputerVision #MachineLearning #OpenSource
The complete asset infrastructure is now accessible on Hugging Face for internal validation and architecture scaling targets.
Dataset Endpoint - sKT-Ai-Labs/W-IMG
#SovereignAI #ComputerVision #MachineLearning #OpenSource
Post
3184
Releasing my first kernel 🔥 MaxSim
Late-interaction retrieval (ColBERT / PyLate) bottlenecks on materializing the full similarity matrix. This kernel avoids it by using tiled scoring with simdgroup_matrix (Metal) and WMMA.
The result is 3–5× speedup compared to naive PyTorch baseline 🔥
Benchmarks:
- SmallRerank (B=32, C=10): up to 3.2× (M3 Pro) / 2.8× (A100)
- HeavyRerank (B=32, C=100): up to 3.8× (M3 Pro) / 5.3× (A100)
- LongDocStress (Ld=1024): up to 6.2× (L4)
Try it out 👇
https://huggingface.co/kernels/erikkaum/maxsim
Late-interaction retrieval (ColBERT / PyLate) bottlenecks on materializing the full similarity matrix. This kernel avoids it by using tiled scoring with simdgroup_matrix (Metal) and WMMA.
The result is 3–5× speedup compared to naive PyTorch baseline 🔥
Benchmarks:
- SmallRerank (B=32, C=10): up to 3.2× (M3 Pro) / 2.8× (A100)
- HeavyRerank (B=32, C=100): up to 3.8× (M3 Pro) / 5.3× (A100)
- LongDocStress (Ld=1024): up to 6.2× (L4)
Try it out 👇
https://huggingface.co/kernels/erikkaum/maxsim
Post
2827
I'm releasing OpenCS2 a 11TB dataset of around 5000 hours of counter strike gameplay recording.
- HD resolution - 1280×720 · 32 fps
- For each frame keyboard and mouse + world state (player position, velocity, weapon ...)
- HD Stereo audio
- All 10 players perspective
https://huggingface.co/collections/blanchon/opencs2
- HD resolution - 1280×720 · 32 fps
- For each frame keyboard and mouse + world state (player position, velocity, weapon ...)
- HD Stereo audio
- All 10 players perspective
https://huggingface.co/collections/blanchon/opencs2
- 1 reply
qgallouedec
posted an update 2 months ago
Post
10539
Shipped hf-sandbox! 🥡
🧪 Running an eval that executes model-generated C on a few thousand prompts? You probably don't want any of that on your laptop.
Just shipped hf-sandbox, a Modal-style sandbox API on top of Hugging Face Jobs. Spin up an isolated, ephemeral container, run untrusted code, get the result back. No Docker on your laptop, no infra to manage.
Just pip install hf-sandbox.
Early days (v0.1); feedback and issues very welcome:
👉 https://github.com/huggingface/hf-sandbox
🧪 Running an eval that executes model-generated C on a few thousand prompts? You probably don't want any of that on your laptop.
Just shipped hf-sandbox, a Modal-style sandbox API on top of Hugging Face Jobs. Spin up an isolated, ephemeral container, run untrusted code, get the result back. No Docker on your laptop, no infra to manage.
Just pip install hf-sandbox.
Early days (v0.1); feedback and issues very welcome:
👉 https://github.com/huggingface/hf-sandbox
- 1 reply
qgallouedec
posted an update 2 months ago
Post
442
**TRL v1.4 is out 🚀** Chunked NLL loss for SFT and a first-class **OpenReward** integration.
**Chunked NLL loss for SFT — drops peak VRAM by up to 14×**
Standard SFT materializes a full
Peak GPU memory, AdamW fp32:
- Qwen3-14B, 8×H100 FSDP2, 16k seq: 58.9 GB → 38.9 GB
- Qwen3-4B, 1×H100 80GB, 16k seq: OOM → 63.8 GB
- Qwen3-32B, 8×H100 FSDP2, 8k seq: OOM → 71.2 GB
End-to-end it's consistently as fast or faster than
Works with PEFT and VLMs out of the box.
**Open Reward Standard environment adapter**
The new
v1.4 also brings MFU helpers for dense + MoE models, GRPO support for Liger 0.8.0 (delta clipping + VESPO + KL bias correction), Tülu 3's length-normalized DPO loss, four more training chat templates (Cohere, Cohere2, Gemma 3, Qwen3-2507), and a 5+ GB CUDA memory leak fix in activation offloading.
Full release notes: https://github.com/huggingface/trl/releases/tag/v1.4.0
**Chunked NLL loss for SFT — drops peak VRAM by up to 14×**
Standard SFT materializes a full
[batch × seq × vocab] logits tensor before computing cross-entropy, which dominates peak memory at long context lengths. The new loss_type="chunked_nll" path drops ignored-label tokens before the lm_head matmul and computes cross-entropy in checkpointed chunks of 256.Peak GPU memory, AdamW fp32:
- Qwen3-14B, 8×H100 FSDP2, 16k seq: 58.9 GB → 38.9 GB
- Qwen3-4B, 1×H100 80GB, 16k seq: OOM → 63.8 GB
- Qwen3-32B, 8×H100 FSDP2, 8k seq: OOM → 71.2 GB
End-to-end it's consistently as fast or faster than
nll, and unlocks sequence lengths that don't fit at all under the standard path.SFTConfig(loss_type="chunked_nll")Works with PEFT and VLMs out of the box.
**Open Reward Standard environment adapter**
The new
trl.experimental.openreward adapter plugs any environment speaking the [Open Reward Standard](https://openrewardstandard.io) protocol into any TRL trainer that takes an environment_factory. One string — a catalog name or a URL — wires the dataset, factory, and reward_func slots; tools are bound dynamically from JSON Schema, no per-env wrapper code:from trl import GRPOTrainer
from trl.experimental.openreward import OpenRewardSpec
spec = OpenRewardSpec("Eigent/SETA", num_tasks=64)
trainer = GRPOTrainer(
...,
train_dataset=spec.train_dataset,
environment_factory=spec.environment_factory,
reward_funcs=spec.reward_funcs,
)v1.4 also brings MFU helpers for dense + MoE models, GRPO support for Liger 0.8.0 (delta clipping + VESPO + KL bias correction), Tülu 3's length-normalized DPO loss, four more training chat templates (Cohere, Cohere2, Gemma 3, Qwen3-2507), and a 5+ GB CUDA memory leak fix in activation offloading.
Full release notes: https://github.com/huggingface/trl/releases/tag/v1.4.0
Aurelien-Morgan
posted an update 2 months ago
Post
1099
@retrain-pipelines v0.2.0 is out !
I'm at Station F at My booth with GOSIM Paris 2026 today & tomorrow.
Come meet me for a live in-person demo and a chat !
I'm at Station F at My booth with GOSIM Paris 2026 today & tomorrow.
Come meet me for a live in-person demo and a chat !
- 1 reply
Sri-Vigneshwar-DJ
posted an update 3 months ago
Post
166
We ran Feather DB v0.8.0 on LongMemEval (ICLR 2025) — 500 questions across real multi-session conversations, up to 115K tokens each.
**Score: 0.693** · GPT-4o full-context baseline: 0.640
Full 500-question run with Gemini-Flash: **$2.40**
Per-axis breakdown:
→ Info-extraction: **0.942**
→ Knowledge-update: **0.714**
→ Multi-session: **0.606**
→ Temporal: **0.477** ← the hard one, Phase 9 addresses this
Architecture: Hybrid BM25+dense · adaptive temporal decay · embedded (no server) · p50 = 0.19ms · MIT
Raw results + audit JSONs: Hawky-ai/longmemeval-results
We ran Feather DB v0.8.0 on LongMemEval (ICLR 2025) — 500 questions across real multi-session conversations, up to 115K tokens each.
**Score: 0.693** · GPT-4o full-context baseline: 0.640
Full 500-question run with Gemini-Flash: **$2.40**
Per-axis breakdown:
→ Info-extraction: **0.942**
→ Knowledge-update: **0.714**
→ Multi-session: **0.606**
→ Temporal: **0.477** ← the hard one, Phase 9 addresses this
Architecture: Hybrid BM25+dense · adaptive temporal decay · embedded (no server) · p50 = 0.19ms · MIT
pip install feather-dbRaw results + audit JSONs: Hawky-ai/longmemeval-results
qgallouedec
posted an update 3 months ago
Post
8157
TRL v1.3 ships day-one training support for Qwen 3.6 🚀
The new Qwen 3.6 family (
Qwen/Qwen3.6-27B, Qwen/Qwen3.6-35B-A3B) reuses the Qwen3.5-MoE architecture but ships a slightly different chat template, so we updated the stack end-to-end: new training template with {% generation %} markers, tool-call response schema routing, tiny test models for the VLM matrix.SFT with assistant-only loss works out of the box:
from trl import SFTConfig, SFTTrainer
trainer = SFTTrainer(
model="Qwen/Qwen3.6-27B",
args=SFTConfig(assistant_only_loss=True),
train_dataset=dataset,
)
trainer.train()So does GRPO tool-calling — just hand
tools=[...] to GRPOTrainer.v1.3 also brings a new experimental TPO trainer (Triple Preference Optimization), speculative decoding in
trl vllm-serve (Qwen3 MTP / Eagle3 drafts), 12 more KTO ↔ DPO alignment PRs (KTO promotion to stable is now in reach), three more {% generation %} chat templates (Gemma/Gemma 2, Phi-3, GLM-4-MoE), and a chunky SFT entropy bug fix.Full release notes: https://github.com/huggingface/trl/releases/tag/v1.3.0
misovalko
authored 3
papers 3 months ago