SAF-OPD: Stable Advantage Fusion for On-Policy Distillation
Paper • 2607.29209 • Published • 29
Dipankar Sarkar PRO
dipankarsarkar
AI & ML interests
Building the AI-native stack. Agents as infrastructure, safety as architecture, performance as plumbing. I publish the receipts: papers, datasets, demos.
Recent Activity
reacted to Cameronk199's post with 🔥 about 5 hours ago
35,882 posts. Four years. Every result linked back to a source record.
I built a searchable timeline for the public Truth Social posts associated with Donald Trump's @realDonaldTrump account, covering February 2022 through August 2, 2026.
The explorer separates original posts, quotes, and retruths. Search exact phrases, chart posting volume, or inspect common terms without pulling down the 1.3 GB media archive.
**Explore:** https://huggingface.co/spaces/Cameronk199/truth-social-timeline-explorer
**Dataset:** https://huggingface.co/datasets/Cameronk199/donald-trump-truth-social-posts
```python
from datasets import load_dataset
posts = load_dataset(
"Cameronk199/donald-trump-truth-social-posts",
"full_archive_posts",
split="train",
)
```
The release also contains 8,004 verified image assets and 5,756 video metadata/transcript records. It does not contain trustworthy engagement counts, so the explorer makes no virality claims.
Independent research project; not affiliated with or endorsed by Donald Trump, Truth Social, TMTG, a campaign, or any government office.
#datasets #nlp #opendata #socialmedia reacted to DedeProGames's post with 👍 about 6 hours ago
🚀 Introducing the GRM-3.2 Family
The GRM-3.2 family is a new generation of reasoning-focused models from OrionLLM, purpose-built for long-horizon agentic tasks, extremely difficult reasoning problems, advanced coding, and local AI workflows across a wide range of hardware constraints.
GRM-3.2-Sky is the flagship model in the family: a 35B-A3B Mixture-of-Experts model built on the Ornith-1.0-35B architecture, designed for elite structured reasoning, complex multi-file coding, advanced mathematics, and sustained coherence across extended agentic workflows. It represents a substantial leap in long-horizon task capability over its predecessor, GRM-2.6-Plus.
GRM-3.2-Cliff is the mid-sized workhorse: a 9B-parameter model optimized for long-horizon agentic tasks and difficult reasoning in low-to-mid GPU environments. It delivers strong multi-step planning, debugging, and terminal-agent performance without demanding flagship-level hardware.
GRM-3.2-Turf is the lightweight edge model: a 1.2B-parameter model based on the LiquidAI/LFM2.5-1.2B-Thinking architecture, engineered for efficient on-device execution, high-fidelity instruction following, and robust tool use on mobile, embedded, and other resource-constrained hardware.
All three models are designed for users who need dependable reasoning engines that can maintain goal-directed behavior, planning quality, and task fidelity across many steps—whether on a server, a local workstation, or an edge device.
Models:
GRM-3.2-Sky: https://huggingface.co/OrionLLM/GRM-3.2-Sky
GRM-3.2-Cliff: https://huggingface.co/OrionLLM/GRM-3.2-Cliff
GRM-3.2-Turf: https://huggingface.co/OrionLLM/GRM-3.2-Turf
Organization:
https://huggingface.co/OrionLLM
reacted to Banaxi-Tech's post with 🔥 about 6 hours ago
We did an experiment, we wanted to see if AI is good enough to train models.
We used GPT 5.6 Sol Max for this because its one of the most powerful ones right now.
Our instructions were, it should write the training code, and start the training process and monitor it by itself.
We also gave it a link to BananaMind 2 Mini to get our architecture right.
The result: It worked, it made the working BananaMind 2 Nano, and even beat our previous MiniBananaMind v4 9M.
Its getting way easier to develop your own models now!Organizations
upvoted a paper about 10 hours ago
upvoted 4
papers about 11 hours ago
Wnuan: Staged Post-Training for Question Answering over Proprietary Enterprise Knowledge
Paper • 2608.01862 • Published • 3
DAPD: Dual-Anchored Policy Distillation
Paper • 2608.01735 • Published • 101
Weak-to-Strong On-Policy Distillation
Paper • 2607.26246 • Published • 50
DiffusionGemma Technical Report
Paper • 2608.00146 • Published • 26
upvoted 5
papers about 12 hours ago
In the Driver's Seat: A Multi-Company Study on the Reality of Autonomous Driving System Testing
Paper • 2607.15820 • Published • 4
Seeing or Knowing? Visual Context Sensitivity in Multimodal Large Language Models
Paper • 2607.26326 • Published • 2
DeepVoyager-VL: Incentivizing Vision-in-the-Loop Search for Long-Horizon Multimodal Agents
Paper • 2608.01827 • Published • 15
RecHarness: A Bandit-Routed Agentic Harness for Self-Evolving Recommender Systems
Paper • 2607.29241 • Published • 7
To Add Is Machine, To Delete Is Human: Measuring and Mitigating Deletion Avoidance in LLM Code Editing
Paper • 2607.28887 • Published • 18
upvoted 5
papers about 13 hours ago
Mental World Modeling
Paper • 2607.27201 • Published • 94
EMBL AI Librarian: Life-Sciences Knowledge Layer for AI Agents
Paper • 2607.28229 • Published • 7
Enhancing Rubric-based RL via Self-Distillation
Paper • 2607.18082 • Published • 22
Would You Walk to the Car Wash? Revealing the Salience Bias of Large Language Models in Commonsense Reasoning
Paper • 2607.28478 • Published • 3
Loud or Silent? A Reusable Framework for Per-Modality Failure Analysis in Multimodal Clinical AI
Paper • 2608.01462 • Published • 3
upvoted 5
papers about 14 hours ago
GradCuit: Credit-Assigned Gradient Flow Enables Robust and Interpretable Test-Time Latent Reasoning
Paper • 2608.02585 • Published • 22
GPTQ-2D: Cubic-Time Two-Sided Adaptive Rounding
Paper • 2607.27042 • Published • 18
Zero-Mem: Zero-Token Memory Operations for LLM Agents
Paper • 2607.29377 • Published • 5
Compute Globally, Materialize Locally: The Memory Contract of Sparse Event-KV
Paper • 2607.23693 • Published • 3
Model or Harness? An Interaction-Centric Taxonomy for Localizing Agent Failures
Paper • 2607.28802 • Published • 9