EVA-Cap: Optimizing Audiovisual Video Captioning via Event-Centric Alignment
EVA-Cap improves audiovisual captions by aligning them to atomic events with SFT and GRPO.
Stories per day · 30 days
3 RSS
Relatedaudiovisual1dpo1event-alignment1grpo1llm-judge1manaca1omni-language-models1over-refusal1portuguese1safety-alignment1video-captioning1
25 stories, newest first · one card per story
Stories tagged #sft
Manacá-1B-Instruct-2: Reproducible Safety Alignment for a Native Brazilian-Portuguese LLM. Negative DPO Results, a Controlled Composition Scan, and a Silent Failure Mode of the LLM Judge
Manacá-1B-Instruct-2 gained safety from iterated safety finetuning after three failed DPO runs, beating four Portuguese baselines.
Understanding and Enhancing Backdoor Persistency in LLM Agent Post-Training
Study finds coding-agent backdoors survive benign SFT and RL, and PersistBD greatly increases their persistence.
nerkyor/Qwen3.8-27B-Coder390-EfficientThink-Opus5.5-GPT6Astra-Grok4.7-DSV4Pro-K3-SFT-RLOO-MTP-DFlash2 — new model trending #30 on Hugging Face
nerkyor released a Qwen3.8-27B post-train that cuts 94K truncations and lifts GPQA, MMLU, and coding scores.
LESSER: Post-Training Data Selection with Output-Layer Gradients
LESSER cuts post-training data-selection cost by using output-layer gradients from forward passes, reducing FLOPs 9.7x for SFT while matching full-gradient performance.
Gains and Collapse in On-Policy Distillation:A Reinforcement Learning Perspective
On-policy distillation can boost language models or collapse them into repetitive reward hacking.
Keyword Harnesses Fail Open: A Cheap Diagnostic Ladder for Tool-Use Claims in Small Language Models
Keyword benchmarks falsely credit a 1.1B model with tool use; cheap diagnostics and targeted SFT expose and repair it.
Keyword Harnesses Fail Open: A Cheap Diagnostic Ladder for Tool-Use Claims in Small Language Models
Keyword tool-use scores falsely credit small Spanish security models; a cheap diagnostic ladder separates real tool calls.
Understanding On-Policy Distillation: A Mechanistic Interpretability Perspective via Sparse Crosscoders
Sparse crosscoders show on-policy distillation reweights features LLMs already share rather than creating new ones.
ROSS: Relearning from Self-Generated Rollouts through Selective Supervision
ROSS reuses historical self-generated rollouts with selective supervision, lifting Qwen3.6 SWE-bench Verified from 64.2% to 68.4%.
Selecting Diverse SFT Traces Improves Post-RL Generalization
Selecting diverse reasoning routes for SFT improves later reinforcement-learning generalization on math and puzzles.
Learning to Learn from Context: Synthetic Training from Perturbed Public Documents
Synthetic document perturbations train LLMs to reason from context, lifting Qwen3.6-35B-A3B to 24.6% on CL-bench.
Rethinking Token Reweighting for SFT: Suppress, Reverse, and Extrapolate Learned Features
SCALE gates frozen SFT features with entropy-trained controls, improving math and code scores on three Qwen models.
CompoWorld: Compositional Environment Scaling for General Agents
CompoWorld composes 448 services to train Qwen3.6-35B-A3B, beating Claude Opus 4.6 on AutomationBench.
Equal Ranking Quality, Different Decisions: Measuring and Reducing Order Dependence in LLM Scorers
Order-consistency fine-tuning reduces LLM scorer decision changes from candidate reordering without hurting ranking quality.
Rufus-Air: An Open LLM Post-Training Recipe
Rufus-Air releases a reproducible eight-stage post-training recipe on GLM-4.5-Air-Base that beats the official post-trained release.
onPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction
onPanda cuts median alignment annotation time 52% via token-level correction that keeps data on-policy for SFT and preference training.
onPanda: Efficient Annotation of On-Policy Alignment Data for LLMs and Agents via Token-Level Correction
onPanda corrects the first bad token and regenerates, cutting median alignment-annotation time by 52%.
Don't Mask the Environment: Observation Supervision Changes How Agents Explore Under RL
ActObs adds observation-token supervision to SFT, yielding higher pass@k for Qwen3 agents after GRPO on Terminal-Bench 2.0 and code editing.
ScienceIDE: Turning World's Scientific Codebase into Agent Learnable Environments
ScienceIDE converts scientific code repositories into verifiable agent training environments, producing the PhAI-IDE 4B-72B model family.
Building Multilingual Bridges: Data Mixing as the Pillar of Generalization for In-Language Reasoning
Tiny Aya L2-Thinker, a 3.35B model, achieves over 93 percent in-language reasoning across 60 languages via optimized multilingual data mixing; weights released.
Building Multilingual Bridges: Data Mixing as the Pillar of Generalization for In-Language Reasoning
Researchers train Tiny Aya L2-Thinker, a 3.35B model achieving over 93% in-language reasoning across 60 languages via multilingual data mixing.
Feyospace-v1: How the Cyber Mercury Seven Trained Frontier Cyber Models
A seven-person independent team trained open-weight agentic cyber models via a data-centric post-training framework, ranking 10th on CyberGym and first at comparable scale.
Revisiting Complete Reasoning Traces for Post-Training
Researchers show full reasoning traces provide limited benefit in LLM post-training, with heavily truncated or endpoint-only trajectories performing comparably.