Calibrated distillation consolidates specialists in dynamic MoE models
On Turing-20B-A2B the method scores 0.9988 across four domains and beats Qwen3-8B on math and code.
42
AI model
Qwen3-8B
- Mentions · 7d3
- Mentions · 30d18
- All mentions48
- First seenAug 24, 2026
- Last mentioned1d ago
Timeline
48 articles · newest firstEnvSched schedules agent RL rollouts by environment and task
On Qwen3-8B it scores 48.81 after 40 steps, above random selection's best score after 120 steps.
46
Study traces GEPA merge instability to configuration, not the operator
A nine-configuration sweep across five datasets finds merge can help both gpt-4.1-mini and Qwen3-8B.
30
NgramRL speeds RL rollouts using drafts from recent history
Warm-cache runs commit 3.60–7.77 tokens per forward, and a prefetcher raises Spider throughput 1.38x.
46
J-Comm links LLM agents with hidden and Jacobian latent channels
Qwen3-8B to Qwen3-4B communication scored highest across seven benchmarks and cut generated text tokens by 19.2–74.8% versus TextMAS.
42
Safety mid-training lowers LLM attack success before post-training
Authors report a 72% to 81% relative attack-success drop versus Qwen3 1.7B, 4B, and 8B bases.
48
StopLoss aborts likely-failing coding-agent runs early
On SWE-bench Verified it cut cost per retained solve by 4.5 percent while keeping 99 percent of successful runs.
40
IF-MAP ranks which agents to improve in multi-agent systems
Prompt optimization of selected agents gained 3.81 points versus 1.85 for uniform selection across six workflows.
52
Waypoint-OPD: Transferring RL-Induced Policy Shifts via Gated On-Policy Distillation
Waypoint-OPD gates RL policy-shift distillation and leads math and code scores for Qwen3 students.
42
Hindsight-Privileged RLVR: Single-Rollout Token Credit Without Group Sampling
HP-RLVR gives single-rollout token credit for RLVR, training Qwen3 faster than GRPO with comparable accuracy.
48
DeLoop cuts repeated tool-call loops in long-horizon agents
Under CodeGym compression, Qwen3-8B tasks containing loops fall from 57.3% to 6.1%.
42
ActionOPD: Distilling Decisions, Not Tokens
ActionOPD distills black-box teacher decisions into Qwen3 agents, gaining 3.3-6.9 points out of distribution.
46
The Thinking Leakage in Reinforcement Learning for LLMs
Qwen3 NO-THINK RL gains largely reactivate existing THINK-mode reasoning rather than new skills.
56
DScout speeds diffusion-drafted speculative decoding on Qwen3
Mean speedups versus autoregressive decoding are 3.84 times on Qwen3-4B and 4.14 times on Qwen3-8B.
48
Do LLMs Act on What They Know? From Partner Representations to Cooperative Actions
Across eight LLMs, probes decode Hanabi partner conventions better than the models act on them.
36
Harness Engineering for Software Engineering via Modular Executable Dev-Primitives
HERMES uses modular Dev-Primitives to beat baseline software-engineering agent harnesses by 12.4% on four benchmarks.
54
Conditional Rank Allocation for Taxonomy-Aware Medical Language Model Adaptation
ARBOR's conditional rank allocation lifts Qwen3-8B medical QA accuracy to 69.69% across four benchmarks.
36
NeutronGym: Physics-Graded Neutron Instrument Design for LLM Agents
NeutronGym grades LLM agents on neutron instrument design via McStas ray-tracing, with RL lifting Qwen3-8B from 11% to 77%.
40
Evaluating Biomedical Reranking for LLM-Based Question Answering over Longitudinal Clinical Notes
Biomedical reranking raised clinical-note Hit@10 from 46.6% to 60.6%, while Qwen3-8B correctness reached only 48.6%.
38
PivotOPD: Learning to Recover from Pivotal Mistakes in Multi-Turn Agents
PivotOPD trains agents to avoid and recover from early pivotal mistakes, lifting ALFWorld and SWE-Bench scores.
52
DAGent: Evaluate-then-Grow Planning for Deep Research Agents
DAGent grows research-agent plans from evidence and beats open baselines on BrowseComp-Plus, GAIA, and xbench-DeepSearch.
48
Agent Error Dataset: Scaling 50,000 Error--Diagnosis Pairs for Failure Analysis and Error-Aware Post-Training
A 50,228-pair Agent Error Dataset diagnoses LLM agent failures and improves correction and post-training.
54
PivotOPD: Learning to Recover from Pivotal Mistakes in Multi-Turn Agents
PivotOPD trains multi-turn agents to avoid pivotal mistakes and recover within a few turns.
52
DAGent: Evaluate-then-Grow Planning for Deep Research Agents
DAGent plans deep-research agent graphs incrementally from evidence confidence, outperforming open baselines on three benchmarks.
55
Better Supervision Is Nearby: Neighborhood On-Policy Self-Distillation
Neighborhood OPSD lifts Qwen3 math scores by routing student prefixes to nearby expert teachers.
52
AdviSD: Learning to Advise Frontier LLMs via Targeted Multi-Turn Self-Distillation
AdviSD trains a small advisor with selective self-distillation, beating advisor-GRPO on BFCL-v3 and EnvScaler.
46
Transformers Stop Thinking Too Early, and a Tiny LoRA Fixes It
A rank-8 LoRA lets frozen transformers follow much longer in-context reference chains.
58
Overcoming Scaling Limits in On-Policy Self-Distillation for LLM Reasoning
OASIS keeps on-policy self-distillation effective at scale by supervising verified reasoning trajectories.
55
AdviSD: Learning to Advise Frontier LLMs via Targeted Multi-Turn Self-Distillation
AdviSD trains a Qwen3-8B advisor to steer frozen Gemini and Claude executors via selective self-distillation.
48
PEARL: Adaptive Prefill-Decode Execution with Elasticity for Agentic Reinforcement Learning
PEARL adapts prefill-decode execution and elastic GPUs, raising agentic RL throughput by up to 2.79x.
51
Same Bytes, Different Authority: Reserved-Token Representations in Chat-Template Prompt Injection
Reserved chat-template tokens give injected instructions far more authority than identical subword text on open-weight models.
68
Draft-KV: Learning Useful Latent Communication Between Language Models
Draft-KV sends a frozen sharer's draft key-value states to a small receiver, lifting MMLU-Redux from 37.45% to 78.04%.
48
Persona Dosing: Calibrated Activation Steering for Graded Trait Control
PersonaDose calibrates activation steering so models express requested persona intensities.
46
Contrastive-LM Releases CLM-8B: An Open System One Model That Scores Agent Actions Up to 9× Faster Than Jev
Contrastive-LM released CLM-8B, an open model that scores agent actions up to 9× faster than Jev.
58
Nokia Open-Sources AnyJev: A Training-Free Layer That Turns Any Open LLM Into a Calibrated Decision Model
Nokia open-sources AnyJev, a training-free library that converts any open LLM into a calibrated decision model for classification tasks.
50
LastOPD: Taming Collapse in Latent On-Policy Distillation
LastOPD briefly aligns only the last-layer latent state when distilling Qwen3 into Qwen3-1.7B, beating token-only OPD on MATH-500.
48
Contrastive-LM/CLM-v0.1-8B — new model trending #30 on Hugging Face
Contrastive-LM released CLM-v0.1-8B, a Qwen3-8B contrastive model for fast state-action scoring and agent verification.
48
Don't Mask the Environment: Observation Supervision Changes How Agents Explore Under RL
ActObs supervises observation tokens during SFT, improving GRPO-trained agent pass@k on Terminal-Bench 2.0 and cross-domain code editing for Qwen3 models.
35
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.
52
GAVEL: Graph World Models for Verified and Efficient Long-Horizon LLM Task Planning
GAVEL framework uses graph world models to improve LLM-based robot planning success rates from ~41% to ~92%.
20
Appears with
- Qwen3-4Bai model · 14
- Qwen3-1.7Bai model · 6
- Alibabavendor · 3
- Claudeai model · 3
- Qwen3-32Bai model · 3
- GPT-5ai model · 3
- Qwen3-235Bai model · 2
- vLLMproduct · 2
- Contrastive-LMorganization · 2
- Geminiai model · 2
- Qwen3-30B-A3Bai model · 2
- Qwen3-235B-A22Bai model · 2
- Hugging Facevendor · 2
- NVIDIAvendor · 2
- Llama-3.1-8Bai model · 2
- Nemotron-3.5ai model · 2
Entities are extracted by the model from each article. First seen , last mentioned . Watching keeps it in this browser only (no account); the watchlist page and dashboard alerts use it.