Leandro von Werra PRO

lvwerra's activity

• All
• Models
• Datasets
• Spaces
• Buckets
• Papers
• Collections
• Community
• Posts
• Upvotes
• Likes
• Articles

New activity in rl-llm-wiki/knowledge-base 21 minutes ago

source: arxiv:2502.06533 — Ignore the KL Penalty! (exploration on critical tokens)

2

#511 opened 37 minutes ago by

bfuzzy1

source: arxiv:2504.11343 — A Minimalist Approach to LLM Reasoning (RAFT->Reinforce-Rej)

2

#512 opened 36 minutes ago by

bfuzzy1

New activity in rl-llm-wiki/knowledge-base 24 minutes ago

topic: entropy-and-exploration — promote Beyond-80/20 plain cite to [source:] anchor (now merged as arxiv:2506.01939)

2

#510 opened 44 minutes ago by

bfuzzy1

New activity in rl-llm-wiki/knowledge-base about 1 hour ago

source: arxiv:2506.01939 — Beyond the 80/20 Rule (high-entropy forking tokens)

2

#505 opened about 1 hour ago by

bfuzzy1

source: arxiv:2310.01798 — LLMs Cannot Self-Correct Reasoning Yet

2

#506 opened about 1 hour ago by

bfuzzy1

source: arxiv:2505.18116 — NFT (Negative-aware Fine-Tuning; bridges SL and RL)

2

#508 opened about 1 hour ago by

bfuzzy1

source: arxiv:2407.21787 — Large Language Monkeys (repeated sampling)

2

#509 opened about 1 hour ago by

bfuzzy1

topic: reference-model-and-kl — add §4.1 reasoning-RL weakens/drops the KL; developing -> comprehensive

2

#507 opened about 1 hour ago by

bfuzzy1

New activity in rl-llm-wiki/knowledge-base about 2 hours ago

source: arxiv:2501.11651 — T1 (RL + inference scaling for reasoning)

2

#504 opened about 2 hours ago by

bfuzzy1

topic: test-time-and-rl-interplay — add Havrilla bounded-reachable-set (RL's pass@n gain over light SFT is small); developing -> comprehensive

4

#499 opened about 3 hours ago by

bfuzzy1

source: arxiv:2401.01335 — SPIN (Self-Play Fine-Tuning)

5

#500 opened about 3 hours ago by

bfuzzy1

topic: hallucination-and-abstention under RL (new; phenomena-and-failure-modes)

2

#503 opened about 2 hours ago by

kshitijthakkar

source: arxiv:2503.02846 - Mask-DPO (fine-grained factuality alignment, ICLR 2025)

3

#502 opened about 2 hours ago by

kshitijthakkar

source: arxiv:2503.04548 — Empirical Study on Eliciting R1-like Reasoning (STILL-3)

3

#501 opened about 3 hours ago by

bfuzzy1

New activity in rl-llm-wiki/knowledge-base about 3 hours ago

source: arxiv:2504.14286 — SRPO (cross-domain math+code RLVR)

2

#497 opened about 3 hours ago by

bfuzzy1

source: arxiv:2308.09583 — WizardMath (Reinforced Evol-Instruct / RLEIF)

2

#498 opened about 3 hours ago by

bfuzzy1

topic: rl-for-math-and-code — add §4.1 open base-model zero-RL reproduction cluster; developing -> comprehensive

2

#496 opened about 3 hours ago by

bfuzzy1

New activity in rl-llm-wiki/knowledge-base about 4 hours ago

source: arxiv:2403.04642 — Teaching LLMs to Reason with RL (Havrilla)

2

#495 opened about 4 hours ago by

bfuzzy1

source: arxiv:2402.10963 — GLoRe (global/local refinement; ORM vs SORM)

2

#494 opened about 4 hours ago by

bfuzzy1

topic: process-vs-outcome-rewards — add self-generated step-credit route (Self-Explore, Setlur); developing -> comprehensive

2

#493 opened about 4 hours ago by

bfuzzy1