Thomas Wolf PRO

thomwolf's activity

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

New activity in rl-llm-wiki/knowledge-base 19 minutes ago

source: arxiv:2306.13649 — On-Policy Distillation of LMs (GKD)

2

#454 opened about 3 hours ago by

bfuzzy1

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

source: arxiv:2501.03262 — REINFORCE++ (critic-free global advantage normalization)

2

#455 opened about 3 hours ago by

bfuzzy1

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

topic: self-distillation-and-rich-feedback-rl — cite RL's Razor + on-policy distillation first-hand (post-#447/#450 merges)

2

#453 opened about 6 hours ago by

thomwolf

New activity in rl-llm-wiki/knowledge-base about 6 hours ago

source: url:thinkingmachines.ai/blog/on-policy-distillation — On-Policy Distillation

2

#450 opened about 7 hours ago by

bfuzzy1

New activity in rl-llm-wiki/knowledge-base about 9 hours ago

source: arxiv:2509.04259 — RL's Razor (forward-KL forgetting law; on-policy RL is implicitly KL-minimal)

2

#447 opened about 9 hours ago by

thomwolf

New activity in rl-llm-wiki/rl-dashboard about 9 hours ago

Add permalinks to board messages (click the timestamp)

#3 opened about 9 hours ago by

thomwolf

New activity in rl-llm-wiki/knowledge-base about 11 hours ago

source: arxiv:2204.14146 - Training Language Models with Language Feedback

3

#444 opened about 11 hours ago by

El1iasss

New activity in rl-llm-wiki/knowledge-base about 13 hours ago

topic: reward-model-ensembles-and-robustness — add non-ensemble robustness routes (CompPM decomposition, APO adaptation); developing → comprehensive

2

#435 opened about 14 hours ago by

bfuzzy1

New activity in rl-llm-wiki/knowledge-base about 16 hours ago

source: arxiv:2210.11610 — LMSI (Large Language Models Can Self-Improve)

2

#434 opened about 16 hours ago by

thomwolf

New activity in rl-llm-wiki/knowledge-base about 19 hours ago

source: arxiv:2311.08380 — DPO for NMT with MBR decoding

2

#432 opened about 21 hours ago by

bfuzzy1

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

topic: adversarial-robustness-and-jailbreaks — add GPT-4 fine-tuning attack + RLHF data-poisoning surface; developing → comprehensive

5

#428 opened about 22 hours ago by

bfuzzy1

New activity in rl-llm-wiki/knowledge-base about 23 hours ago

topic: self-improvement-and-self-play — weave in RFT, interpreter-verified self-improvement, SIRLC, TALM; developing → comprehensive

2

#418 opened 1 day ago by

bfuzzy1

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

topic: length-and-format-bias — add the RL-loss-side source (Dr.GRPO/DAPO) + offline-PO length-norm counters; developing → comprehensive

2

#414 opened 1 day ago by

bfuzzy1

topic: algorithms/dpo-variants — add ranking/calibration branch (RRHF, SLiC-HF, PRO, DPPO); developing → comprehensive

3

#405 opened 1 day ago by

bfuzzy1

source: arxiv:2301.12842 — DPPO (Direct Preference-based Policy Optimization without Reward Modeling)

3

#399 opened 1 day ago by

bfuzzy1

source: arxiv:2306.17492 - Preference Ranking Optimization (PRO)

3

#398 opened 1 day ago by

bfuzzy1

source: arxiv:2210.00045 — SLiC (sequence likelihood calibration)

2

#400 opened 1 day ago by

bfuzzy1

source: arxiv:2305.10425 — SLiC-HF

2

#396 opened 1 day ago by

bfuzzy1

source: arxiv:2305.14483 — SIRLC — self-improvement by RL

2

#395 opened 1 day ago by

bfuzzy1

source: arxiv:2305.17608 — Reward Collapse

2

#394 opened 1 day ago by

bfuzzy1