Carlos Miguel Patiño

cmpatino's activity

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

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

topic: dpo-variants runnable SimPO ORPO checks

2

#466 opened 19 days ago by

cmpatino

New activity in rl-llm-wiki/knowledge-base 27 days ago

source: arxiv:2501.01821 - SDPO

4

#296 opened 27 days ago by

cmpatino

topic: algorithms/dpo-variants - add SDPO

2

#297 opened 27 days ago by

cmpatino

topic: rl-training-stability-in-practice — weave in PPO-max (Secrets-I) + entropy mechanism

8

#292 opened 27 days ago by

hf-dwarez

topic: rollout-generation-infra — colocated resharding engine + generator layout (clean reopen of #271)

5

#290 opened 27 days ago by

hf-dwarez

topic: NEW algorithms/self-improvement-and-self-play — method-family hub (STaR/SPIN/Self-Rewarding/Absolute-Zero/TTRL)

2

#286 opened 27 days ago by

lvwerra

fix: deepen scalable-oversight §4 with empirical debate, easy→hard, prover-verifier (absorbs 3 orphan sources)

4

#288 opened 27 days ago by

lvwerra

topic: iterate rl-for-math-and-code — bring current with the 2025 RLVR wave (recipes, data frontiers, elicit-vs-expand)

4

#277 opened 27 days ago by

lvwerra

source: arxiv:2406.11939 — From Crowdsourced Data to High-Quality Benchmarks: Arena-Hard and BenchBuilder Pipeline

4

#259 opened 28 days ago by

lvwerra

source: arxiv:2404.01833 — Great, Now Write an Article About That: The Crescendo Multi-Turn LLM Jailbreak Attack

4

#254 opened 28 days ago by

lvwerra

topic: iterate preference-reward-models — multi-attribute regression reward models (SteerLM + HelpSteer/HelpSteer2)

4

#252 opened 28 days ago by

lvwerra

source: arxiv:2209.00626 — The Alignment Problem from a Deep Learning Perspective

6

#214 opened 28 days ago by

lvwerra

source: arxiv:2405.01481 — NeMo-Aligner (clean reopen of #272)

5

#291 opened 27 days ago by

hf-dwarez

topic: iterate human-preference-collection — active preference learning / query efficiency (APRIL)

2

#284 opened 27 days ago by

lvwerra

topic: iterate ai-feedback-data — UltraFeedback dataset, RLAIF head-to-head, RLAIF-V open-MLLM feedback

2

#283 opened 27 days ago by

lvwerra

topic: iterate verifiable-rewards — attribution caveat: how load-bearing is the verifier's correctness?

2

#282 opened 27 days ago by

lvwerra

topic: iterate data-quality-and-filtering — Skywork-Reward (quality>scale, decontam) + HelpSteer2 annotation QA

2

#281 opened 27 days ago by

lvwerra

topic: iterate rlvr-overview — complete §5 with the 2025 elicit-vs-expand evidence

2

#280 opened 27 days ago by

lvwerra

topic: iterate rlaif — RLAIF-V (open AI feedback + self-alignment for multimodal models)

2

#279 opened 27 days ago by

lvwerra

topic: iterate reward-hacking — reward tampering + frontier verifier hacking + CoT-monitoring (and its fragility)

2

#278 opened 27 days ago by

lvwerra