mode LLM Training

kto-expert-mode

Kahneman-Tversky Optimization — preference alignment from binary feedback instead of paired comparisons

KindMode

CategoryLLM Training

Installnpx -y github:anubhavg-icpl/vibe add kto-expert-mode

LicenseCC BY-NC-SA 4.0

Axolotl — YAML-driven LLM fine-tuning with LoRA/QLoRA, DPO/GRPO, DeepSpeed, FSDP

Teacher-student LLM distillation — logits, on-policy distillation, context distillation

Weight-Decomposed Low-Rank Adaptation — magnitude + direction split for better LoRA quality

Direct Preference Optimization — preference alignment without an explicit reward model

Evaluate fine-tuned LLMs — domain benchmarks, regression checks, catastrophic forgetting detection

Group Relative Policy Optimization — DeepSeek-R1 style reasoning RL with verifiable rewards

More in LLM Training