論文標(biāo)題：Direct Preference Optimization: Your Language Model is Secretly a Reward Model論文鏈接...
發(fā)簡信
IP屬地：浙江
-
直接偏好優(yōu)化技術(shù)DPO基礎(chǔ)理論及推導(dǎo)
-
生成式大模型的RLHF技術(shù)（一）：基礎(chǔ)一号坡、概述 大語言模型（LLMs）在預(yù)訓(xùn)練的過程中通常會捕捉數(shù)據(jù)的特征，而這些訓(xùn)練數(shù)據(jù)通常既包含高質(zhì)量的也包含低質(zhì)量的描扯，因此模型有時(shí)會產(chǎn)生不被期望的行為，如編造事實(shí)趟薄，生成有偏見...
-
LoRA：大模型下游任務(wù)的低秩適應(yīng)論文標(biāo)題：LoRA: Low-Rank Adaptation of Large Language Models論文鏈接：https://arxiv.org/abs/2106....
-
思維鏈Prompting促進(jìn)大型語言模型的推理能力論文標(biāo)題：Chain-of-Thought Prompting Elicits Reasoning in Large Language Models論文鏈接：https://...
-
Megatron-LM：Transformer模型專用分布式張量模型并行方法論文標(biāo)題：Megatron-LM: Training Multi-Billion Parameter Language Models Using Model Parallel...
-
思維樹：大模型的復(fù)雜推理技術(shù)論文標(biāo)題：Tree of Thoughts: Deliberate Problem Solving with Large Language Models論文鏈接：https:...
-
LIMA：小規(guī)模監(jiān)督數(shù)據(jù)指令微調(diào)論文標(biāo)題：LIMA: Less Is More for Alignment論文鏈接：https://arxiv.org/abs/2305.11206[https://arxi...
-
語言模型的自洽性思維鏈推理技術(shù)論文標(biāo)題：Self-Consistency Improves Chain of Thought Reasoning in Language Models論文鏈接：https:...
-
GPipe：微批量流水線并行論文標(biāo)題：GPipe: Easy Scaling with Micro-Batch Pipeline Parallelism論文鏈接：https://arxiv.org/ab...