發(fā)簡信
IP屬地：西藏
-
ACL 2025 - 基于片段監(jiān)督偏好優(yōu)化的字幕翻譯時延對齊論文標題：Fine-grained Video Dubbing Duration Alignment with Segment Supervis...
-
直接偏好優(yōu)化技術(shù)DPO基礎(chǔ)理論及推導
論文標題：Direct Preference Optimization: Your Language Model is Secretly a R...
-
自適應(yīng)視圖增強的謠言檢測圖對比學習方法論文標題：Propagation Tree Is Not Deep: Adaptive Graph Contrastive Learning A...
-
生成式大模型的RLHF技術(shù)（一）：基礎(chǔ)一、概述 大語言模型（LLMs）在預訓練的過程中通常會捕捉數(shù)據(jù)的特征，而這些訓練數(shù)據(jù)通常既包含高質(zhì)量的也包含低質(zhì)量的，因此模型有時會產(chǎn)生不被期望...
-
LoRA：大模型下游任務(wù)的低秩適應(yīng)論文標題：LoRA: Low-Rank Adaptation of Large Language Models論文鏈接：https://arxi...
-
Megatron-LM：Transformer模型專用分布式張量模型并行方法論文標題：Megatron-LM: Training Multi-Billion Parameter Language Models Using...
-
思維樹：大模型的復雜推理技術(shù)論文標題：Tree of Thoughts: Deliberate Problem Solving with Large Language Mo...
-
LIMA：小規(guī)模監(jiān)督數(shù)據(jù)指令微調(diào)論文標題：LIMA: Less Is More for Alignment論文鏈接：https://arxiv.org/abs/2305.112...
-
語言模型的自洽性思維鏈推理技術(shù)論文標題：Self-Consistency Improves Chain of Thought Reasoning in Language Mo...