On the Opportunities and Challenges of Prefill-to-Autoregressive Cross-Model Transfer
🤝 Open to Collab
Junrulu
AI & ML interests
NLP & LLM & MLLM
Recent Activity
authored a paper about 4 hours ago
Accelerating Mixture of Block Attention via Dynamic Routing, KV-Centric Execution and Runtime Pipelining submitted a paper about 5 hours ago
Accelerating Mixture of Block Attention via Dynamic Routing, KV-Centric Execution and Runtime Pipelining upvoted a paper about 7 hours ago
Accelerating Mixture of Block Attention via Dynamic Routing, KV-Centric Execution and Runtime PipeliningOrganizations
ContextPilot
Teaching Agents for Proactive Context Management via Fine-grained RL
-
ContextPilot: Teaching Agents for Proactive Context Management via Fine-grained RL
Paper • 2608.28476 • Published • 30 -
tencent/ContextPilot-E4B
Text Generation • 8B • Updated • 2.4k • 9 -
tencent/ContextPilot-8B
Text Generation • 8B • Updated • 854 • 12 -
tencent/ContextPilot-14B
Text Generation • 15B • Updated • 2.74k • 23
Youtu-LLM
Unlocking the Native Agentic Potential for Lightweight Large Language Models
-
Youtu-LLM: Unlocking the Native Agentic Potential for Lightweight Large Language Models
Paper • 2512.24618 • Published • 154 -
tencent/Youtu-LLM-2B
Text Generation • 2B • Updated • 3.49k • 241 -
tencent/Youtu-LLM-2B-Base
Text Generation • 2B • Updated • 1.01k • 43 -
tencent/Youtu-LLM-A7B-Base
Image-Text-to-Text • 118B • Updated • 2.19k • 41
SamPO
Eliminating Biased Length Reliance of Direct Preference Optimization via Down-Sampled KL Divergence
-
Eliminating Biased Length Reliance of Direct Preference Optimization via Down-Sampled KL Divergence
Paper • 2406.10957 • Published • 2 -
jiazhengli/Pythia-2.8B-HH-RLHF-Iterative-SamPO
Text Generation • 3B • Updated • 20 -
jiazhengli/Pythia-2.8B-TLDR-Iterative-SamPO
Text Generation • 3B • Updated • 15 -
Junrulu/Llama-3-8B-Instruct-Iterative-SamPO
Text Generation • 8B • Updated • 115 • 1
ElephantBench
Blind Men and the Elephant: Probing the Epistemic Myopia of LLMs under Long-Tail Divergent Knowledge
SSA
Sparse Sparse Attention by Aligning Full and Sparse Attention Outputs in Feature Space
RoleMRC
A Fine-Grained Composite Benchmark for Role-Playing and Instruction-Following
MemoChat
Tuning LLMs to Use Memos for Consistent Long-Range Open-Domain Conversation
PLatour
On the Opportunities and Challenges of Prefill-to-Autoregressive Cross-Model Transfer
ElephantBench
Blind Men and the Elephant: Probing the Epistemic Myopia of LLMs under Long-Tail Divergent Knowledge
ContextPilot
Teaching Agents for Proactive Context Management via Fine-grained RL
-
ContextPilot: Teaching Agents for Proactive Context Management via Fine-grained RL
Paper • 2608.28476 • Published • 30 -
tencent/ContextPilot-E4B
Text Generation • 8B • Updated • 2.4k • 9 -
tencent/ContextPilot-8B
Text Generation • 8B • Updated • 854 • 12 -
tencent/ContextPilot-14B
Text Generation • 15B • Updated • 2.74k • 23
SSA
Sparse Sparse Attention by Aligning Full and Sparse Attention Outputs in Feature Space
Youtu-LLM
Unlocking the Native Agentic Potential for Lightweight Large Language Models
-
Youtu-LLM: Unlocking the Native Agentic Potential for Lightweight Large Language Models
Paper • 2512.24618 • Published • 154 -
tencent/Youtu-LLM-2B
Text Generation • 2B • Updated • 3.49k • 241 -
tencent/Youtu-LLM-2B-Base
Text Generation • 2B • Updated • 1.01k • 43 -
tencent/Youtu-LLM-A7B-Base
Image-Text-to-Text • 118B • Updated • 2.19k • 41
RoleMRC
A Fine-Grained Composite Benchmark for Role-Playing and Instruction-Following
SamPO
Eliminating Biased Length Reliance of Direct Preference Optimization via Down-Sampled KL Divergence
-
Eliminating Biased Length Reliance of Direct Preference Optimization via Down-Sampled KL Divergence
Paper • 2406.10957 • Published • 2 -
jiazhengli/Pythia-2.8B-HH-RLHF-Iterative-SamPO
Text Generation • 3B • Updated • 20 -
jiazhengli/Pythia-2.8B-TLDR-Iterative-SamPO
Text Generation • 3B • Updated • 15 -
Junrulu/Llama-3-8B-Instruct-Iterative-SamPO
Text Generation • 8B • Updated • 115 • 1
MemoChat
Tuning LLMs to Use Memos for Consistent Long-Range Open-Domain Conversation