jeongseokoh/llama3.1_8b_sft_SPEED-28-BoS_AttnGate
Updated
SPEED
-
-
jeongseokoh/llama3.1_8b_sft_SPEED-24-BoS_AttnGate
Updated -
jeongseokoh/llama3.1_8b_sft_SPEED-24-BoS_SpecialTokens-10
Updated • 1 -
jeongseokoh/llama3.1_8b_sft_SPEED-16-BoS_SpecialTokens-10
Updated • 1 -
jeongseokoh/llama3.1_8b_sft_SPEED-28-BoS_SpecialTokens-10
Updated -
jeongseokoh/llama3.1_8b_sft_vanilla_SpecialTokens-10
Updated -
jeongseokoh/llama3.1_8b_sft_SPEED-28-BoS
Updated • 3 -
jeongseokoh/llama3.1_8b_sft_SPEED-24-BoS
Updated • 9 -
jeongseokoh/llama3.1_8b_sft_SPEED-20-BoS
Updated • 3 -
jeongseokoh/llama3.1_8b_sft_SPEED-16-BoS
Updated • 2 -
jeongseokoh/llama3.1_8b_sft_SPEED-28
Updated • 1 -
jeongseokoh/llama3.1_8b_sft_SPEED-24
Updated • 2 -
jeongseokoh/llama3.1_8b_sft-vanilla
Updated • 4 -
jeongseokoh/llama3.1_8b_sft_SPEED-16
Updated • 8 -
jeongseokoh/llama3.1_8b_sft_SPEED-20-System
Updated • 3 -
jeongseokoh/llama3.1_8b_sft_SPEED-20
Updated • 2 -
jeongseokoh/llama3.1_8b_sft_SPEED-28-System
Updated • 2 -
jeongseokoh/llama3.1_8b_sft_SPEED-24-System
Updated • 2 -
Shallow Prefill, Deep Decoding: Efficient Long-Context Inference via Layer-Asymmetric KV Visibility
Paper • 2605.06105 • Published • 3 -
jeongseokoh/Qwen3-8B_SPEED-24-BoS
Text Generation • Updated • 1