jeongseokoh/llama3.1_8b_sft_SPEED-28-BoS_AttnGate
Updated • 32
SPEED
-
-
jeongseokoh/llama3.1_8b_sft_SPEED-24-BoS_AttnGate
Updated • 67 -
jeongseokoh/llama3.1_8b_sft_SPEED-24-BoS_SpecialTokens-10
Updated • 33 -
jeongseokoh/llama3.1_8b_sft_SPEED-16-BoS_SpecialTokens-10
Updated • 40 -
jeongseokoh/llama3.1_8b_sft_SPEED-28-BoS_SpecialTokens-10
Updated • 28 -
jeongseokoh/llama3.1_8b_sft_vanilla_SpecialTokens-10
Updated • 2 -
jeongseokoh/llama3.1_8b_sft_SPEED-28-BoS
Updated • 248 -
jeongseokoh/llama3.1_8b_sft_SPEED-24-BoS
Updated • 821 -
jeongseokoh/llama3.1_8b_sft_SPEED-20-BoS
Updated • 121 -
jeongseokoh/llama3.1_8b_sft_SPEED-16-BoS
Updated • 60 -
jeongseokoh/llama3.1_8b_sft_SPEED-28
Updated • 11 -
jeongseokoh/llama3.1_8b_sft_SPEED-24
Updated • 245 -
jeongseokoh/llama3.1_8b_sft-vanilla
Updated • 277 -
jeongseokoh/llama3.1_8b_sft_SPEED-16
Updated • 27 -
jeongseokoh/llama3.1_8b_sft_SPEED-20-System
Updated • 35 -
jeongseokoh/llama3.1_8b_sft_SPEED-20
Updated • 21 -
jeongseokoh/llama3.1_8b_sft_SPEED-28-System
Updated • 31 -
jeongseokoh/llama3.1_8b_sft_SPEED-24-System
Updated • 27 -
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 • 63