kshitijthakkar/deepseek-v4-mini-300M-init
Text Generation • 0.3B • Updated • 18
DeepSeek V4 Replicas
Small-scale faithful replicas of the DeepSeek-V4 architecture for ablation and weight-transfer research.
Note Randomly-initialized 317M / 170M-active scaffold; reference architecture for V4 weight-transfer experiments.
kshitijthakkar/deepseek-v4-mini-1B-init
Text Generation • 1B • Updated • 12Note Random-init 1.02B / 0.46B-active DeepSeek-V4 architecture scaffold.
kshitijthakkar/deepseek-v4-mini-3B-init
Text Generation • 3B • Updated • 8Note Random-init 3.17B / 1.10B-active DeepSeek-V4 architecture scaffold.
kshitijthakkar/deepseek-v4-mini-6B-init
Text Generation • 8B • Updated • 13Note Random-init 8.06B / 2.21B-active DeepSeek-V4 architecture scaffold.
kshitijthakkar/deepseek-v4-mini-300M-from-flash
Text Generation • 0.3B • Updated • 14 • 2Note Sliced from V4-Flash. frozen_keys.json lists tensors copied verbatim.
kshitijthakkar/deepseek-v4-mini-1B-from-flash
Text Generation • 1B • Updated • 298Note Partially sliced from deepseek-ai/DeepSeek-V4-Flash; 16 of 27 chosen-layer shards applied. Re-run on Linux GPU for a complete slice.