verifiedhigh · 0.92updated 2026-10-03

config.json, deepseek-ai/DeepSeek-V3.2-Exp

repo

Fields

sources/arch-hf-config-deepseek-v32

URL

https://huggingface.co/deepseek-ai/DeepSeek-V3.2-Exp/raw/main/config.json

Publisher

Hugging Face

Kind

repo

Accessed

2026-10-03

notes

Identical MLA/MoE hyperparameters to DeepSeek-V3 (61 layers, kv_lora_rank 512, qk_rope_head_dim 64, 256 routed experts + 1 shared, 8 per token), confirming DSA was added on top of the same MLA shape.

Notes

Identical MLA/MoE hyperparameters to DeepSeek-V3 (61 layers, kv_lora_rank 512, qk_rope_head_dim 64, 256 routed experts + 1 shared, 8 per token), confirming DSA was added on top of the same MLA shape.

Cross-references

This record references no other records.