inference/research
verifiedhigh · 0.90updated 2026-10-18

moonshotai/Kimi-K2.6 config.json

repo

Fields

id
sources/hf-kimi-k2-6-config
Publisher
Moonshot AI
Kind
repo
Published
2026-04-14
Accessed
2026-10-18
notes
Fetched 2026-10-18 from the same commit as the card (7eb5002f6aad). Top-level model_type kimi_k25 with a nested text_config of model_type kimi_k2 plus a vision_config. text_config: hidden_size 7168, 61 layers, 64 attention heads, n_routed_experts 384, num_experts_per_tok 8, n_shared_experts 1, moe_intermediate_size 2048, first_k_dense_replace 1, moe_layer_freq 1, vocab 163840, max_position_embeddings 262144. MLA: kv_lora_rank 512, q_lora_rank 1536, qk_nope_head_dim 128, qk_rope_head_dim 64, v_head_dim 128. rope_scaling type yarn, factor 64.0, original_max_position_embeddings 4096, beta_fast 32, beta_slow 1, mscale 1, mscale_all_dim 1, rope_theta 50000. num_nextn_predict_layers 0.

Notes

Fetched 2026-10-18 from the same commit as the card (7eb5002f6aad). Top-level model_type kimi_k25 with a nested text_config of model_type kimi_k2 plus a vision_config. text_config: hidden_size 7168, 61 layers, 64 attention heads, n_routed_experts 384, num_experts_per_tok 8, n_shared_experts 1, moe_intermediate_size 2048, first_k_dense_replace 1, moe_layer_freq 1, vocab 163840, max_position_embeddings 262144. MLA: kv_lora_rank 512, q_lora_rank 1536, qk_nope_head_dim 128, qk_rope_head_dim 64, v_head_dim 128. rope_scaling type yarn, factor 64.0, original_max_position_embeddings 4096, beta_fast 32, beta_slow 1, mscale 1, mscale_all_dim 1, rope_theta 50000. num_nextn_predict_layers 0.

Cross-references

This record references no other records.