Kimi Linear
Kimi Linear is a MoE model (48B total, 3B active) by MoonshotAI using a hybrid linear attention architecture to achieve a 1M token context length. It uses Kimi Delta Attention (KDA), a refined version of Gated DeltaNet that reduces KV cache size by up to 75% and boosts decoding throughput by up to 6x for long contexts.
This guide shows how to fine-tune it with Axolotl with multi-turn conversations and proper masking.
Note: Axolotl uses experimental training code for Kimi Linear as their original modeling code is inference-only.
Getting started
Install Axolotl following the installation guide.
Install CCE via docs
Check that FLA 0.4.1 is installed — the KDA layers import their kernels from it:
pip install fla-core==0.4.1 flash-linear-attention==0.4.1Axolotl already pins both on x86_64, so a standard install covers this. The version matters:
fused_kda_gatechanged signature in 0.4.1, so 0.4.0 raises aTypeErrorat the first KDA layer. FLA is not published for aarch64, which leaves Kimi Linear unsupported there.Run the finetuning example:
axolotl train examples/kimi-linear/kimi-48b-lora.yaml
This config uses about 98.7GiB VRAM.
Let us know how it goes. Happy finetuning!
TIPS
Optimization Guides
See 👉 docs.
Limitations
This is not yet compatible with MoE kernels from transformers v5.