Moonshot AIMoonshot AI

Kimi Linear 48B A3B

現行
65.1Kダウンロード562いいねOct 2025公開日1.0M トークンコンテキストMITライセンス76 優秀品質

Kimi Linear 48B A3B (48B parameters) requires approximately 33.2 GB of VRAM with Q4_K_M quantization. For the best balance of quality and speed, we recommend hardware with at least 39 GB of VRAM.

はじめに

— コピー&ペーストでローカル実行

Copy-paste commands to run Kimi Linear 48B A3B on your machine.

Run

docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \ --hf-repo "moonshotai/Kimi-Linear-48B-A3B-Instruct" \ --hf-file "Kimi-Linear-48B-A3B-Instruct-Q4_K_M.gguf" \ -c 4096 -ngl 99

Quick specs

Parameters48B
Architecturelinear
Context1.0M tokens
Modalitytext
Min RAM18.7 GB
Rec. RAM29.3 GB (Q4_K_M)
LicenseMIT
FamilyKimi Linear
Code Chat Reasoning

About this model

Kimi Linear is Moonshot AI's long-context efficient architecture release, using Kimi Delta Attention to cut KV-cache pressure and improve decoding throughput at very long sequence lengths.

  • 48B total params with 3B activated and 1M context
  • Kimi Delta Attention for lower KV-cache usage
  • Designed for efficient long-context inference and high decode throughput

あなたのハードウェア

検出中...

おすすめ

最適なハードウェア

Kimi Linear 48B A3Bのおすすめ

このモデルを実行

量子化オプション

量子化レベル別VRAM推定値

No hardware detected — fit column shows raw VRAM estimates

QuantBitsVRAMQualityFit
Q2_K
2
18.7 GB
Low
Q3_K_S
3
23.5 GB
Low
NVFP4
4
26.9 GB
Medium
Q4_K_M
4
29.3 GB
Medium
Q5_K_M
5
34.6 GB
High
Q6_K
6
39.4 GB
High
Q8_0
8
51.4 GB
Very High
F16
16
98.4 GB
Maximum

Quality benchmarks

Kimi Linear 48B A3B benchmark scores

Benchmark verified

Reasoning

MMLU-Pro51.0%
GPQA Diamond
MATH-500
ARC Challenge

Source: official · 2025-10-30

ハードウェア互換性

全ハードウェアの適合度推定

カリキュレーターを開く

Computing compatibility...

メモリ内訳

Reference: RTX 2060 6GB

Weights29.3 GB
KV Cache0.9 GB
Runtime2.4 GB
Headroom0.6 GB

よくある質問

FAQ — Kimi Linear 48B A3B

関連項目