InternScience
Agents-A1 4B
最先端
65.2Kダウンロード64いいねJul 2026公開日262K トークンコンテキストApache 2.0ライセンス62 良好品質
Agents-A1 4B (4.5B parameters) requires approximately 4.7 GB of VRAM with Q4_K_M quantization. For the best balance of quality and speed, we recommend hardware with at least 6 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 Agents-A1 4B on your machine.
Run
docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
--hf-repo "InternScience/Agents-A1-4B" \
--hf-file "Agents-A1-4B-Q4_K_M.gguf" \
-c 4096 -ngl 99Quick specs
Parameters4.5B
Architecturedense
Context262K tokens
Modalitytext+vision
Min RAM1.8 GB
Rec. RAM2.7 GB (Q4_K_M)
LicenseApache 2.0
FamilyAgents-A1
✓ Reasoning
About this model
- •4.5B dense and vision-capable — fits in ~3GB at Q4.
- •262K context, unusually long for a 4B-class model.
- •Hybrid linear + full attention (1-in-4 layers) keeps long-context KV cache small.
- •Shares the Agents-A1 long-horizon agentic training recipe.
関連モデル
おすすめ
最適なハードウェア
Agents-A1 4B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
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q1_0_G128 | 1.125 | 0.6 GB | Very Low | — |
Q2_0_G128 | 1.71 | 1.2 GB | Low | — |
Q2_K | 2 | 1.8 GB | Low | — |
Q3_K_S | 3 | 2.2 GB | Low | — |
NVFP4 | 4 | 2.5 GB | Medium | — |
Q4_K_M | 4 | 2.7 GB | Medium | — |
Q5_K_M | 5 | 3.2 GB | High | — |
Q6_K | 6 | 3.7 GB | High | — |
Q8_0 | 8 | 4.8 GB | Very High | — |
F16 | 16 | 9.2 GB | Maximum | — |
Quality benchmarks
Agents-A1 4B benchmark scores
Coding
SWE-bench Verified—
HumanEval+—
Aider Polyglot—
LiveCodeBench59.6%
General
Chatbot Arena—
IFEval94.8%
Source: official · 2026-07-13
ハードウェア互換性
全ハードウェアの適合度推定
Computing compatibility...
メモリ内訳
Reference: RTX 2060 6GB
Weights2.7 GB
KV Cache0.5 GB
Runtime0.9 GB
Headroom0.6 GB
よくある質問
FAQ — Agents-A1 4B
関連項目