IBM

Granite Code 34B

現行

HuggingFace

Ollama

May 2024公開日8K トークンコンテキストApache 2.0ライセンス64 良好品質

Granite Code 34B (34B parameters) requires approximately 26.2 GB of VRAM with Q4_K_M quantization. For the best balance of quality and speed, we recommend hardware with at least 31 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 Granite Code 34B on your machine.

Run

ollama run granite-code:34b

Quick specs

Parameters34B

Architecturedense

Context8K tokens

Modalitytext

Min RAM13.3 GB

Rec. RAM20.7 GB (Q4_K_M)

LicenseApache 2.0

FamilyGranite

✓ Code

About this model

Granite Code 34B is IBM's largest code generation model, strong across 100+ programming languages.

Granite Code 34Bのおすすめ

NVIDIA A100 40GBA

40 GB

RTX PRO 5000 Blackwell 48GBA

48 GB

MacBook Pro M4 Max 64GBA

このモデルを実行

Granite Code 34B on NVIDIA A100 40GB Granite Code 34B on RTX PRO 5000 Blackwell 48GB Granite Code 34B on MacBook Pro M4 Max 64GB

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
Q2_K	2	13.3 GB	Low	—
Q3_K_S	3	16.7 GB	Low	—
NVFP4	4	19.0 GB	Medium	—
Q4_K_M	4	20.7 GB	Medium	—
Q5_K_M	5	24.5 GB	High	—
Q6_K	6	27.9 GB	High	—
Q8_0	8	36.4 GB	Very High	—
F16	16	69.7 GB	Maximum	—

Quality benchmarks

Granite Code 34B benchmark scores

Benchmark verified

Coding

SWE-bench Verified—

HumanEval+54.8%

Aider Polyglot—

LiveCodeBench—

Source: official · 2024-05-07

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

Weights20.7 GB

KV Cache3.7 GB

Runtime1.2 GB

Headroom0.6 GB

よくある質問