LG AI Research32BQ4_K_M 下约 20 GB 显存

EXAONE 3.5 32B — 显存、速度与本地部署

作者: Jakub Rusinowski · 最后更新:

Full-power EXAONE. Leads real-world usability benchmarks in its class. Excellent long-context performance at 32K context.

EXAONE 3.5 32B 在 Q4_K_M 下约需 20 GB 显存——量化权重加框架开销,不含 KV 缓存。在 Apple Silicon 上,这部分来自统一内存。

逻辑88
创意85
编程87

按量化级别的显存与速度

计算基准:NVIDIA RTX 4090 (24 GB)。包含 8K 上下文的 KV 缓存,因此数值高于上方的主数字。

量化显存速度(估算)适配
Q2_K
2.63 bpw
13.5 GB~57 tok/s可运行
Q3_K_M
3.41 bpw
16.6 GB~46 tok/s可运行
Q4_K_M
4.83 bpw
22.3 GB~34 tok/s勉强
Q5_K_M
5.67 bpw
25.6 GB~4 tok/s需卸载
Q6_K
6.56 bpw
29.2 GB~4 tok/s需卸载
Q8_0
8.50 bpw
36.9 GB~3 tok/s需卸载
F16
16.00 bpw
66.9 GB—放不下

黑色标记 = NVIDIA RTX 4090 (24 GB) 上的可用显存。 估算来自内存带宽屋顶线模型,详见 方法说明页. EXAONE 3.5 32B 显存计算器 →

运行 EXAONE 3.5 32B

目录中能运行 EXAONE 3.5 32B 的最便宜 GPU 是 AMD Radeon RX 7900 XTX (24 GB).

购买此硬件 AMD Radeon RX 7900 XTX 24GB — 24 GB VRAM · 355 W board power立即云端部署 RunPod 上的 RTX 4090 — 低至 $0.34/小时 · 价格核实于 2026-07

或在 Vast.ai 比较,低至 $0.35/小时 (typical low · varies)

作为亚马逊联盟成员,我们从符合条件的购买中获得收入。云 GPU 链接为推荐链接——我们可能获得佣金,您无需额外付费。

联盟营销声明: 本页部分链接为联盟推广链接——如果你通过它们购买,LLM Configurator 可能会获得佣金,而你无需支付任何额外费用。作为亚马逊联盟成员(Amazon Associate),LLM Configurator 会从符合条件的购买中获得收益。
AMD Radeon RX 7900 XTX 24GB
24 GB VRAM · 355 W board power
2026年价格波动较大——请以当前商品页价格为准。

如何运行 EXAONE 3.5 32B

安装 Ollama,然后运行:

ollama run exaone3.5:32b
Hugging Face 上的权重: LGAI-EXAONE/EXAONE-3.5-32B-Instruct ↗

规格

Preview — The model is released, but these specs are thin or rest on a single source. Individual fields may be wrong.

Preview — The model is released, but these specs are thin or rest on a single source. Individual fields may be wrong.

参数量
32 Billion
上下文窗口
32,768
架构
Dense, Decoder-only
提供商
LG AI Research
许可证
MIT
Practical deployment
Single consumer GPU
规格量化
Q4_K_M
系统内存
48 GB
记录更新于
2026-02-10
许可证MIT允许商业使用

Commercial use permitted. No usage restrictions beyond attribution.

质量与使用场景

评分由模型作者或独立评测方发布——衡量质量而非吞吐量,并非我们实测。

最适合reasoningchatcodingmultilingualrag

我的 GPU 能运行 EXAONE 3.5 32B 吗?

EXAONE 3.5 的其他尺寸

EXAONE 3.5 32B — 常见问题

How much VRAM does EXAONE 3.5 32B need?

About 20 GB at Q4_K_M — quantized weights plus framework overhead, before any KV cache. The cache grows with context length and is added on top; the table above folds it in. Apple Silicon counts unified memory toward the same figure.

Does EXAONE 3.5 32B run on an RTX 4090 (24 GB)?

Yes. EXAONE 3.5 32B needs about 20 GB at Q4_K_M, inside a 24 GB card, at an estimated 34 tokens/sec.

How do I run EXAONE 3.5 32B locally?

Install Ollama and run `ollama run exaone3.5:32b`. That pulls the weights and starts a local OpenAI-compatible endpoint; after the download nothing leaves the machine.

What other sizes does EXAONE 3.5 come in?

EXAONE 3.5 2.4B (2 GB), EXAONE 3.5 7.8B (6 GB), EXAONE 3.5 32B (20 GB). Every size shares the family's training and licence; the larger ones score higher and need proportionally more memory.