Llama 3.3 70B

Name: Llama 3.3 70B
Rating: 82 (44 reviews)
Author: Meta

当前

HuggingFace

Ollama

909.0K下载量2.8K点赞Dec 2024发布日期128K tokens上下文Community许可证80 优秀质量

Llama 3.3 70B (70B parameters) requires approximately 49.4 GB of VRAM with Q4_K_M quantization. For the best balance of quality and speed, we recommend hardware with at least 57 GB of VRAM.

快速开始

— 复制粘贴即可本地运行

Copy-paste commands to run Llama 3.3 70B on your machine.

Run

ollama run llama3.3

Quick specs

Parameters70B

Architecturedense

Context128K tokens

Modalitytext

Min RAM27.3 GB

Rec. RAM42.7 GB (Q4_K_M)

LicenseCommunity

FamilyLlama

✓ Chat✓ Reasoning

About this model

Llama 3.3 70B is Meta's most capable single-GPU-class model, offering improved reasoning and instruction following over Llama 3.1 70B. Supports 128K context with enhanced multilingual and code capabilities.

Llama 3.3 70B 的最佳选择

运行此模型

Llama 3.3 70B on NVIDIA H100 80GB Llama 3.3 70B on NVIDIA H800 80GB Llama 3.3 70B on NVIDIA GH200 96GB

量化选项

各量化级别的 VRAM 估算

No hardware detected — fit column shows raw VRAM estimates

Quant	Bits	VRAM	Quality	Fit
Q2_K	2	27.3 GB	Low	—
Q3_K_S	3	34.3 GB	Low	—
NVFP4	4	39.2 GB	Medium	—
Q4_K_M	4	42.7 GB	Medium	—
Q5_K_M	5	50.4 GB	High	—
Q6_K	6	57.4 GB	High	—
Q8_0	8	74.9 GB	Very High	—
F16	16	143.5 GB	Maximum	—

Quality benchmarks

Llama 3.3 70B benchmark scores

Benchmark verified

Coding

SWE-bench Verified—

HumanEval+88.4%

Aider Polyglot—

LiveCodeBench—

Reasoning

MMLU-Pro68.9%

GPQA Diamond50.5%

MATH-50077.0%

ARC Challenge—

General

Chatbot Arena—

IFEval92.1%

Source: official · 2024-12-06

硬件兼容性

全部硬件的适配估算

打开计算器

Computing compatibility...

内存详细分析

Reference: RTX 2060 6GB

Weights42.7 GB

KV Cache4.9 GB

Runtime1.2 GB

Headroom0.6 GB

常见问题

FAQ — Llama 3.3 70B

另请参阅

量化指南评分方法打开计算器