Swiss AI

Apertus v1.5 70B

Name: Apertus v1.5 70B
Rating: 76 (44 reviews)
Author: Swiss AI

Aktuell

HuggingFace

1.0KDownloads34LikesJul 2026Veröffentlicht262K TokenKontextApache 2.0Lizenz69 GutQualität

Apertus v1.5 70B (72B parameters) requires approximately 50.3 GB of VRAM with Q4_K_M quantization. For the best balance of quality and speed, we recommend hardware with at least 58 GB of VRAM.

Loslegen

— kopieren & einfügen, um lokal auszuführen

Copy-paste commands to run Apertus v1.5 70B on your machine.

Run

docker run --rm -it ghcr.io/ggerganov/llama.cpp:full \
  --hf-repo "swiss-ai/Apertus-v1.5-70B" \
  --hf-file "Apertus-v1.5-70B-Q4_K_M.gguf" \
  -c 4096 -ngl 99

Quick specs

Parameters72B

Architecturedense

Context262K tokens

Modalitytext+vision

Min RAM28.1 GB

Rec. RAM43.9 GB (Q4_K_M)

LicenseApache 2.0

FamilyApertus

✓ Chat✓ Reasoning

About this model

Apertus 1.5 70B is the large member of Switzerland's fully open model family from the Swiss AI initiative, built on 17T pretraining tokens plus a 2T-token multimodal continued-pretraining mix on top of Apertus 1.0. It is a decoder-only transformer using the xIELU activation and the AdEMAMix optimizer, trained exclusively on fully open data, with 262K context and multilingual plus vision support.

•Fully open: open weights, open training data, and published EU Code of Practice documentation.
•72B dense parameters including a ~294M vision encoder for image-text-to-text.
•262,144-token context window.
•Continued pretraining of Apertus 1.0 with a 2T-token multimodal mix; 17T pretraining tokens total.
•Apache 2.0, developed by the Swiss AI initiative (EPFL / ETH Zurich).

Verwandte Modelle

Schnellauswahl

Bestes BudgetA

MacBook Pro M4 Max 96GB~$2,499 — 15 tok/s

Beste GesamtwahlS

NVIDIA H100 80GB~$40,000 — 70 tok/s

Beste Hardware

Top-Empfehlungen für Apertus v1.5 70B

Dieses Modell ausführen

Apertus v1.5 70B on NVIDIA H100 80GB Apertus v1.5 70B on NVIDIA H800 80GB Apertus v1.5 70B on NVIDIA GH200 96GB

Quantisierungsoptionen

VRAM-Schätzungen nach Quantisierungsstufe

No hardware detected — fit column shows raw VRAM estimates

Quant	Bits	VRAM	Quality	Fit
Q1_0_G128	1.125	10.4 GB	Very Low	—
Q2_0_G128	1.71	19.2 GB	Low	—
Q2_K	2	28.1 GB	Low	—
Q3_K_S	3	35.3 GB	Low	—
NVFP4	4	40.3 GB	Medium	—
Q4_K_M	4	43.9 GB	Medium	—
Q5_K_M	5	51.8 GB	High	—
Q6_K	6	59.0 GB	High	—
Q8_0	8	77.0 GB	Very High	—
F16	16	147.6 GB	Maximum	—

Hardware-Kompatibilität

Eignungsschätzungen für alle Hardware

Rechner öffnen

Computing compatibility...

Speicheraufschlüsselung

Reference: RTX 2060 6GB

Weights43.9 GB

KV Cache4.9 GB

Runtime0.9 GB

Headroom0.6 GB

Häufig gestellte Fragen

FAQ — Apertus v1.5 70B

Siehe auch

Quantisierungsleitfaden Bewertungsmethodik Rechner öffnen