Yz

exaone3.5

EXAONE 3.5 is a collection of instruction-tuned bilingual (English and Korean) generative models ranging from 2.4B to 32B parameters, developed and released by LG AI Research.

2.4b 7.8b 32b
Hızlı Kurulum (Ollama kuruluysa) 
ollama run exaone3.5

Ollama kurulu değil mi? ollama.com/download — Windows, macOS ve Linux için ücretsiz. İlk çalıştırmada model indirilir, sonrası tamamen çevrimdışıdır.

Varyantlar

Boyut büyüdükçe kalite artar, donanım ihtiyacı yükselir. Başlangıç için küçük varyantı deneyin.

EtiketBoyutBağlamGirdiKomut
latest 4.8GB 32K Text ollama run exaone3.5:latest
2.4b 1.6GB 32K Text ollama run exaone3.5:2.4b
7.8b 4.8GB 32K Text ollama run exaone3.5:7.8b
32b 19GB 32K Text ollama run exaone3.5:32b
2.4b-instruct-q4_K_M 1.6GB 32K Text ollama run exaone3.5:2.4b-instruct-q4_K_M
2.4b-instruct-q8_0 2.8GB 32K Text ollama run exaone3.5:2.4b-instruct-q8_0
2.4b-instruct-fp16 5.3GB 32K Text ollama run exaone3.5:2.4b-instruct-fp16
7.8b-instruct-q4_K_M 4.8GB 32K Text ollama run exaone3.5:7.8b-instruct-q4_K_M
7.8b-instruct-q8_0 8.3GB 32K Text ollama run exaone3.5:7.8b-instruct-q8_0
7.8b-instruct-fp16 16GB 32K Text ollama run exaone3.5:7.8b-instruct-fp16
32b-instruct-q4_K_M 19GB 32K Text ollama run exaone3.5:32b-instruct-q4_K_M
32b-instruct-q8_0 34GB 32K Text ollama run exaone3.5:32b-instruct-q8_0
32b-instruct-fp16 64GB 32K Text ollama run exaone3.5:32b-instruct-fp16

Model Detayları ve Benchmarklar (kaynak: ollama.com)

EXAONE 3.5 is a collection of instruction-tuned bilingual (English and Korean) generative models ranging from 2.4B to 32B parameters, developed and released by LG AI Research. EXAONE 3.5 language models include:

  1. 2.4B model optimized for deployment on small or resource-constrained devices
  2. 7.8B model matching the size of its predecessor but offering improved performance
  3. 32B model delivering powerful performance.

All models support long-context processing of up to 32K tokens. Each model demonstrates state-of-the-art performance in real-world use cases and long-context understanding, while remaining competitive in general domains compared to recently released models of similar sizes.

benchmarks

References

Paper

Hugging Face

Blog