Model
Model explorer

Llama-3-ELYZA-JP-8B

OPEN
ELYZA · ELYZA LLM for JP family · released Jun 26, 2024

Llama-3 8B continually pretrained and instruction-tuned for Japanese; comparable to GPT-3.5 Turbo/Claude 3 Haiku on ELYZA Tasks 100 (the flagship 70B model, not this 8B, is the one claimed to exceed GPT-4).

ReasoningCodingVisionFunction callingTool useAgentic
324.8
Elo · unrated
Parameters
8B
Active params
8B (dense)
Context
8K tokens
Architecture
8B dense transformer (Llama-3 architecture, Japanese continual pretrain + instruction tuning)
License
Llama 3 Community License
Languages
2+
API price (in/out)
No hosted API
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
MT-BenchHuman preference7.78#50
best: Hunyuan-Large (A52B) · 9.4
Run it locally
VRAM @ Q4
5 GB
VRAM @ FP16
16.1 GB
Fits on (Q4)
RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB
Throughput data unavailable.
Quantizations
GGUF Q4 · AWQ
ELYZA LLM for JP family
Elo progression across releases
Llama-3-ELYZA-JP-8BJun 2024324.8Llama-3.1-ELYZA-JP-70BOct 20241190.8
Resources
HFelyza/Llama-3-ELYZA-JP-8BDOCSelyza.ai
API price weights · each benchmark row carries its own source badge (see methodology)