Meta Llama 3.1 8B Instruct

unsloth · released 2024-07-23 · llama3.1 license

Llama 3.1 8B Instruct is Meta's general-purpose 8B (Dec-2023 cutoff): vendor MMLU-Pro 48.3, IFEval 80.4, strong instruction-following and tool use, weaker on graduate science (GPQA 30.4).

Key specs

TypeLocal open-weight
Parameters8.03B total
Architecturellama
Context window131K tokens
Knowledge cutoff2023-12-01
Modalitiestext
Recommended backends
Minimum viable rig

Benchmark scores

GPQA Diamond
SWE-bench Verified
AIME
MMLU-Pro48.3%
BFCL v3 (tool use)
Composite score6.15
Community ratingNo reviews yet

VRAM & disk per quantization

QuantVRAMDiskRAMContext
Q4_K_M6.2 GB4.7 GB131K

Strengths & weaknesses

Strengths: Strong all-round small model (MMLU-Pro 48.3, IFEval 80.4); Good tool use (BFCL 76.1) and 128K context; Widely supported

Weaknesses: Low GPQA (vendor 30.4); Pre-reasoning-era; no AIME results