v2.2 · 2026-06-12

Meta Llama 3.3 70B Instruct

llama-3-3-70b-instruct

TEXT-CHAT

Meta Llama 3.3 70B Instruct is a 70B text chat model profile for local AI planning. This page records provider, license, context, quantization, hardware-fit estimates, setup hints, source links, and caveats so readers do not choose a model by name alone.

PROVIDER

Meta

FAMILY

llama

MODEL TYPE

text-chat

PARAMETERS

70B

MODALITIES

text

ARCHITECTURE

decoder-only Transformer (LlamaForCausalLM) with grouped-query attention

CONTEXT WINDOW

128,000 tokens

TRAINING TOKENS

15.0T

RELEASE DATE

2024-12-06

LICENSE & LINKS

Llama 3.3 Community Licenselicense text

MODEL CARD

huggingface.co/meta-llama/Llama-3.3-70B-Instruct

HUGGING FACE

meta-llama/Llama-3.3-70B-Instruct

GGUF REPOSITORIES

bartowski/Llama-3.3-70B-Instruct-GGUF community conversion

Community GGUF conversion with Q4_K_M, Q5_K_M and Q8_0 file-size table.

SETUP HINTS

OLLAMA

ollama run llama3.3:70b

LM STUDIO

Use bartowski/Llama-3.3-70B-Instruct-GGUF and select Q4_K_M only on high-memory machines.

LLAMA.CPP

llama-server -hf bartowski/Llama-3.3-70B-Instruct-GGUF:Q4_K_M

QUANTIZATION FILE SIZES (GGUF)

Q4_K_M

42.52 GB

Q5_K_M

49.95 GB

Q8_0

74.98 GB

Quantization note: Q4_K_M

RAM / VRAM ESTIMATES

MIN RAM

48 GB

COMFORTABLE RAM

64 GB

MIN VRAM

48 GB

COMFORTABLE VRAM

80 GB

These are conservative local-inference estimates, not official hardware requirements.

HARDWARE FIT

CPU only (no GPU)

limited

8 GB RAM

16 GB RAM

32 GB RAM

8 GB VRAM

12 GB VRAM

24 GB VRAM

partial_offload_only

Apple Silicon (unified memory)

requires_64gb_or_more_unified_memory

Hardware fit values are conservative local-inference estimates based on GGUF size plus runtime and KV-cache overhead. Actual requirements depend on context length, quantization, runtime, GPU offload, and other running apps.

BEST FOR

·highest-quality Llama 70B local chat target in this batch

·multilingual drafting and analysis

·agentic or tool-use workflows on high-memory systems

·local alternative when 405B is impractical

AVOID IF

·you have less than about 48 GB available memory for the model

·you need fast consumer-GPU inference

·you require an Apache-2.0 license

CAVEATS

·RAM and VRAM estimates are conservative local-inference estimates based on GGUF file size plus runtime and KV-cache overhead, not official Meta requirements.

·Full-GPU Q4_K_M inference generally exceeds 24 GB VRAM; 24 GB cards require CPU offload, smaller quantization, shorter context, or a different model.

SOURCE URLS