Skip to content
English

Switch theme

// IBM Granite · Text LLM

granite-3.0-2b-instruct: VRAM & local deployment

2.63B parameters. About 2.9 GB at 4-bit and 7.0 GB at BF16 with one 8K-token request.

released 2024-10-02licence: Commercial OKmemory figures are estimates

01Specs

PublisherIBM · ibm-granite
Family / seriesIBM Granite · granite-3.0
TypeText LLM (text-generation)
Parameters2.63B Source: counted from official safetensors weights
Context length4K tokens (4,096)max_position_embeddings in config.json
Architecture40 layers · GQA
Released weightsBF16 · 5.3 GBdownload
LicenceApache-2.0 · Commercial OKPermissive licence (Apache-2.0 / MIT / CC-BY / OpenMDW): commercial use allowed.
Released2024-10-02 (Hugging Face repo created)

02How much VRAM it needs (estimate)

Total = weights + 10% + 0.5GB overhead + KV cache for one request (FP16 KV). More concurrent users multiply the KV column.

PrecisionWeightsTotal @ 8KSmallest GPUMac
BF16 / FP16 (original)5.3 GB 7.0 GB 8GB GPUMac 16GB
FP8 / INT8 / Q8_02.8 GB 4.3 GB 8GB GPUMac 16GB
4-bit (Q4_K_M / AWQ / GPTQ-Int4)1.6 GB 2.9 GB 8GB GPUMac 16GB

KV cache per request: 8K 0.7 GB

Bits per weight: BF16 16, FP8 8.5, 4-bit 4.85. “Fits” = total ≤ 90% of memory; Mac GPU can use ~2/3 of unified memory (≤36GB) or ~3/4 (larger). For MoE, all experts must be in memory, so size by total parameters. Method →

03How to run it

Only runtimes the publisher or the runtime’s official library names for this model.

Which inference engine to choose →

04Official repositories & quantised builds

2 official repo(s) from ibm-granite.

05Other IBM Granite models

All 67 →
ModelParams4-bit totalReleased
granite-3.0-8b-instruct · same series8.17B7.3 GB2024-10-02
granite-switch-4.1-8b8.78B7.8 GB2026-10-14
granite-switch-4.1-30b30B22.7 GB2026-10-14
granite-speech-5.0-2b2.31B2.0 GB2026-10-07
ensemble-forecasting-with-ibm-granite-time-series——2026-10-01
granite-speech-5.0-470m-turboctc-nc473M0.8 GB2026-08-17
granite-4.2-3b3.66B3.6 GB2026-08-07
granite-4.2-8b8.79B7.7 GB2026-08-07
granite-4.2-30b29.3B22.2 GB2026-08-07

06Sources

Data checked 2024-12-19. Parameters, context and licence are copied from the official repository; memory figures are our estimates.

Data updated: Self-host open models(UTC+8)