Skip to content
English

Switch theme

// Liquid LFM · Embedding

LFM2-ColBERT-350M: VRAM & local deployment

353M parameters. About 0.7 GB at 4-bit and 1.3 GB at BF16.

released 2025-10-28licence: Custom licence – read itmemory figures are estimates

01Specs

PublisherLiquid AI · LiquidAI
Family / seriesLiquid LFM · LFM2-ColBERT
TypeEmbedding (sentence-similarity)
Parameters353M Source: counted from official safetensors weights
Released weightsFP32 · 1.4 GBdownload
Licencelfm1.0 · Custom licence – read itThe publisher uses its own licence; we have not classified it. Read the licence text.
Released2025-10-28 (Hugging Face repo created)

02How much VRAM it needs (estimate)

Embedding models are small; memory grows with batch size and sequence length.

PrecisionWeightsWeights + overheadSmallest GPUMac
BF16 / FP16 (original)0.7 GB 1.3 GB 8GB GPUMac 16GB
FP8 / INT8 / Q8_00.4 GB 0.9 GB 8GB GPUMac 16GB
4-bit (Q4_K_M / AWQ / GPTQ-Int4)0.2 GB 0.7 GB 8GB GPUMac 16GB

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 LiquidAI.

05Other Liquid LFM models

All 26 →
ModelParams4-bit totalReleased
LFM2.5-24B-A2B24B / 2B16.7 GB2026-10-12
LFM2.5-VL-3B3.12B2.7 GB2026-08-11
LFM2.5-2.6B2.7B2.4 GB2026-07-28
LFM2.5-Encoder-350M-Diffusion354M0.8 GB2026-07-10
LFM2.5-230M230M0.8 GB2026-06-24
LFM2.5-8B-A1B8.47B / 1B6.2 GB2026-05-28
LFM2.5-1.2B-JP-2026061.17B1.4 GB2026-05-26
LFM2.5-ColBERT-350M353M0.7 GB2026-05-20

06Sources

Data checked 2026-08-18. Parameters, context and licence are copied from the official repository; memory figures are our estimates.

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