HomeModel Memory Fit › Qwen2.5-VL 3B Instruct
Model Memory Fit · VLM

Qwen2.5-VL 3B Instruct: memory fit on every Jetson module.

Method v1.0 · dataset 2026-09-07 · verified 2026-09-07 · last updated September 2026

3.76 B parameters (text 3.13 B + vision tower) · 36 layers · 128,000-token max context. Hugging Face: Qwen/Qwen2.5-VL-3B-Instruct.

Model facts

FactValueNote
Parameters3.755 Btext 3.126 B + vision 0.629 B (estimated)
Layers36
Hidden size2048
Attention heads16
KV heads2
Head dimension128
Max context128,000 tokens
dtypebfloat16
Vocabulary151,936

Qwen (Alibaba) · hf:Qwen/Qwen2.5-VL-3B-Instruct · config.json · verified 2026-09-07

“"hidden_size": 2048, "num_attention_heads": 16, "num_key_value_heads": 2, "max_position_embeddings": 128000”

Parameters source

  • Hugging Face · hf:Qwen/Qwen2.5-VL-3B-Instruct · verified 2026-09-07 · model.safetensors.index.json · class A
    “"total_size": 7509245952”

Vision tower

FactValue
Layers32
Hidden size1280
Patch size14
Image size
Model type

Vision config source

  • Qwen (Alibaba) · hf:Qwen/Qwen2.5-VL-3B-Instruct · vision_config · verified 2026-09-07 · config.json vision_config · class A
    “"depth": 32, "hidden_size": 1280, "num_heads": 16, "patch_size": 14”

mmproj (vision weights): GGUF mmproj Q8_0 (vision tower), 0.845 GB.

mmproj source

Image tokens: 1280 tokens per image (dynamic: (H/28)×(W/28) tokens after 2×2 spatial merge; 1280 assumed for a ~1 MP image; class D).

Published quantised artefacts

A file the publisher or a community mirror actually ships, used as the class-A weight figure when the requested quant matches. Any quant without a row here falls back to parameters × bytes-per-parameter (class D for block quants).

QuantFormatFileSizeSource
GGUF Q4_K_M / INT4 (AWQ, MLC q4f16)GGUF Q4_K_M (text)Qwen2.5-VL-3B-Instruct-Q4_K_M.gguf1.93 GBggml-org (llama.cpp maintainers) · hf:ggml-org/Qwen2.5-VL-3B-Instruct-GGUF · verified 2026-09-07

Jetson tokens/s measurements

Published or archived throughput numbers, not modelled. Class C (external measured benchmark).

No Jetson-measured tokens/s reported for this model yet.

Every module × every quantisation

Verdict and total memory at 4096-token context, 1 sequence, llama.cpp, headless. Q4 cells link to the static breakdown page; every other cell links to the live engine at that quantisation.

Notes

params_b (3.75B) is the FULL model (text backbone + vision encoder) from safetensors total_size; the '3B' name refers to the LLM backbone only. Vision GGUF requires a separate mmproj file loaded alongside the text GGUF in llama.cpp.

Method and data: Model Memory Fit methodology. Full registry: model-memory.json.

Change the context, concurrency or runtime.

The live engine covers any context length, sequence count, KV precision and runtime, with a permanent link.

OPEN MODEL MEMORY FIT →