Qwen3.5 2B (Q3_K_XL) — 14.4 GBon NVIDIA L40S

Code Multilingual Thinking Tool Calls Vision

Q3_K_XL NVIDIA L40S

Overview

Qwen3.5 2B is a 2.27B parameter dense language model by Qwen, with code, multilingual, thinking, tool-calls, vision capabilities. It supports a context window of up to 262,144 tokens.

Qwen3.5 2B is a lightweight model from Alibaba's Qwen 3.5 family built on the Gated Delta Networks hybrid architecture with 2.27 billion parameters, balancing capability and efficiency for edge deployment. It is natively multimodal, processing text, images, and video, with built-in thinking capabilities for chain-of-thought reasoning. The model supports a 262K context window and covers over 201 languages, handling code generation and multilingual tasks with ease. Released under the Apache 2.0 license, it runs in roughly 2 GB of VRAM at Q4, making it practical for self-hosted deployment on modest hardware.

At Q3_K_XL quantization (low quality tier), the model weighs 1.08 GB. This fits within the 48 GB of VRAM on NVIDIA L40S, enabling full GPU inference.

The NVIDIA L40S is a datacenter GPU with 48 GB of GDDR6 VRAM and 864 GB/s memory bandwidth. It delivers 362 FP16 TFLOPS with Ada Lovelace architecture. A versatile GPU for AI inference, training, and graphics workloads. Handles quantized models up to 40B parameters comfortably.

Hardware Requirements

Model size	1.08 GB
VRAM available	48 GB
VRAM used	14.4 GB
System RAM
Min RAM required	0 GB
GPU layers	24 / 24
Context size	262,144
Backend	cuda13
Flash attention	Yes
Reading from disk	Yes

Deploy

Prerequisites

Ensure your GPU nodes are prepared with the NVIDIA container toolkit:

ansible-playbook prositronic.infra.nvidia_container_toolkit

Command

helmfile --state-values-file <(curl -s https://www.prositronic.eu/values/qwen3-5-2b/q3_k_xl/nvidia-l40s.yaml) apply

Generated `values.yaml`

/values/qwen3-5-2b/q3_k_xl/nvidia-l40s.yaml

Loading values…

Install llama.cpp

Install llama.cpp from the official build instructions for your platform and backend.

Download Model

curl -L -o qwen3-5-2b.gguf "https://huggingface.co/unsloth/Qwen3.5-2B-GGUF/resolve/main/Qwen3.5-2B-UD-Q3_K_XL.gguf"

Start Server

llama-server \
  -m qwen3-5-2b.gguf \
  --n-gpu-layers 24 \
  --ctx-size 262144 \
  --flash-attn

Verify

curl http://localhost:8080/health

Frequently Asked Questions

How much VRAM does Qwen3.5 2B (Q3_K_XL) need?

The Q3_K_XL quantization of Qwen3.5 2B requires 1.08 GB. All 24 layers fit in the 48 GB of VRAM available on NVIDIA L40S, enabling full GPU acceleration.

Can I run Qwen3.5 2B on NVIDIA L40S?

Yes. NVIDIA L40S provides 48 GB of VRAM, which is enough to run Qwen3.5 2B (Q3_K_XL) with all layers on the GPU for optimal performance.

What is quantization?

Quantization reduces a model's numerical precision from its original floating-point format to a more compact representation. This shrinks the file size and VRAM footprint, making it possible to run large models on consumer hardware. The trade-off is a small reduction in output quality. Q3_K_XL compresses Qwen3.5 2B from its original size down to 1.08 GB.

What quantization should I choose for Qwen3.5 2B?

Q3_K_XL is a low-quality quantization. Higher-quality quants (Q8, Q6) preserve more model accuracy but need more VRAM. Lower quants (Q4, Q3, Q2) reduce VRAM usage at the cost of some quality. Choose based on your available hardware and quality requirements.

What is flash attention and why is it enabled?

Flash attention is a memory-efficient algorithm that speeds up the attention mechanism in transformer models. It reduces VRAM usage during inference by avoiding the materialisation of the full attention matrix. For Qwen3.5 2B on NVIDIA L40S, flash attention is enabled to maximise context length and throughput within the available 48 GB of VRAM.

How do I run Qwen3.5 2B (Q3_K_XL) with Ollama?

Run ollama run qwen3.5:2b-q3_k_xl to start Qwen3.5 2B (Q3_K_XL). Ollama handles downloading the model weights automatically on first run.

Last updated: March 13, 2026