llmrun
AsilarknessMaskedDiffusionLM

DiffuRefill 1B — Hardware Requirements & GPU Compatibility

Chat

DiffuRefill 1B is a 1B-parameter open language model from Asilarkness. It supports a context window of up to 2,048 tokens. At BF16 it needs about 2.60 GB of VRAM — see which GPUs and Macs can run it below.

1B params · dense2K contextApache 2.037.6K downloads3 likes
S

Runs great on a 24 GB GPU

Runs from8 GB BF16 · 2KBest on24 GB BF16 · ~252 tok/s · 2KMacs from8 GB BF16 · ~28 tok/s

Apache 2.0

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run DiffuRefill 1B

One recommended way. Every other method, OS and quant is in the step-by-step guide.Each way opens a step-by-step guide for your OS and quant.

Recommended

vLLM
one command · needs vLLM installed
$vllm serve Asilarkness/DiffuRefill-1B --max-model-len 2048

vLLM repo with its own quantisation

Download: Weights on Hugging Face

install, OS, other quants

Ways to runOther ways

each opens step-by-step

Run in cloud

Runs on RTX 3060 12GB (9 GB headroom) · BF16

Generation speed
~90 tok/s
generation speed
Cost per 1M output tokens
$0.19
per 1M output tokens

How Much VRAM Does DiffuRefill 1B Need?

Select a quantization level; the listed hardware adheres to it.

QuantizationVRAM+32KOn RTX 4090 (reference)
2.6 GB—16-bit · bfloat — Brain floating point 16 — preferred for trainingFits: 2K context · ~252 tok/s
est.calculated, no published file yet

Selected BF16

Brain floating point 16

2.6GBat 2K context

Weights
2.0 GB
KV cache
0.3 GB
Overhead
0.3 GB

Fits RTX 4090 (reference) · 21.4 GB free · ~252 tok/s

≥50 tok/s needs ≥200 GB/s · RTX 5090 ~448, M5 Ultra ~331

Which Hardware Can Run DiffuRefill 1B?

BF16 · 2.6 GB

DiffuRefill 1B (BF16) requires 2.6 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 4+ GB is recommended. 114 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 185 devices with unified memory can run DiffuRefill 1B, including Mac Studio M5 Ultra (256 GB), Mac Studio M5 Ultra (512 GB).

By VRAM tier · typical GPUs and Macs at BF16
8 GB · RTX 4060BF16 · 2.6 GB · ~68 tok/s · 2K8 GB tier →
12 GB · RTX 3060 12GBBF16 · 2.6 GB · ~90 tok/s · 2K12 GB tier →
16 GB · RTX 5060 Ti 16GBBF16 · 2.6 GB · ~112 tok/s · 2K16 GB tier →
24 GB · RTX 4090referenceBF16 · 2.6 GB · ~252 tok/s · 2K24 GB tier →
Mac 32 GB · Mac Mini M4BF16 · 2.6 GB · ~32 tok/s · 2KMacs →
Mac 64 GB · Mac Studio M4 MaxBF16 · 2.6 GB · ~147 tok/s · 2KMacs →

Alternatives to DiffuRefill 1B

Similar models

4
ModelScoreVRAMtok/sWhy it is similar
Gemma 3 1B ITGoogle—2.1 GB~993same size
LLaMA3.1 8B Instruct DFlash UltraChatz-lab—2.7 GB~269same size
Vaultgemma 1BGoogle—3.7 GB~286same size
Qwen3 8B Speculator.eagle3RedHatAI—2.1 GB~978same size

Specifications

Publisher
Asilarkness
Parameters
1B
Architecture
MaskedDiffusionLM
Context Length
2,048 tokens
Vocabulary Size
73,760
Release Date
2026-09-10
License
Apache 2.0

Frequently Asked Questions

How much VRAM does DiffuRefill 1B need?

DiffuRefill 1B requires 2.6 GB of VRAM at BF16.

VRAM = Weights + KV Cache + Overhead

Weights = 1B × 16 bits ÷ 8 = 2 GB

KV Cache + Overhead ≈ 0.6 GB (at 2K context + ~0.3 GB framework)

Fit ratings and hardware model lists check this model with room for a 16K-token context, which needs a little more memory.

VRAM usage by quantization

2.6 GB

Learn more about VRAM estimation →

Can I run DiffuRefill 1B on a Mac?

Yes — MacBook Air 13" M3 (8 GB) and 158 other Macs can run DiffuRefill 1B. Apple Silicon uses unified memory, so the model shares RAM with the system. At BF16 you need at least 2.6 GB of usable unified memory (RAM minus macOS overhead).

Can I run DiffuRefill 1B locally?

Yes — DiffuRefill 1B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is DiffuRefill 1B?

At BF16, DiffuRefill 1B can reach ~1846 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~252 tok/s. Speed depends mainly on GPU memory bandwidth. Real-world results typically within ±20%.

tok/s = (bandwidth GB/s ÷ model GB) × efficiency

Example: NVIDIA B200 → 8000 ÷ 2.6 × 0.65 = ~2000 tok/s

Estimated speed at BF16 (2.6 GB)

~2000 tok/s
~252 tok/s
~2000 tok/s
~1846 tok/s

Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.

Learn more about tok/s estimation →

What's the download size of DiffuRefill 1B?

At BF16, the download is about 2.00 GB.

Which GPUs can run DiffuRefill 1B?

114 consumer GPUs can run DiffuRefill 1B at BF16. Top options include AMD Radeon RX 5600 XT, AMD Radeon RX 5700 XT, AMD Radeon RX 580 8GB. 114 GPUs have plenty of headroom for comfortable inference.

Which devices can run DiffuRefill 1B?

185 devices with unified memory can run DiffuRefill 1B at BF16 (2.6 GB), including AMD Ryzen 7 8845HS (Hawk Point) Laptop, AMD Ryzen AI 7 350 (Krackan Point) Laptop, AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10. Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.