llmrun
arianrajeQwen 3Qwen3Mamba3ForCausalLM

Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0 — Hardware Requirements & GPU Compatibility

ChatFunctions

Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0 is a 4.7B-parameter open language model from arianraje in the Qwen 3 family. It supports a context window of up to 40,960 tokens. At Q4_K_M it needs about 3.33 GB of VRAM — see which GPUs and Macs can run it below.

4.7B params · dense41K contextApache 2.048.3K downloads1 likes
Based on Qwen3 4B
S

Runs great on a 24 GB GPU

Runs from8 GB Q4_K_M · 40KBest on24 GB Q4_K_M · ~197 tok/s · 40KMacs from8 GB Q4_K_M · ~22 tok/s

Apache 2.0

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0

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 arianraje/qwen3-4b-mamba3-hybrid-stage2b-kd-bias0 --max-model-len 16384

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 (7 GB headroom) · Q4_K_M

Generation speed
~70 tok/s
generation speed
Cost per 1M output tokens
$0.25
per 1M output tokens

How Much VRAM Does Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0 Need?

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

QuantizationVRAM+32KOn RTX 4090 (reference)
2.5 GB5.3 GB2-bit · K-quant — 2-bit quantization with K-quant improvementsFits: 40K context · ~262 tok/s
2.8 GB5.6 GB3-bit medium — 3-bit medium quantizationFits: 40K context · ~234 tok/s
3.3 GB6.2 GB4-bit medium — 4-bit medium quantization — most popular sweet spotFits: 40K context · ~197 tok/s
3.9 GB6.7 GB5-bit medium — 5-bit medium quantization — good quality/size tradeoffFits: 40K context · ~169 tok/s
4.4 GB7.2 GB6-bit, very good — 6-bit quantization, very good qualityFits: 40K context · ~149 tok/s
5.2 GB8.1 GB8-bit, near-lossless — 8-bit quantization, near-losslessFits: 40K context · ~125 tok/s
10.0 GB12.8 GB16-bit · bfloat — Brain floating point 16 — preferred for trainingFits: 40K context · ~66 tok/s
est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

4.7GBat 16K context

Weights
2.8 GB
KV cache
1.5 GB
Overhead
0.3 GB

Fits RTX 4090 (reference) · 19.4 GB free · ~197 tok/s

≥50 tok/s needs ≥256 GB/s · RTX 5090 ~350, M5 Ultra ~258

Which Hardware Can Run Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0?

Q4_K_M · 3.3 GB

Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0 (Q4_K_M) requires 3.3 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 5+ GB is recommended. Using the full 41K context window can add up to 3.6 GB, bringing total usage to 6.9 GB. 114 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce GTX 1650 SUPER. 185 devices with unified memory can run Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0, including Mac Studio M5 Ultra (256 GB), Mac Studio M5 Ultra (512 GB), MacBook Air 13" M3 (8 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M
8 GB · RTX 4060Q4_K_M · 4.7 GB · ~53 tok/s · 40K8 GB tier →
12 GB · RTX 3060 12GBQ4_K_M · 4.7 GB · ~70 tok/s · 40K12 GB tier →
16 GB · RTX 5060 Ti 16GBQ4_K_M · 4.7 GB · ~87 tok/s · 40K16 GB tier →
24 GB · RTX 4090referenceQ4_K_M · 4.7 GB · ~197 tok/s · 40K24 GB tier →
Mac 32 GB · Mac Mini M4Q4_K_M · 4.7 GB · ~25 tok/s · 40KMacs →
Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 4.7 GB · ~115 tok/s · 40KMacs →

Alternatives to Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0

Qwen 3 family · 79 models →

Similar models

6
ModelScoreVRAMtok/sWhy it is similar
Qwen3.5 4BAlibaba98.53.4 GB~201same size
WeVisOne VL 4BTencent—4.7 GB~193same size
FrogNano 4B 2609Microsoft—3.4 GB~201same size
Molmo2 4BAllen AI—4.7 GB~193same size
Qwen3.6 27B MTPLX Optimized SpeedYoussofal—4 GB~185same size
AFM 4.5BArcee AI—4.3 GB~204same size

Specifications

Publisher
arianraje
Family
Qwen 3
Parameters
4.7B
Architecture
Qwen3Mamba3ForCausalLM
Context Length
40,960 tokens
Vocabulary Size
151,936
Release Date
2026-09-10
License
Apache 2.0

Frequently Asked Questions

How much VRAM does Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0 need?

Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0 requires 3.3 GB of VRAM at Q4_K_M, or 10.0 GB at BF16. Full 41K context adds up to 3.6 GB (6.9 GB total).

VRAM = Weights + KV Cache + Overhead

Weights = 4.7B × 4.8 bits ÷ 8 = 2.8 GB

KV Cache + Overhead ≈ 0.5 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.

KV Cache + Overhead ≈ 4.1 GB (at full 41K context)

VRAM usage by quantization

3.3 GB
6.9 GB

Learn more about VRAM estimation →

What's the best quantization for Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0?

For Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0, Q4_K_M (3.3 GB) offers the best balance of quality and VRAM usage. Q5_K_M (3.9 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 2.5 GB.

VRAM requirement by quantization

Q2_K
2.5 GB
Q4_K_M ★
3.3 GB
Q5_K_M
3.9 GB
Q6_K
4.4 GB
Q8_0
5.2 GB
BF16
10.0 GB

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0 on a Mac?

Yes — MacBook Air 13" M3 (8 GB) and 158 other Macs can run Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 3.3 GB of usable unified memory (RAM minus macOS overhead).

Can I run Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0 locally?

Yes — Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0 can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0?

At Q4_K_M, Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0 can reach ~1441 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~197 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 ÷ 3.33 × 0.65 = ~1562 tok/s

Estimated speed at Q4_K_M (3.3 GB)

~1562 tok/s
~197 tok/s
~1562 tok/s
~1441 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 Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0?

At Q4_K_M, the download is about 2.84 GB. The full-precision BF16 version is 9.48 GB. The smallest option (Q2_K) is 2.02 GB.

Which GPUs can run Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0?

114 consumer GPUs can run Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0 at Q4_K_M. Top options include AMD Radeon RX 5600 XT, AMD Radeon RX 5700 XT, AMD Radeon RX 580 8GB, NVIDIA GeForce GTX 1050 Ti. 107 GPUs have plenty of headroom for comfortable inference.

Which devices can run Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0?

185 devices with unified memory can run Qwen3 4B Mamba3 Hybrid Stage2b Kd Bias0 at Q4_K_M (3.3 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.