llmrun
SupraLabsQwen3ForCausalLM

Supra2.5 Pro Instruct — Hardware Requirements & GPU Compatibility

Chat

Supra2.5 Pro Instruct is a 67M-parameter open language model from SupraLabs. It supports a context window of up to 512 tokens. At Q4_K_M it needs about 0.40 GB of VRAM — see which GPUs and Macs can run it below.

67M params · dense1K contextApache 2.01.2K downloads24 likes
S

Runs great on a 24 GB GPU

Runs from8 GB Q4_K_MBest on24 GB Q4_K_M · ~1638 tok/sMacs from8 GB Q4_K_M · ~179 tok/s

Apache 2.0

Reference GPU · RTX 4090 24 GB

Estimates, ±20% · how we estimate

Run Supra2.5 Pro Instruct

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 SupraLabs/Supra2.5-Pro-Instruct --max-model-len 512

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 (11 GB headroom) · Q4_K_M

Generation speed
~585 tok/s
generation speed
Cost per 1M output tokens
$0.028
per 1M output tokens

How Much VRAM Does Supra2.5 Pro Instruct Need?

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

QuantizationVRAM+32KOn RTX 4090 (reference)
0.4 GB—2-bit · K-quant — 2-bit quantization with K-quant improvementsFits: — context · ~1680 tok/s
0.4 GB—3-bit medium — 3-bit medium quantizationFits: — context · ~1638 tok/s
0.4 GB—4-bit medium — 4-bit medium quantization — most popular sweet spotFits: — context · ~1638 tok/s
0.4 GB—5-bit medium — 5-bit medium quantization — good quality/size tradeoffFits: — context · ~1598 tok/s
0.4 GB—6-bit, very good — 6-bit quantization, very good qualityFits: — context · ~1560 tok/s
0.4 GB—8-bit, near-lossless — 8-bit quantization, near-losslessFits: — context · ~1524 tok/s
0.5 GB—16-bit · bfloat — Brain floating point 16 — preferred for trainingFits: — context · ~1310 tok/s
est.calculated, no published file yet

Selected Q4_K_M

4-bit medium quantization

0.4GBat 1K context

Weights
0.1 GB
KV cache
0.0 GB
Overhead
0.3 GB

Fits RTX 4090 (reference) · 23.7 GB free · ~1638 tok/s

≥50 tok/s needs ≥31 GB/s · RTX 5090 ~2912, M5 Ultra ~2149

Which Hardware Can Run Supra2.5 Pro Instruct?

Q4_K_M · 0.4 GB

Supra2.5 Pro Instruct (Q4_K_M) requires 0.4 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 1+ 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 Supra2.5 Pro Instruct, including Mac Studio M5 Ultra (256 GB), Mac Studio M5 Ultra (512 GB).

By VRAM tier · typical GPUs and Macs at Q4_K_M
8 GB · RTX 4060Q4_K_M · 0.4 GB · ~442 tok/s · —8 GB tier →
12 GB · RTX 3060 12GBQ4_K_M · 0.4 GB · ~585 tok/s · —12 GB tier →
16 GB · RTX 5060 Ti 16GBQ4_K_M · 0.4 GB · ~728 tok/s · —16 GB tier →
24 GB · RTX 4090referenceQ4_K_M · 0.4 GB · ~1638 tok/s · —24 GB tier →
Mac 32 GB · Mac Mini M4Q4_K_M · 0.4 GB · ~210 tok/s · —Macs →
Mac 64 GB · Mac Studio M4 MaxQ4_K_M · 0.4 GB · ~956 tok/s · —Macs →

Alternatives to Supra2.5 Pro Instruct

Similar models

1
ModelScoreVRAMtok/sWhy it is similar
Llama 68MJackFram—0 GB~16380same size

Specifications

Publisher
SupraLabs
Parameters
67M
Architecture
Qwen3ForCausalLM
Context Length
512 tokens
Vocabulary Size
16,384
Release Date
2026-11-16
License
Apache 2.0

Frequently Asked Questions

How much VRAM does Supra2.5 Pro Instruct need?

Supra2.5 Pro Instruct requires 0.4 GB of VRAM at Q4_K_M, or 0.5 GB at BF16.

VRAM = Weights + KV Cache + Overhead

Weights = 67M × 4.8 bits ÷ 8 = 0 GB

KV Cache + Overhead ≈ 0.4 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

0.4 GB

Learn more about VRAM estimation →

What's the best quantization for Supra2.5 Pro Instruct?

For Supra2.5 Pro Instruct, Q4_K_M (0.4 GB) offers the best balance of quality and VRAM usage. Q5_K_M (0.4 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 0.4 GB.

VRAM requirement by quantization

Q2_K
0.4 GB
Q4_K_M ★
0.4 GB
Q5_K_M
0.4 GB
Q6_K
0.4 GB
Q8_0
0.4 GB
BF16
0.5 GB

★ Recommended — best balance of quality and VRAM usage.

Learn more about quantization →

Can I run Supra2.5 Pro Instruct on a Mac?

Yes — MacBook Air 13" M3 (8 GB) and 158 other Macs can run Supra2.5 Pro Instruct. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 0.4 GB of usable unified memory (RAM minus macOS overhead).

Can I run Supra2.5 Pro Instruct locally?

Yes — Supra2.5 Pro Instruct can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.

How fast is Supra2.5 Pro Instruct?

At Q4_K_M, Supra2.5 Pro Instruct can reach ~12000 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~1638 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 ÷ 0.4 × 0.65 = ~13000 tok/s

Estimated speed at Q4_K_M (0.4 GB)

~13000 tok/s
~1638 tok/s
~13000 tok/s
~12000 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 Supra2.5 Pro Instruct?

At Q4_K_M, the download is about 0.04 GB. The full-precision BF16 version is 0.13 GB. The smallest option (Q2_K) is 0.03 GB.

Which GPUs can run Supra2.5 Pro Instruct?

114 consumer GPUs can run Supra2.5 Pro Instruct at Q4_K_M. 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 Supra2.5 Pro Instruct?

185 devices with unified memory can run Supra2.5 Pro Instruct at Q4_K_M (0.4 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.