Quark 135M — Hardware Requirements & GPU Compatibility
Chat
Quark 135M is a 135M-parameter open language model from ThingAI. It supports a context window of up to 2,048 tokens. At Q4_K_M it needs about 0.43 GB of VRAM — see which GPUs and Macs can run it below.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 2K)Best on 24 GB (Q4_K_M · ~1524 tok/s · 2K)Macs from 8 GB (Q4_K_M · ~167 tok/s)
Apache 2.0
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Quark 135M
One recommended way. Every other method, OS and quant is in the step-by-step guide.
Recommended
vllm serve ThingAI/Quark-135m --max-model-len 2048vLLM repo with its own quantisation
Download: Weights on Hugging Face
Other ways
each opens step-by-stepRun in cloud
Fits on RTX 3060 12GB (11 GB headroom) · Q4_K_M
- Generation speed
- ~544 tok/s
- generation speed
- Cost per 1M output tokens
- $0.042
- per 1M output tokens
How Much VRAM Does Quark 135M Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 0.4 GB | — | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 2K context · ~1638 tok/s | |
| 0.4 GB | — | 3-bit medium — 3-bit medium quantization | Fits: 2K context · ~1598 tok/s | |
| 0.4 GB | — | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 2K context · ~1524 tok/s | |
| 0.4 GB | — | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 2K context · ~1489 tok/s | |
| 0.5 GB | — | 6-bit, very good — 6-bit quantization, very good quality | Fits: 2K context · ~1424 tok/s | |
| 0.5 GB | — | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 2K context · ~1365 tok/s | |
| 0.6 GB | — | 16-bit · bfloat — Brain floating point 16 — preferred for training | Fits: 2K context · ~1057 tok/s |
Selected Q4_K_M
4-bit medium quantization0.4GBat 2K context
- Weights
- 0.1 GB
- KV cache
- 0.0 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 23.6 GB free · ~1524 tok/s
≥50 tok/s needs ≥33 GB/s · RTX 5090 ~2709, RTX 3090 Ti ~1524
Which Hardware Can Run Quark 135M?
Q4_K_M · 0.4 GBQuark 135M (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. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 163 devices with unified memory can run Quark 135M, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
≤ 11 GB
≤ 11 GB tier →fits · 7.6 GB free · up to 2K ctx
16 GPUs · 1 Mac class · 2 device families
RTX 3080~1149Simulate output on RTX 3080RTX 2080 Ti~931Simulate output on RTX 2080 TiRTX 3070 Ti~920Simulate output on RTX 3070 TiM3*8 GB~167Simulate output on iMac 24" M3 (8 GB)Orin Nano~154Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~111Simulate output on iPhone 1720 GB
fits · 19.6 GB free · up to 2K ctx
1 GPU
RX 7900 XT~1116Simulate output on RX 7900 XT32–40 GB
fits · 31.6 GB free · up to 2K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~2709Simulate output on RTX 5090M5 Max36 GB~749Simulate output on Mac Studio M5 Max (36 GB)M532 GB~250Simulate output on MacBook Air 13" M5 (32 GB)Orin~310Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~204Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Similar models
1| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| SmolLM 135M Instruct FP32modularai | — | 0.4 GB | ~1524 | same size |
Specifications
- Publisher
- ThingAI
- Parameters
- 135M
- Architecture
- LlamaForCausalLM
- Context Length
- 2,048 tokens
- Vocabulary Size
- 49,152
- Release Date
- 2026-05-04
- License
- Apache 2.0
Frequently Asked Questions
How much VRAM does Quark 135M need?
Quark 135M requires 0.4 GB of VRAM at Q4_K_M, or 0.6 GB at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 135M × 4.8 bits ÷ 8 = 0.1 GB
KV Cache + Overhead ≈ 0.3 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
What's the best quantization for Quark 135M?
For Quark 135M, 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
★ Recommended — best balance of quality and VRAM usage.
Can I run Quark 135M on a Mac?
Yes — MacBook Air 13" M3 (8 GB) and 145 other Macs can run Quark 135M. 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 Quark 135M locally?
Yes — Quark 135M can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Quark 135M?
At Q4_K_M, Quark 135M can reach ~11163 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~1524 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 ÷ 0.43 × 0.65 = ~12093 tok/s
Estimated speed at Q4_K_M (0.4 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Quark 135M?
At Q4_K_M, the download is about 0.08 GB. The full-precision BF16 version is 0.27 GB. The smallest option (Q2_K) is 0.06 GB.
Which GPUs can run Quark 135M?
52 consumer GPUs can run Quark 135M at Q4_K_M. Top options include AMD Radeon RX 6700 XT, AMD Radeon RX 6800, AMD Radeon RX 6800 XT. 52 GPUs have plenty of headroom for comfortable inference.
Which devices can run Quark 135M?
163 devices with unified memory can run Quark 135M at Q4_K_M (0.4 GB), including AMD Ryzen AI 9 HX 370 (Strix Point) Laptop, ASUS Ascent GX10, Beelink GTR9 Pro (Ryzen AI Max+ 395, 128 GB), Framework Desktop (Ryzen AI Max+ 395, 128 GB). Apple Silicon Macs use unified memory shared between CPU and GPU, making them well-suited for local LLM inference.