HarmBench Llama 2 13B Cls — Hardware Requirements & GPU Compatibility
Chat
HarmBench Llama 2 13B Cls is a 13.0B-parameter open language model from cais in the Llama 2 family. It supports a context window of up to 2,048 tokens. At Q4_K_M it needs about 9.79 GB of VRAM — see which GPUs and Macs can run it below.
Runs great on a 24 GB GPU
Runs from 8 GB (IQ3_M · 2K)Best on 24 GB (Q4_K_M · ~67 tok/s · 2K)Macs from 16 GB (Q4_K_M · ~14 tok/s)
MIT
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run HarmBench Llama 2 13B Cls
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
ollama run hf.co/mradermacher/HarmBench-Llama-2-13b-cls-i1-GGUF:Q4_K_MQ4_K_M fits fully on the RTX 4090 · 14.2 GB free · 2K context
Download: HarmBench-Llama-2-13b-cls.i1-Q4_K_M.gguf7.87 GBmradermacher
Ways to runOther ways
each opens step-by-stepRun in cloud
Your own private copy · Q4_K_M · 24 GB GPU
- The easiest way to try this model
- Running in a few minutes, easy setup
- Chat right here or connect any app
$5+ bonus for new users
Runs on RTX 3080 Ti (12 GB) (2 GB headroom) · Q4_K_M
- Generation speed
- ~61 tok/s
- generation speed
- Cost per 1M output tokens
- $0.56
- per 1M output tokens
How Much VRAM Does HarmBench Llama 2 13B Cls Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 7.5 GB | — | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 2K context · ~87 tok/s | |
| 7.7 GB | — | 3-bit small — 3-bit small quantization | Fits: 2K context · ~85 tok/s | |
| 8.3 GB | — | 3-bit medium — 3-bit medium quantization | Fits: 2K context · ~79 tok/s | |
| 8.5 GB | — | 4-bit · legacy — 4-bit legacy quantization | Fits: 2K context · ~77 tok/s | |
| 9.8 GB | — | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 2K context · ~67 tok/s | |
| 11.3 GB | — | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 2K context · ~58 tok/s | |
| 12.7 GB | — | 6-bit, very good — 6-bit quantization, very good quality | Fits: 2K context · ~52 tok/s | |
| 15.0 GB | — | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 2K context · ~44 tok/s |
Selected Q4_K_M
4-bit medium quantization9.8GBat 2K context
- Weights
- 7.8 GB
- KV cache
- 1.7 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 14.2 GB free · ~67 tok/s
≥50 tok/s needs ≥752 GB/s · RTX 5090 ~119, RTX 3090 Ti ~67
Which Hardware Can Run HarmBench Llama 2 13B Cls?
Q4_K_M · 9.8 GBHarmBench Llama 2 13B Cls (Q4_K_M) requires 9.8 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 13+ GB is recommended. 40 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti, NVIDIA GeForce RTX 3080 Ti. 150 devices with unified memory can run HarmBench Llama 2 13B Cls, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), Mac Mini M2 Pro (16 GB).
12 GB
12 GB tier →fits · 2.2 GB free · up to 2K ctx
10 GPUs
RTX 3080 Ti~61Simulate output on RTX 3080 TiRTX 5070~45Simulate output on RTX 5070RTX 4070~34Simulate output on RTX 407020 GB
fits · 10.2 GB free · up to 2K ctx
1 GPU
RX 7900 XT~49Simulate output on RX 7900 XT32–40 GB
fits · 22.2 GB free · up to 2K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~119Simulate output on RTX 5090M5 Max36 GB~33Simulate output on Mac Studio M5 Max (36 GB)M532 GB~11Simulate output on MacBook Air 13" M5 (32 GB)Orin~14Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~9Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Alternatives to HarmBench Llama 2 13B Cls
Llama 2 family · 11 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Gemma 3 12B ITGoogle | 74.2 | 10.9 GB | ~82 | same size |
| Mistral Nemo Base 2407Mistral AI | — | 11 GB | ~81 | same size |
| Phi 3 Medium 128k InstructMicrosoft | — | 12 GB | ~72 | same size |
| Baichuan2 13B Basebaichuan-inc | — | 11.5 GB | ~76 | same size |
| Vicuna 13B V1.3LMSYS | — | 8.6 GB | ~76 | same size |
| Vllm Translategemma 12B ITchbae624 | — | 9.8 GB | ~73 | same size |
Specifications
- Publisher
- cais
- Family
- Llama 2
- Parameters
- 13.0B
- Architecture
- LlamaForCausalLM
- Context Length
- 2,048 tokens
- Vocabulary Size
- 32,000
- Release Date
- 2024-02-03
- License
- MIT
Frequently Asked Questions
How much VRAM does HarmBench Llama 2 13B Cls need?
HarmBench Llama 2 13B Cls requires 9.8 GB of VRAM at Q4_K_M, or 28.0 GB at BF16.
VRAM = Weights + KV Cache + Overhead
Weights = 13.0B × 4.8 bits ÷ 8 = 7.8 GB
KV Cache + Overhead ≈ 2 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
Can NVIDIA GeForce RTX 4090 run HarmBench Llama 2 13B Cls?
Yes, at Q8_0 (15.0 GB) or lower. Higher quantizations like BF16 (28.0 GB) exceed the NVIDIA GeForce RTX 4090's 24 GB.
What's the best quantization for HarmBench Llama 2 13B Cls?
For HarmBench Llama 2 13B Cls, Q4_K_M (9.8 GB) offers the best balance of quality and VRAM usage. Q5_K_S (10.9 GB) provides better quality if you have the VRAM. The smallest option is IQ2_XXS at 5.6 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run HarmBench Llama 2 13B Cls on a Mac?
Yes — Mac Mini M2 Pro (16 GB) and 133 other Macs can run HarmBench Llama 2 13B Cls. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 9.8 GB of usable unified memory (RAM minus macOS overhead).
Can I run HarmBench Llama 2 13B Cls locally?
Yes — HarmBench Llama 2 13B Cls can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is HarmBench Llama 2 13B Cls?
At Q4_K_M, HarmBench Llama 2 13B Cls can reach ~490 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~67 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 ÷ 9.8 × 0.65 = ~531 tok/s
Estimated speed at Q4_K_M (9.8 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of HarmBench Llama 2 13B Cls?
At Q4_K_M, the download is about 7.81 GB. The full-precision BF16 version is 26.03 GB. The smallest option (IQ2_XXS) is 3.58 GB.
Which GPUs can run HarmBench Llama 2 13B Cls?
40 consumer GPUs can run HarmBench Llama 2 13B Cls at Q4_K_M. Top options include AMD Radeon RX 6800, AMD Radeon RX 6800 XT, AMD Radeon RX 6900 XT, AMD Radeon RX 6700 XT. 26 GPUs have plenty of headroom for comfortable inference.
Which devices can run HarmBench Llama 2 13B Cls?
150 devices with unified memory can run HarmBench Llama 2 13B Cls at Q4_K_M (9.8 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.