Llama 7B — Hardware Requirements & GPU Compatibility
Chat
This is a community reupload of Meta's original Llama 1 7B model, published by the huggyllama account on Hugging Face. The original Llama 1 was a 6.7-billion parameter base model released by Meta in early 2023, trained on 1 trillion tokens of publicly available data. It pioneered the wave of open-weight large language models. As a first-generation Llama model, it has been superseded by Llama 2 and Llama 3 in terms of quality and capability. It remains of historical and research interest as the model that catalyzed the open-source LLM ecosystem. This upload provides convenient access in Hugging Face Transformers format.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 2K)Best on 24 GB (Q4_K_M · ~147 tok/s · 2K)Macs from 8 GB (Q4_K_M · ~16 tok/s)
Other
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Llama 7B
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/tensorblock/llama-7b-GGUFQ4_K_M fits fully on the RTX 4090 · 19.5 GB free · 2K context
Download: llama-7b-Q4_K_M.gguf4.04 GBtensorblock
Ways to runOther ways
each opens step-by-stepRun in cloud
Runs on RTX 3080 Ti (12 GB) (7 GB headroom) · Q4_K_M
- Generation speed
- ~133 tok/s
- generation speed
- Cost per 1M output tokens
- $0.25
- per 1M output tokens
How Much VRAM Does Llama 7B Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 3.1 GB | — | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 2K context · ~208 tok/s | |
| 3.6 GB | — | 3-bit medium — 3-bit medium quantization | Fits: 2K context · ~182 tok/s | |
| 4.5 GB | — | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 2K context · ~147 tok/s | |
| 5.3 GB | — | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 2K context · ~124 tok/s | |
| 6.1 GB | — | 6-bit, very good — 6-bit quantization, very good quality | Fits: 2K context · ~107 tok/s | |
| 7.4 GB | — | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 2K context · ~88 tok/s | |
| 14.8 GB | — | 16-bit · full — Full half-precision — baseline for inference | Fits: 2K context · ~44 tok/s |
Selected Q4_K_M
4-bit medium quantization4.5GBat 2K context
- Weights
- 3.8 GB
- KV cache
- 0.4 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 19.6 GB free · ~147 tok/s
≥50 tok/s needs ≥342 GB/s · RTX 5090 ~262, RTX 3090 Ti ~147
Which Hardware Can Run Llama 7B?
Q4_K_M · 4.5 GBLlama 7B (Q4_K_M) requires 4.5 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 6+ 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 Llama 7B, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB), MacBook Air 13" M3 (8 GB).
≤ 11 GB
≤ 11 GB tier →fits · 3.5 GB free · up to 2K ctx
16 GPUs · 1 Mac class · 2 device families
RTX 3080~111Simulate output on RTX 3080RTX 2080 Ti~90Simulate output on RTX 2080 TiRTX 3070 Ti~89Simulate output on RTX 3070 TiM3*8 GB~16Simulate output on iMac 24" M3 (8 GB)Orin Nano~15Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~11Simulate output on iPhone 1720 GB
fits · 15.6 GB free · up to 2K ctx
1 GPU
RX 7900 XT~108Simulate output on RX 7900 XT32–40 GB
fits · 27.6 GB free · up to 2K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~262Simulate output on RTX 5090M5 Max36 GB~72Simulate output on Mac Studio M5 Max (36 GB)M532 GB~24Simulate output on MacBook Air 13" M5 (32 GB)Orin~30Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~20Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
How good is Llama 7B?
llmrun Score blends public benchmark boards into one number; 100 = the average open model.
Not enough data yet for an llmrun Score.
A Score needs results on enough independent boards. How the Composite Score is built
Benchmarks
The 6 boards that weigh most in the llmrun Score, of 10 with a result for this model; badges show who ran the test.
- MMLU
Epoch AI · Feb 2023Lab-reported
Lab-reported35.6%
#77/85 open
- BIG-Bench Hard
Epoch AI · Feb 2023Lab-reported
Lab-reported33.5%
#35/38 open
- GSM8K
Epoch AI · Feb 2023Lab-reported
Lab-reported11.0%
#58/62 open
- HellaSwag
Epoch AI · Feb 2023Lab-reported
Lab-reported76.2%
#24/47 open
- ARC Challenge
Epoch AI · Feb 2023Lab-reported
Lab-reported47.6%
#33/56 open
- WinoGrande
Epoch AI · Feb 2023Lab-reported
Lab-reported70.1%
#26/51 open
Alternatives to Llama 7B
Llama family · 20 models →Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Mathstral 7B v0.1Mistral AI | — | 6.8 GB | ~133 | same size |
| Mistral 7B Instruct v0.3Mistral AI | — | 6.8 GB | ~133 | same size |
| CodeQwen1.5 7BAlibaba | — | 5.7 GB | ~137 | same size |
| Starling LM 7B BetaNexusflow | — | 5.7 GB | ~133 | same size |
| Llava 1.5 7B HFllava-hf | — | 5 GB | ~141 | same size |
| Chatglm3 6BZ.ai | — | 5.2 GB | ~159 | same size |
Specifications
- Publisher
- huggyllama
- Family
- Llama
- Parameters
- 6.7B
- Architecture
- LlamaForCausalLM
- Context Length
- 2,048 tokens
- Vocabulary Size
- 32,000
- Release Date
- 2023-04-03
- License
- Other
Frequently Asked Questions
How much VRAM does Llama 7B need?
Llama 7B requires 4.5 GB of VRAM at Q4_K_M, or 14.8 GB at FP16.
VRAM = Weights + KV Cache + Overhead
Weights = 6.7B × 4.8 bits ÷ 8 = 4 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.
VRAM usage by quantization
What's the best quantization for Llama 7B?
For Llama 7B, Q4_K_M (4.5 GB) offers the best balance of quality and VRAM usage. Q5_K_M (5.3 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 3.1 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Llama 7B on a Mac?
Yes — Mac Mini M2 (8 GB) and 145 other Macs can run Llama 7B. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 4.5 GB of usable unified memory (RAM minus macOS overhead).
Can I run Llama 7B locally?
Yes — Llama 7B can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Llama 7B?
At Q4_K_M, Llama 7B can reach ~1079 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~147 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 ÷ 4.45 × 0.65 = ~1169 tok/s
Estimated speed at Q4_K_M (4.5 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Llama 7B?
At Q4_K_M, the download is about 4.04 GB. The full-precision FP16 version is 13.48 GB. The smallest option (Q2_K) is 2.86 GB.
Which GPUs can run Llama 7B?
52 consumer GPUs can run Llama 7B 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 Llama 7B?
163 devices with unified memory can run Llama 7B at Q4_K_M (4.5 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.