Llava 1.5 7B HF — Hardware Requirements & GPU Compatibility
Vision
LLaVA-1.5-7B is a 7-billion-parameter vision-language chat model made by fine-tuning Vicuna-7B, itself based on Llama 2, on GPT-generated multimodal instruction data, paired with a CLIP ViT-L/14 vision encoder via an MLP projector. It answers questions about images, describes visual content, and follows multi-turn multimodal instructions, and was one of the first widely used open vision-language chat models. At 7B it runs comfortably on a single consumer GPU once quantized. Context length is limited to 4,096 tokens, short by current standards given its 2023-era Llama 2 backbone. It is released under the Llama 2 Community License, permitting commercial use with some restrictions, including a separate license for very large companies; published in December 2023. Unlike newer small vision-language models, LLaVA-1.5 uses a simple architecture with no token-compression for image patches.
Runs great on a 24 GB GPU
Runs from 8 GB (Q4_K_M · 4K)Best on 24 GB (Q4_K_M · ~141 tok/s · 4K)Macs from 8 GB (Q3_K_M · ~19 tok/s)
Llama 2 Community
Reference GPU · RTX 4090 24 GB
Estimates, ±20% · how we estimate
Run Llava 1.5 7B HF
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/colossatr0n/llava-1.5-7b-hf-Q8_0-GGUFQ4_K_M fits fully on the RTX 4090 · 19.0 GB free · 4K context
Download: llava-1.5-7b-hf-Q8_0-Q4_K_M.gguf4.24 GBcolossatr0n
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
- ~127 tok/s
- generation speed
- Cost per 1M output tokens
- $0.27
- per 1M output tokens
How Much VRAM Does Llava 1.5 7B HF Need?
Select a quantization level; the listed hardware adheres to it.
| Quantization | VRAM | +32K | On RTX 4090 (reference) | |
|---|---|---|---|---|
| 3.3 GB | — | 2-bit · K-quant — 2-bit quantization with K-quant improvements | Fits: 4K context · ~199 tok/s | |
| 3.8 GB | — | 3-bit medium — 3-bit medium quantization | Fits: 4K context · ~173 tok/s | |
| 4.7 GB | — | 4-bit medium — 4-bit medium quantization — most popular sweet spot | Fits: 4K context · ~141 tok/s | |
| 5.5 GB | — | 5-bit medium — 5-bit medium quantization — good quality/size tradeoff | Fits: 4K context · ~118 tok/s | |
| 6.4 GB | — | 6-bit, very good — 6-bit quantization, very good quality | Fits: 4K context · ~102 tok/s | |
| 7.8 GB | — | 8-bit, near-lossless — 8-bit quantization, near-lossless | Fits: 4K context · ~84 tok/s | |
| 15.5 GB | — | 16-bit · full — Full half-precision — baseline for inference | Fits: 4K context · ~42 tok/s |
Selected Q4_K_M
4-bit medium quantization5.0GBat 4K context
- Weights
- 4.0 GB
- KV cache
- 0.7 GB
- Overhead
- 0.3 GB
Fits RTX 4090 (reference) · 19.0 GB free · ~141 tok/s
≥50 tok/s needs ≥358 GB/s · RTX 5090 ~250, RTX 3090 Ti ~141
Which Hardware Can Run Llava 1.5 7B HF?
Q4_K_M · 4.7 GBLlava 1.5 7B HF (Q4_K_M) requires 4.7 GB of VRAM to load the model weights. For comfortable inference with headroom for KV cache and system overhead, 7+ GB is recommended. 52 GPUs can run it, including NVIDIA GeForce RTX 5090, NVIDIA GeForce RTX 3090 Ti. 151 devices with unified memory can run Llava 1.5 7B HF, including Mac Studio (M3 Ultra, 256GB), Mac Studio (M3 Ultra, 512GB).
≤ 11 GB
≤ 11 GB tier →fits · 3.0 GB free · up to 4K ctx
16 GPUs · 2 device families
RTX 3080~106Simulate output on RTX 3080RTX 2080 Ti~86Simulate output on RTX 2080 TiRTX 3070 Ti~85Simulate output on RTX 3070 TiOrin Nano~14Simulate output on NVIDIA Jetson Orin Nano 8GB (Super)iPhone~10Simulate output on iPhone 1720 GB
fits · 15.0 GB free · up to 4K ctx
1 GPU
RX 7900 XT~103Simulate output on RX 7900 XT32–40 GB
fits · 27.0 GB free · up to 4K ctx
1 GPU · 3 Mac classes · 4 device families
RTX 5090~250Simulate output on RTX 5090M5 Max36 GB~69Simulate output on Mac Studio M5 Max (36 GB)M532 GB~23Simulate output on MacBook Air 13" M5 (32 GB)Orin~29Simulate output on NVIDIA Jetson AGX Orin 32GBSnapdragon X Elite (X1E-80-100)~19Simulate output on Snapdragon X Elite Copilot+ PC
By VRAM tier · typical GPUs and Macs at Q4_K_M
Similar models
6| Model | Score | VRAM | tok/s | Why it is similar |
|---|---|---|---|---|
| Qwen2.5 7B InstructAlibaba | 65.1 | 5.8 GB | ~131 | same size |
| Olmo 3 7B InstructAllen AI | — | 8.4 GB | ~114 | same size |
| Falcon H1 7B InstructTII UAE | — | 6.3 GB | ~130 | same size |
| DeepSeek R1 Distill Qwen 7BDeepSeek | — | 5.8 GB | ~131 | same size |
| Eurus 2 7B PRIMEPRIME-RL | — | 5.1 GB | ~131 | same size |
| Mathstral 7B v0.1Mistral AI | — | 6.8 GB | ~133 | same size |
Specifications
- Publisher
- llava-hf
- Family
- LLaVA
- Parameters
- 7.1B
- Architecture
- LlavaForConditionalGeneration
- Context Length
- 4,096 tokens
- Vocabulary Size
- 32,064
- Release Date
- 2023-12-05
- License
- Llama 2 Community
Frequently Asked Questions
How much VRAM does Llava 1.5 7B HF need?
Llava 1.5 7B HF requires 4.7 GB of VRAM at Q4_K_M, or 15.5 GB at FP16.
VRAM = Weights + KV Cache + Overhead
Weights = 7.1B × 4.8 bits ÷ 8 = 4.2 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 Llava 1.5 7B HF?
For Llava 1.5 7B HF, Q4_K_M (4.7 GB) offers the best balance of quality and VRAM usage. Q5_K_M (5.5 GB) provides better quality if you have the VRAM. The smallest option is Q2_K at 3.3 GB.
VRAM requirement by quantization
★ Recommended — best balance of quality and VRAM usage.
Can I run Llava 1.5 7B HF on a Mac?
Yes — Mac Mini M2 (8 GB) and 145 other Macs can run Llava 1.5 7B HF, the smallest at Q3_K_M. Apple Silicon uses unified memory, so the model shares RAM with the system. At Q4_K_M you need at least 4.7 GB of usable unified memory (RAM minus macOS overhead), from Mac Mini M2 Pro (16 GB).
Can I run Llava 1.5 7B HF locally?
Yes — Llava 1.5 7B HF can run locally on consumer hardware. Popular tools include Ollama, LM Studio, and llama.cpp.
How fast is Llava 1.5 7B HF?
At Q4_K_M, Llava 1.5 7B HF can reach ~1030 tok/s on AMD Instinct MI350X. On NVIDIA GeForce RTX 4090: ~141 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 ÷ 4.66 × 0.65 = ~1116 tok/s
Estimated speed at Q4_K_M (4.7 GB)
Real-world results typically within ±20%. Speed depends on batch size, quantization kernel, and software stack.
What's the download size of Llava 1.5 7B HF?
At Q4_K_M, the download is about 4.24 GB. The full-precision FP16 version is 14.13 GB. The smallest option (Q2_K) is 3.00 GB.
Which GPUs can run Llava 1.5 7B HF?
52 consumer GPUs can run Llava 1.5 7B HF 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 Llava 1.5 7B HF?
151 devices with unified memory can run Llava 1.5 7B HF at Q4_K_M (4.7 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.