Yes. The RTX A4000 runs ZwZ2-4B fully in VRAM at FP16 / BF16 (about 9.7 GB of 16 GB), at roughly 27 tokens/s.
- best quant
- FP16 / BF16
- needs
- 9.7 GB
- usable
- 16 GB
- est. speed
- ~27 tok/s
Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the ZwZ2-4B page →
Every quantisation of ZwZ2-4B on the RTX A4000
| Quant | Memory | Fit | Speed |
|---|---|---|---|
| Q2_K | 3.2 GB | fits | ~165 tok/s |
| Q3_K_M | 3.5 GB | fits | ~111 tok/s |
| Q4_K_M | 4.2 GB | fits | ~90 tok/s |
| Q5_K_M | 4.7 GB | fits | ~76 tok/s |
| Q6_K | 5.1 GB | fits | ~66 tok/s |
| Q8_0 | 5.9 GB | fits | ~51 tok/s |
| FP16 / BF16 pick | 9.7 GB | fits | ~27 tok/s |
Speed is a bandwidth-based estimate for generating text (448 GB/s); prompt processing and long contexts are slower.
Measured speed on the RTX A4000
No community reports for this pair yet.
Report your own speed →Run ZwZ2-4B on the RTX A4000
Using FP16 / BF16, the best fit for RTX A4000.
ollama run hf.co/inclusionAI/ZwZ2-4B-GGUF:F16 llama-server -hf inclusionAI/ZwZ2-4B-GGUF:F16Downloads the GGUF on first run. Installed through llama.app? The same flags work with `llama serve`.
lms get inclusionAI/ZwZ2-4B-GGUF@F16Then load it from the LM Studio app or with `lms load`.
vllm serve inclusionAI/ZwZ2-4Bservices:
vllm:
image: vllm/vllm-openai:latest
command: inclusionAI/ZwZ2-4B
ports:
- "8000:8000"
ipc: host
volumes:
- ~/.cache/huggingface:/root/.cache/huggingface
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]API on port 8000.
sglang serve inclusionAI/ZwZ2-4B --host 0.0.0.0 --port 30000services:
sglang:
image: lmsysorg/sglang:latest
command: sglang serve inclusionAI/ZwZ2-4B --host 0.0.0.0 --port 30000
ports:
- "30000:30000"
ipc: host
volumes:
- ~/.cache/huggingface:/root/.cache/huggingface
deploy:
resources:
reservations:
devices:
- driver: nvidia
count: all
capabilities: [gpu]API on port 30000.
from transformers import pipeline
pipe = pipeline("image-text-to-text", model="inclusionAI/ZwZ2-4B", dtype="auto", device_map="auto")
messages = [
{
"role": "user",
"content": [
{"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
{"type": "text", "text": "Describe this image."},
],
}
]
out = pipe(text=messages, max_new_tokens=256)
print(out[0]["generated_text"][-1]["content"])Vision models that fit the RTX A4000 comfortably
All that fit →| Model | Params | Best quant | Memory | Speed |
|---|---|---|---|---|
| 4.7B | Q8_0 | ≈5.9 GBFits in memory | ~50 tok/s | |
| 4.3B | Q8_0 | ≈5.2 GBFits in memory | ~54 tok/s | |
| 4.4B | Q8_0 | ≈5.8 GBFits in memory | ~52 tok/s | |
| 4.3B | Q8_0 | ≈5.6 GBFits in memory | ~54 tok/s | |
| 3.8B | Q8_0 | ≈4.0 GBFits in memory | ~62 tok/s | |
| 5B | Q8_0 | ≈5.6 GBFits in memory | ~47 tok/s |
Other GPUs that run ZwZ2-4B
Frequently asked questions
- Can the RTX A4000 run ZwZ2-4B?
- Yes. The RTX A4000 runs ZwZ2-4B fully in VRAM at FP16 / BF16 (about 9.7 GB of 16 GB), at roughly 27 tokens/s.
- How fast does ZwZ2-4B run on the RTX A4000?
- An estimate from memory bandwidth gives about 27 tokens/s at FP16 / BF16.
- Which quantisation of ZwZ2-4B should I use on the RTX A4000?
- FP16 / BF16: the largest one that fits. Expected quality: lossless compared with the full-precision weights.
Memory and speed are estimates from the model's size and the card's memory bandwidth; real numbers depend on the runtime, drivers and context length. Community figures are medians of user reports.