Hy4 models
2 Hy4 models from Tencent on Hugging Face, from 780.0B to 803.5B parameters, published by Tencent in BF16, with a 1M-token context. The smallest official model, Hy4-preview, needs about 1743 GB of VRAM at its published precision.
Part of the Hunyuan series · Previous generation: Hy3
Pick a size
One row per official Hy4 size: the VRAM it needs at each precision, the cheapest GPU that holds it today, and the KV cache for a 32K-token context.
| Model | Parameters | Native VRAM | FP8 VRAM | INT4 VRAM | Live GPU fit (native) | Est. $/hr | KV cache at 32K |
|---|---|---|---|---|---|---|---|
| Hy4-preview | 780.0B | 1743 GB | 872 GB | 436 GB | No live fit | – | 2.7 GB |
VRAM is weights times a flat 1.2 overhead; the KV cache is a separate per-model figure at 16-bit, shown where the architecture is published. See the methodology.
Official models (1)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| Hy4-preview | 780.0B | BF16 | 1743 GB | No live fit | – |
Quantized versions (1)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| Hy4-preview-FP8 | 803.5B | F8_E4M3 | 898 GB | No live fit | – |
VRAM is for the precision the model is published in: the weights times a flat 1.2 overhead, with the KV cache not included. See the methodology. The fit shown is the lowest-priced single GPU type that holds the model at that precision, or the lowest-priced multi-GPU set (up to 8) when none does.
Run and fine-tune Hy4
- How to run Hy4: engine support, chat template and context flags, with a serving command sized per model.
- How to fine-tune Hy4: LoRA and QLoRA memory per model, and the license limits on derivatives.
More from Tencent
- tencent (3 models)
- Hy3 (3 models)
- Hunyuan Instruct (2 models)
- Hy-MT2 (3 models)