GPT-NeoX models
2 GPT-NeoX models from EleutherAI on Hugging Face, from 204M to 20.7B parameters, published by EleutherAI in F16, with a 2K-token context. The smallest official model, gpt-neox-20b, needs about 46.4 GB of VRAM at its published precision; the cheapest live fit is RTX A6000 at $0.363/hr.
Pick a size
One row per official GPT-NeoX 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 |
|---|---|---|---|---|---|---|---|
| gpt-neox-20b | 20.7B | 46.4 GB | 23.2 GB | 11.6 GB | RTX A6000 | $0.363/hr | 33.0 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 |
|---|---|---|---|---|---|
| gpt-neox-20b | 20.7B | F16 | 46.4 GB | RTX A6000 | $0.363/hr |
Fine-tunes and community models (1)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| japanese-gpt-neox-small | 204M | F32 | 0.9 GB | RTX 3070 | $0.088/hr |
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.