StarCoder2 models
2 StarCoder2 models from bigcode on Hugging Face, from 3.0B to 7.2B parameters, published by bigcode in F32 and BF16, with a 16K-token context. The smallest official model, starcoder2-3b, needs about 13.5 GB of VRAM at its published precision; the cheapest live fit is V100 at $0.110/hr.
Pick a size
One row per official StarCoder2 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 |
|---|---|---|---|---|---|---|---|
| starcoder2-3b | 3.0B | 13.5 GB | 3.4 GB | 1.7 GB | V100 | $0.110/hr | not published for this architecture |
| starcoder2-7b | 7.2B | 16.0 GB | 8.0 GB | 4.0 GB | RTX A5000 | $0.176/hr | not published for this architecture |
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 (2)
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| starcoder2-3b | 3.0B | F32 | 13.5 GB | V100 | $0.110/hr |
| starcoder2-7b | 7.2B | BF16 | 16.0 GB | RTX A5000 | $0.176/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.
More from bigcode
- bigcode (1 models)