Falcon models
4 Falcon models on Hugging Face, from 6.9B to 41.8B parameters. At the precision each one is published in, the smallest needs about 16.1 GB of VRAM (falcon-7b, cheapest live fit: RTX A5000) and the largest about 93.5 GB (falcon-40b, cheapest live fit: RTX PRO 6000). The cheapest way to run falcon-7b is $0.176/hr.
Falcon models
| Model | Parameters | Published as | VRAM needed | Live GPU fit | Est. $/hr |
|---|---|---|---|---|---|
| falcon-7b-instruct-sharded | 6.9B | F32 | 30.9 GB | V100 | $0.187/hr |
| falcon-7b | 7.2B | BF16 | 16.1 GB | RTX A5000 | $0.176/hr |
| falcon-7b-instruct | 7.2B | BF16 | 16.1 GB | RTX A5000 | $0.176/hr |
| falcon-40b | 41.8B | BF16 | 93.5 GB | RTX PRO 6000 | $1.38/hr |
VRAM is for the precision the model is published in, with the same overhead and an 8,192-token context assumed on every page; 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.
