- Newest
- Most votes
- Most comments
When deploying DeepSeek-R1 distilled Llama models with Amazon Bedrock Custom Model Import, the pricing model and deployment process are as follows:
-
Amazon Bedrock Custom Model Import doesn't launch specific machine types that you manage. Instead, it uses a fully managed infrastructure to host your custom models. The underlying hardware is abstracted away, and you're billed based on Custom Model Units (CMUs) rather than specific instance types.
-
The total cost for deploying a model like DeepSeek-R1-Distill-Llama-8B involves several components:
a) Custom Model Import: There is no charge for importing the model itself.
b) Inference costs: You are billed based on two factors:
- The number of active model copies
- The duration of their activity
Billing occurs in 5-minute windows, starting from the first successful invocation of each model copy. The pricing per model copy per minute varies based on factors including architecture, context length, region, and compute unit version.
c) Custom Model Units (CMUs): The number of CMUs required depends on the model's architecture, parameter count, and context length. For example, a Llama 3.1 8B 128K model (which is similar in size to DeepSeek-R1-Distill-Llama-8B) requires 2 Custom Model Units.
d) Storage costs: There's a monthly storage cost per Custom Model Unit.
Let's break down an example cost calculation:
Assuming you're using the US East (N. Virginia) or US West (Oregon) regions:
- Price per Custom Model Unit per minute: $0.0785
- Monthly storage cost per Custom Model Unit: $1.95
For a DeepSeek-R1-Distill-Llama-8B model (assuming it requires 2 CMUs like the Llama 3.1 8B model):
-
If the model is active for 1 hour per day: Inference cost: 2 CMUs * $0.0785 per minute * 60 minutes = $9.42 per hour Daily cost: $9.42 * 1 hour = $9.42 Monthly inference cost: $9.42 * 30 days = $282.60
-
Monthly storage cost: 2 CMUs * $1.95 = $3.90
Total estimated monthly cost: $282.60 (inference) + $3.90 (storage) = $286.50
This is a simplified example, and actual costs may vary based on usage patterns, scaling, and other factors.
Regarding deploying DeepSeek distilled models in the standard Amazon Bedrock marketplace, currently, these models are not available as pre-built options in the standard marketplace. They need to be deployed using the Custom Model Import feature, which allows you to use your own model weights within Amazon Bedrock for supported architectures, serving them alongside the hosted foundation models in a fully managed way.
Sources
Deploy DeepSeek-R1 Distilled Llama models in Amazon Bedrock | AWS Machine Learning Blog
Deploy DeepSeek-R1 distilled Llama models with Amazon Bedrock Custom Model Import | AWS Machine Learning Blog
Build Generative AI Applications with Foundation Models – Amazon Bedrock Pricing – AWS
Relevant content
- asked 2 months ago
- asked a year ago
- AWS OFFICIALUpdated 10 months ago
- AWS OFFICIALUpdated a year ago
- AWS OFFICIALUpdated 9 months ago
- AWS OFFICIALUpdated a month ago