This service offers a hosted version of the DeepSeek-R1-Distill-Qwen-1.5B model (https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B), which operates within your private cloud. After you subscribe to the listing, a CloudFormation deployment will initiate in your AWS account, setting up an EKS cluster running an inference service for the DeepSeek-R1-Distill-Qwen-1.5B model. Once the installation is complete, an API endpoint will be made available for seamless service queries!

DeepSeek-R1-Distill-Qwen-1.5B model is fine-tuned based on the open-source Qwen model, using samples generated by DeepSeek-R1. The DeepSeek team showed that the reasoning patterns discovered with reinforcement learning in a giant 671 B model can be compressed into tiny dense models without much loss. This 1.5 B checkpoint is the smallest of those distillations.

DeepSeek R1-Distill-Qwen-1.5 B punches way above its weight in math- and code-heavy reasoning while still fitting on a single laptop GPU (~4 GB in 8-bit). Use it whenever you need solid chain-of-thought performance under tight VRAM / latency budgets.

Architecture: 1.78 B-param decoder Transformer (Qwen 2.5-Math-1.5 B base) distilled from the 671 B-param DeepSeek R1 reasoning model

Context length: 32,768 tokens (inherits Qwen 2.5 long-context support)