Launching today
deepseek-private-converter
OpenAI-DeepSeek private in 10min,96% cache, 10-month payback
1 follower
OpenAI-DeepSeek private in 10min,96% cache, 10-month payback
1 follower
OpenAI ā DeepSeek private in 10min. 96% Trie cache, $38.08/GPU-h, 10-month payback. MIT, Huawei 910B compatible.
Free Options
Launch Team / Built With
Customer.ioAutomate Messaging Everywhere ā Startups Get 12 Months Free
Promoted
Hey PH! š Sa here.
I was paying $38.08/GPU-hour and burning cash on OpenAI bills.
So I built a 1-file converter to move our whole stack to private DeepSeek:
openai_to_deepseek.py - 10 minutes to switch
What it does:
ā 96% Trie cache hit: 77K prompt = 73.9K cached, so 1 req/s ā 25 req/s
ā Real $38.08/GPU-h cost, 10-month payback math inside
ā MIT licensed, works on both NVIDIA and Huawei Ascend 910B
ā Includes customer_service.py example - we run our bot on it
Why private?
No data leaves our VPC
Fixed cost, no per-token surprise
25x throughput with same GPU
Repo is fully open: https://github.com/sagi668899-ops/deepseek-private-converter
CI is green, docs + economics breakdown included.
Try it: python openai_to_deepseek.py --help
Would love your feedback - what's stopping you from going private?