i have created a simple way to try the new LLM called DeepSeek V3.
here is a picture of the blocks:
you have to obtain your own deepseek v3 api key. start from https://platform.deepseek.com/ and look for api key. (Note: you do have to pay for it but during this introductory period the rate is a fraction of the normal rate! $0.014 vs $0.07 per 1M input tokens!)
here’s the program
demo_deepseek.aia (3.8 KB)
and here’s a sample run (enter your request in first box, click button, response goes to text box below)
2 Likes
Hi, how can I abort the request if its taking too long for deepseek to response. For example, I only set 40 seconds. If there is no response within that, the request will be aborted. Thank you.
use the timeout option see highlighted parts in this snapshot
here’s a sample screenshot of output - you get a momentary error message that flashes but you do have the ability to catch the error (see textbox after the button)
here’s the updated aia
demo_deepseek.aia (4.1 KB)
Hi there again,
Deepseek just launched V3-0324 version. I changed the model to deepseek-chat-v3-0324 and url to https://api.deepseek.com/v1/chat/completions but I am having this error:
Segment: text length is negative six
When I tried to revert from the original blocks, with no changes, everything is fine. I just wanna try the new version and check if there is an improvement.
Thank you.
go to this bottom of this page:
(api usage has not changed )
i believe you can change the model to deepseek-reasoner as an alternative if you want to see the reasonings etc.
there are many spaces to try in huggingface but i haven’t figured out how to translate those to equivalent niotron flows, yet.
1 Like
