deepseek-v4.1POST

deepseek-recipe serves V4.1 Flash 1017 locally via vLLM on dual DGX Spark

The open-source deepseek-recipe project runs V4.1 Flash 1017 locally with vLLM on dual DGX Spark, claiming API-identical prompt tokens and fixed thinking, effort, JSON schema and tools without speed loss.

Sufyan

@sfxnzArchived 9 days agoOriginally posted 2026.10.23

Sufyan @sfxnzOriginally posted 2026.10.23

Open original on X

DeepSeek V4.1 Flash 1017 now handles requests like the official API. deepseek-recipe now fronts vLLM: • prompts token identical to the API • thinking, effort, JSON schema, tools fixed • Chat, Anthropic & Responses API's • no speed cost Local. Free https://github.com/sfxnz/DeepSeek-V4.1-Flash-1017-EXL3-2.7bpw-vLLM-2x-DGX-Spark

x.com/sfxnz/status/2113305564323197286Archived 2026.10.23 12:08