Ask HN: Best way to add vision support to Deepseek v4 Flash on DGX?
StartupTim · reddit · 2026-08-24
A user is running Deepseek v4 Flash 0731 on 2x DGX Sparks with 1M context and vLLM TP2 but needs vision support.
Current blockers:
- Available vision encoders either require disabling the thinking mode in vLLM or rely on sglang, which is difficult to deploy on their RDMA cluster.
- Workarounds like Image MCP and vision plugins underperform compared to native multimodal models.
The user seeks advice on the best engineering approach to enable image/vision capabilities for DSv4F without sacrificing performance.
More from coding & agent
- Vercel Goes Full Backend: Python GA, Containers, and Daemons — cramforce · 2026-10-16
- WorldMemArena Accepted at NeurIPS: Memory Storage ≠ Memory Use — xwang_lk · 2026-10-16
- Vercel Ship Demos Container Deploys for Go, Ruby, Java and More — edwin · 2026-10-16
- TRL adds day-0 support for fine-tuning Muse Spark Open's 816B MoE with LoRA on 64 H100s — QGallouedec · 2026-10-16
- Databricks engineers land impactful PRs on day 2 with AI dev tools, vs week 3 before — Yuchenj_UW · 2026-10-16
- Perplexity open-sources SPACE, the sandbox platform powering its Computer agents — AravSrinivas · 2026-10-16