Fireworks AI adds serverless training for DeepSeek V4.1 Flash, Muse Spark Open and Nemotron 3.5 Super VL
sophiamyang · x · 2026-10-19
Fireworks AI rolled out several Training platform updates:
- Muse Spark Open is now available via the Serverless Training API
- DeepSeek V4.1 Flash 1017 replaces the prior version on serverless training
- Nemotron 3.5 Super VL supports Serverless, Dedicated and Managed training, including full-parameter runs
Per the docs, managed training, the Training API and serverless training draw from the same base model catalog, with per-model availability across SFT, DPO and RL methods and LoRA/full-parameter modes. Muse Glimmer 30B SFT is free via UI and Serverless Training API until 11/30. The platform also documents dedicated dataset schemas and launch flows for vision-language model SFT.
More from Infra
- Local LLMs on a 96GB M3 Ultra: Qwen3.8-27B Q8_0 Wins for Coding, With Full Config Shared — FanDiscombobulated38 · 2026-10-20
- Hugging Face's Local AI 202 shows massive speed-ups with speculative decoding — huggingface · 2026-10-20
- Open-weights Muse Spark on B300 costs $0.12 vs $0.88 per correct task, 7.1x cheaper — ryanshrout · 2026-10-20
- HARNESS: pure-Rust LLM inference engine claims 70B models on 8GB VRAM — 9gxa05s8fa8sh · 2026-10-20
- Running Strata on Two Ancient Quadro M6000s: 34.5 tok/s Decode After Lifting the Arch Restriction — jmuff98 · 2026-10-20
- Polymarket puts 28% odds on a US state enacting a data center moratorium this year — Polymarket · 2026-10-20