{
  "slug": "ember-1",
  "order": 106,
  "name": "Ember-1 (2.8T)",
  "baseTitle": "Ember-1",
  "company": "Fireworks AI",
  "country": "United States",
  "flag": "🇺🇸",
  "date": "2026-09-23",
  "decoderType": "Sparse hybrid",
  "attention": "3:1 Kimi Delta Attention and Gated MLA with NoPE",
  "layerMix": "69 KDA + 24 Gated MLA",
  "scale": "2.8T total, 104B active (3.7% active)",
  "totalParams": 2800000000000,
  "activeParams": 104000000000,
  "context": "1,048,576",
  "contextTokens": 1048576,
  "license": "N/A",
  "kvCacheText": "27 KiB",
  "kvCacheKiB": 27,
  "kvCacheLevel": "Low",
  "summary": "Fireworks' Kimi K3 derivative, post-trained for shorter reasoning traces. Architecture details follow the published Kimi K3 base model; the diagram focuses on its text backbone.",
  "keyDetail": "Inherits Kimi K3's eight Block Attention Residual groups and Stable LatentMoE, with one dense layer followed by 92 MoE layers.",
  "image": {
    "full": "https://sebastianraschka.com/llm-architecture-gallery/images/architectures/ember-1.webp",
    "thumb": "https://sebastianraschka.com/llm-architecture-gallery/images/architectures/thumbnails/ember-1.webp",
    "alt": "Figure 1: Ember-1",
    "width": 960,
    "height": 884
  },
  "articleUrl": null,
  "links": [
    {
      "label": "Kimi K3 base config",
      "href": "https://huggingface.co/moonshotai/Kimi-K3/blob/main/config.json",
      "title": "moonshotai/Kimi-K3"
    },
    {
      "label": "Tech report",
      "href": "https://fireworks.ai/blog/ember-1",
      "title": null
    }
  ],
  "concepts": [
    {
      "slug": "nope",
      "label": "NoPE",
      "name": "No Positional Embeddings",
      "url": "https://llm-atlas-7or.pages.dev/concepts/nope/"
    },
    {
      "slug": "mla",
      "label": "MLA",
      "name": "Multi-Head Latent Attention",
      "url": "https://llm-atlas-7or.pages.dev/concepts/mla/"
    },
    {
      "slug": "gated-attention",
      "label": "Gated Attention",
      "name": "Gated Attention",
      "url": "https://llm-atlas-7or.pages.dev/concepts/gated-attention/"
    },
    {
      "slug": "hybrid-attention",
      "label": "Hybrid Attention",
      "name": "Hybrid Attention",
      "url": "https://llm-atlas-7or.pages.dev/concepts/hybrid-attention/"
    },
    {
      "slug": "shortconv",
      "label": "ShortConv",
      "name": "Short Convolution",
      "url": "https://llm-atlas-7or.pages.dev/concepts/shortconv/"
    },
    {
      "slug": "attention-residuals",
      "label": "Attention Residuals",
      "name": "Attention Residuals",
      "url": "https://llm-atlas-7or.pages.dev/concepts/attention-residuals/"
    },
    {
      "slug": "latent-moe",
      "label": "LatentMoE",
      "name": "Latent MoE",
      "url": "https://llm-atlas-7or.pages.dev/concepts/latent-moe/"
    },
    {
      "slug": "moe",
      "label": "MoE",
      "name": "Mixture of Experts",
      "url": "https://llm-atlas-7or.pages.dev/concepts/moe/"
    }
  ],
  "aai": null,
  "sourceUrl": "https://sebastianraschka.com/llm-architecture-gallery/#card-ember-1",
  "family": "hybrid-moe",
  "url": "https://llm-atlas-7or.pages.dev/models/ember-1/",
  "markdownUrl": "https://llm-atlas-7or.pages.dev/models/ember-1.md"
}