---

license: other

license_name: muse-model-license

pipeline_tag: text-generation

tags:

• gguf
• moe
• text-generation

base_model: meta-models/Muse-Spark-Open-816B-A42B

---

Muse Spark Open 816B — Interchained GGUF

Welcome. This is Meta's Muse Spark Open 816B (816B parameters, 42B active), converted to GGUF and quantized by Interchained. Every file here is cut from official weights, and we're glad you found us.

Built with Muse Spark Open.

What you're getting

Official meta-models/Muse-Spark-Open-816B-A42B weights, pulled gated with our own access, converted with Hearth and quantized with stock llama-quantize. No third-party sources, no weight surgery. What Meta shipped is what's in these files, and we document every step so you can verify that yourself.

| File | Type | Size | Status |

|------|------|------|--------|

| Interchained-Muse-Spark-Open-816B-A42B-F16-00001-of-00005.gguf … -00005-of-00005.gguf | F16 master (5 shards) | ~330 GB each | ✅ Live |

| Interchained-Muse-Spark-Open-816B-A42B-Q4_0.gguf | Q4_0 | ~450 GB | ✅ Live |

| Interchained-Muse-Spark-Open-816B-A42B-Q4_K_M-00001-of-00003.gguf … -00003-of-00003.gguf | Q4_K_M (3 shards) | ~249 GB ×2 + 2.4 GB | ✅ Live |

| Interchained-Muse-Spark-Open-816B-A42B-Q5_K_M-00001-of-00002.gguf … -00002-of-00002.gguf | Q5_K_M (2 shards) | ~293 GB ×2 | ✅ Live |

| Interchained-Muse-Spark-Open-816B-A42B-Q6_K | Q6_K | ~740 GB | 🔨 Coming soon |

The F16 master is the source of truth behind every quant: one master, many quants, zero drift. If you want maximum fidelity, start with the F16. If you want to actually run it, the quants are the practical path.

> Note: split GGUFs load transparently. Point llama.cpp at the first shard (-00001-of-00005.gguf) and it pulls in the rest automatically, same as a single file.

Provenance

• Source weights: meta-models/Muse-Spark-Open-816B-A42B (gated, official)
• Downloaded: 2026-10-16. 78 safetensors files, 1.64 TB, verified complete
• Converted by: Interchained via Hearth (hearth convert). 1,429 tensors, F16
• Quantized by: Interchained via stock llama-quantize (Hearth managed runtime)
• License: Muse Model License. Please see LICENSE.md and USE_POLICY.md

Running it

This is a datacenter-class model. The F16 master wants serious hardware; the quants are the practical way in. Load Q4_0 or Q4_K_M in llama.cpp, Ollama, LM Studio, or anything that speaks GGUF:

llama-cli -m Interchained-Muse-Spark-Open-816B-A42B-Q4_0.gguf -p "Hello, giant."

Text-only for now. No vision projector yet.

Talk to us

We'd genuinely like to hear from you. Whether you're running these files, hitting an issue, want a different quant, or just want to talk shop, come say hello:

• Website: interchained.org
• Email: dev@interchained.org or interchained@gmail.com
• Telegram: t.me/interchained

If you build something with these models, we'd love to hear about it. If something's wrong, we want to know that too. This is a front door, not a wall.

---

Built by Interchained. Clean conversions, no shortcuts.