---

license: mit

pipeline_tag: text-generation

base_model: deepreinforce-ai/Ornith-1.0-397B-FP8

tags:

• gguf
• long-context
• yarn
• ornith
• llama.cpp
• ollama

---

<div align="center">

<img src="banner.jpeg" alt="Ornith" width="720"/>

Ornith-1.0-397B: 1M Context GGUF (YaRN)




![Context: 1M]()

</div>

All GGUF files live on ModelScope

Download here: modelscope.ai/models/satgeze/Ornith-1.0-397B-1M-GGUF

This Hugging Face repo is a pointer page. The 397B quants are large (426 GB across IQ1_M, IQ2_M, and Q4_K_M splits), and the full set is hosted and verified on the ModelScope mirror to keep this account inside storage limits. Nothing was discontinued; the files are simply one click away.

Every file, every mirror

Nothing was discontinued: every quant is one click away. Hugging Face carries the curated picks, ModelScope always carries everything, and Ollama serves ready-to-run tags.

| File | Size | Hugging Face | ModelScope | Ollama |

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

| ornith-1.0-397b-1M-IQ1_M-00001-of-00003.gguf | 39.8 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-IQ1_M-00002-of-00003.gguf | 39.5 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-IQ1_M-00003-of-00003.gguf | 12.1 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-IQ2_M-00001-of-00004.gguf | 39.3 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-IQ2_M-00002-of-00004.gguf | 40.0 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-IQ2_M-00003-of-00004.gguf | 39.6 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-IQ2_M-00004-of-00004.gguf | 14.3 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-Q4_K_M-00001-of-00007.gguf | 39.0 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-Q4_K_M-00002-of-00007.gguf | 39.5 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-Q4_K_M-00003-of-00007.gguf | 38.9 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-Q4_K_M-00004-of-00007.gguf | 39.0 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-Q4_K_M-00005-of-00007.gguf | 39.4 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-Q4_K_M-00006-of-00007.gguf | 39.3 GB | on ModelScope | download | - |

| ornith-1.0-397b-1M-Q4_K_M-00007-of-00007.gguf | 6.8 GB | on ModelScope | download | - |

What the ModelScope repo contains

DeepReinforce's Ornith-1.0-397B (397B MoE, Qwen3.5 family) with YaRN RoPE scaling baked into GGUF metadata for a 1,048,576-token context window, 4x the native 262,144. Weights bit-identical to the original; only rope metadata differs.

| Quant | Size | Runs on |

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

| IQ1_M (3 split files) | 91 GB | 128 GB Mac (tested: it runs) |

| IQ2_M (4 split files) | 132 GB | 192 GB+ unified or multi-GPU |

| Q4_K_M (7 split files) | 241 GB | Server class, 256 GB+ |

Needle certification for this size class is pending a GPU-pod session; the 9B and 35B repos carry fully certified ladders of the same recipe. MTP note: the official 397B checkpoint includes an MTP layer (config mtp_num_hidden_layers: 1) that GGUF conversions dropped; restoring it by extraction is on the roadmap.

Run it

llama-server -m ornith-1.0-397b-1M-IQ1_M-00001-of-00003.gguf -c 262144 -np 1 --jinja

llama.cpp loads split GGUFs automatically from the first part. Start at 262K and raise context per available memory; KV runs about 30 KB per token (15 of 60 layers full attention).

How to actually use a 1M-context model

Habits that measurably help, from our RULER, hop and adherence testing across this fleet:

1. Re-state standing instructions near the end of long prompts; recency beats depth.
2. One big reference dump beats a long accumulated conversation. Fresh session per task.
3. After any compaction or summarization, repeat your active rules yourself.
4. Prefill at 500K+ takes real time on any hardware; stage your questions accordingly.
5. Know your quant: the results tables on this card show what each quant actually holds at depth; pick the strongest one your memory allows.

Credits

Ornith-1.0: DeepReinforce (MIT), post-trained on Qwen 3.5. 1M YaRN extension: SatGeze. Method and harness: github.com/satindergrewal/aviary-1m.

Sister repos with full ladders and MTP: 9B | 35B

---

Repo history note: the commit history was squashed (July 2026) to reclaim storage from deleted large files, since Hugging Face bills every revision including deleted ones. The real build history of this repo is longer than the log shows. Method, tooling, and full records: aviary-1m.