Ornith-1.0-35B: 1M Context + MTP + Vision

Mirrors: Hugging Face | ModelScope (full quant ladder always available on ModelScope)

DeepReinforce's Ornith-1.0-35B (35B MoE, 3B active, Qwen3.5 family) with YaRN RoPE scaling baked into the GGUF metadata for a 1,048,576-token context window, needle-certified through the full range, now shipping MTP speculative decoding and a vision tower. Weights are bit-identical to the source builds; only rope metadata differs, so llama.cpp and Ollama apply the extension with no extra flags.

Capability	Status	Evidence in this repo
1M context	Certified	50/50 needles through 1M, f16 KV (heatmap + results.jsonl)
MTP speculative decoding	Two vetted builds	Q6_K: 267.9 to 382.9 tok/s (+43%). APEX-Compact: +14%, needle-perfect 70/70 through 1M (524K on 5090, 786K/1M on M3 Max)
Vision	Verified	mmproj tower reads image text and identifies objects correctly

Files: MTP-first

File	Size	Pick it when
`ornith-1.0-35b-1M-MTP-Q6_K.gguf`	29.2 GB	You have the memory (64 GB Mac, RTX Pro 6000). Fastest: +43% decode, Q6 quality
`ornith-1.0-35b-1M-MTP-APEX-Compact.gguf`	17.0 GB	You want 1M-class context on a 32 GB card. SC117 APEX imatrix quant, needle-perfect to 524K on an RTX 5090
`mmproj-ornith-35b-f16.gguf`	899 MB	Vision, attach with `--mmproj`

This Hugging Face repo intentionally carries only the MTP builds. The full 10-quant ladder (IQ2_M through bf16, every quant needle-tested) lives on the ModelScope mirror.

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.

On Ollama every tag ships with the vision tower bundled and the 1M rope metadata baked in.

File	Size	Hugging Face	ModelScope	Ollama
`mmproj-ornith-35b-f16.gguf`	899 MB	download	download	bundled in every tag
`ornith-1.0-35b-1M-IQ2_M.gguf`	11.7 GB	on ModelScope	download	`ollama run satgeze/ornith-35b-1m:iq2_m`
`ornith-1.0-35b-1M-IQ3_XXS.gguf`	13.6 GB	on ModelScope	download	`ollama run satgeze/ornith-35b-1m:iq3_xxs`
`ornith-1.0-35b-1M-IQ4_XS.gguf`	18.7 GB	on ModelScope	download	`ollama run satgeze/ornith-35b-1m:iq4_xs`
`ornith-1.0-35b-1M-MTP-APEX-Compact.gguf`	17.0 GB	download	download	`ollama run satgeze/ornith-35b-1m`
`ornith-1.0-35b-1M-MTP-Q6_K.gguf`	29.2 GB	download	download	`ollama run satgeze/ornith-35b-1m:q6_k-mtp`
`ornith-1.0-35b-1M-Q2_K.gguf`	12.9 GB	on ModelScope	download	`ollama run satgeze/ornith-35b-1m:q2_k`
`ornith-1.0-35b-1M-Q3_K_M.gguf`	16.8 GB	on ModelScope	download	`ollama run satgeze/ornith-35b-1m:q3_k_m`
`ornith-1.0-35b-1M-Q4_K_M.gguf`	21.2 GB	on ModelScope	download	`ollama run satgeze/ornith-35b-1m:q4_k_m`
`ornith-1.0-35b-1M-Q5_K_M.gguf`	24.7 GB	on ModelScope	download	`ollama run satgeze/ornith-35b-1m:q5_k_m`
`ornith-1.0-35b-1M-Q6_K.gguf`	28.5 GB	on ModelScope	download	`ollama run satgeze/ornith-35b-1m:q6_k`
`ornith-1.0-35b-1M-Q8_0.gguf`	36.9 GB	on ModelScope	download	`ollama run satgeze/ornith-35b-1m:q8_0`
`ornith-1.0-35b-1M-bf16.gguf`	69.4 GB	on ModelScope	download	`ollama run satgeze/ornith-35b-1m:bf16`

Needle-in-a-haystack certification

Native 262,144, YaRN factor 4 to 1,048,576: 10/10 needles at every rung through 1M, f16 KV, temperature 0 (results.jsonl). The APEX-Compact build was ladder-tested separately: 50/50 from 64K through 524K on the RTX 5090 (results-apex.jsonl); its 786K and 1M rungs are running on a 128 GB Mac at publish time and this card will update when they land.

MTP: same output, more tokens per second

Ornith's base family (Qwen3.5) ships a multi-token-prediction layer in the official checkpoints; Ornith's own GGUF releases dropped it, and community builds by wang-yang (Q6_K) and SC117 (APEX) restored it. The trunk verifies every drafted token, so output is identical to standard decoding, only faster.

Measured here (RTX 5090, 16K ctx): Q6_K 267.9 to 382.9 tok/s (+43 percent). APEX-Compact gains +14 percent; its custom imatrix shifts token probabilities away from the draft head, lowering acceptance, and its strength is instead the 17 GB footprint. Both builds were re-baked to 1M, coherence-gated, and NIAH-spot-checked here before shipping.

llama-server -m ornith-1.0-35b-1M-MTP-Q6_K.gguf \
  -c 1048576 -np 1 --jinja \
  --spec-type draft-mtp --spec-draft-n-max 3 \
  --mmproj mmproj-ornith-35b-f16.gguf

Ollama (1M and vision work; Ollama has no speculative decoding yet, so MTP adds no speed there):

FROM ./ornith-1.0-35b-1M-MTP-Q6_K.gguf
RENDERER qwen3.5
PARSER qwen3.5
PARAMETER num_ctx 262144

How this was built

1M context: YaRN rope-scaling metadata (factor 4.0, original context 262,144) written into the GGUF header with gguf-py. No weight changes, no fine-tuning.
MTP: adopted the community GGUFs carrying the official Qwen3.5 MTP layer, verified provenance and license, re-baked 1M metadata, then gated: coherence check, speed A/B, needle spot-checks with speculation active.
Certification: multi-needle harness (10 needles, depths 5 to 95 percent, temperature 0, seeded prompts) against llama-server, f16 KV only. The harness ships in this repo (niah_test.py) along with prefill timing data (prefill_timing.jsonl).

For base-model capability benchmarks see the official Ornith-1.0-35B card; weights here are bit-identical apart from rope metadata, so base capability carries over modulo quantization.

Credits

Ornith-1.0: DeepReinforce (MIT), post-trained on Qwen 3.5. MTP layer: Qwen. MTP GGUF builds: wang-yang and SC117. 1M YaRN extension, vetting, and certification: SatGeze.