---

license: mit

base_model: ornith-ai/Ornith-1.5-35B-A3B

base_model_relation: quantized

tags:

• gguf
• llama.cpp
• reap
• expert-pruning
• moe
• qwen3.5
• text-generation

language:

• en

pipeline_tag: text-generation

quantized_by: Ttimms

license_link: https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B/blob/main/LICENSE

---

Ornith-1.5-35B-A3B · REAP-50 — GGUF

GGUF quants of the 50%-REAP-pruned Ornith-1.5-35B-A3B (Gated-DeltaNet + MoE

hybrid): 256 → 128 experts, MTP head removed, ~19B parameters. Likely the

first GGUF of any 50%-REAP-pruned model.

• Flagship (NVFP4A16, vLLM): Ttimms/Ornith-1.5-35B-A3B-REAP-50-NVFP4A16
• Pruned bf16 source: Ttimms/Ornith-1.5-35B-A3B-REAP-50-bf16
• Pipeline + benchmarks: <https://github.com/t-timms/ornith-nvfp4>

Base model quality (NVFP4A16, before GGUF quant): HumanEval+ 84.2%, MBPP+ 89.2% — single greedy draws; see the NVFP4A16 card on run-to-run spread.

Architecture

graph TD
    Base["ornith-ai/Ornith-1.5-35B-A3B<br/>qwen3_5_moe - 256 experts - MTP head - ~65 GB bf16 - MIT"]
    subgraph Build ["Build - RTX 5070 Ti, SM120"]
        REAP["REAP expert prune 50%  256 -> 128"]
        MTP["strip MTP head (1 -> 0, reload-verified)"]
        StripV["strip vision tower (333 tensors)"]
        Quant["GPTQ-NVFP4A16 quantize (~5.5 h, 12.47 GiB)"]
    end
    subgraph HF ["Published formats"]
        A16["REAP-50-NVFP4A16 - 12.47 GiB (vLLM)"]
        GGUF["REAP-50-GGUF (Q4_K_M / Q5_K_M / Q6_K / Q8_0)"]
        BF16["REAP-50-bf16 (pruned source)"]
    end
    Bench["HumanEval+ 84.2% - MBPP+ 89.2% - SWE-bench Verified 44.0%"]
    Base --> REAP --> MTP --> StripV --> Quant --> A16
    StripV --> BF16
    BF16 -. convert .-> GGUF
    A16 --> Bench

Download

Grab one file, not the whole repo.

| File | Quant | Size | Notes |

|---|---|--:|---|

| ornith-reap50-Q4_K_M.gguf | Q4_K_M | 11.4 GB | recommended — fits 16 GB with room for context |

| ornith-reap50-Q5_K_M.gguf | Q5_K_M | 13.3 GB | higher quality, tighter KV budget |

| ornith-reap50-Q6_K.gguf | Q6_K | 15.3 GB | near-lossless; little headroom on 16 GB |

| ornith-reap50-Q8_0.gguf | Q8_0 | 19.7 GB | needs >16 GB or CPU offload |

MoE with ~3B active parameters — CPU / Apple-Silicon inference is fast for the size.

For full precision, re-quantizing, or merging, use the

pruned bf16 source

and run convert_hf_to_gguf.py --no-mtp yourself.

Run it in

• llama.cpp — llama-server -m <file>.gguf -ngl 99
• LM Studio (bundles a recent llama.cpp)
• koboldcpp
• Jan
• text-generation-webui
• Ollama — once its bundled llama.cpp includes this arch (see Requirements)

Requirements

Needs a recent llama.cpp (the qwen35moe hybrid arch). Verified with

llama-server from a source build. Ollama support pending its bundled-llama.cpp

bump.

llama-server -m ornith-reap50-Q4_K_M.gguf -ngl 99 --port 8080

The chat template defaults to thinking on; pass

"chat_template_kwargs": {"enable_thinking": false} for grounded output.

License

MIT (inherits from Ornith-1.5-35B-A3B). Pruning: REAP

(github.com/CerebrasResearch/reap, with a router-renormalization fix).