license: apache-2.0 base_model: huihui-ai/Huihui-Ornith-1.5-35B-A3B-abliterated tags:

• qwen35moe
• moe
• abliterated
• uncensored
• rocmfp4
• rocmfpx
• llama.cpp-rocm
• gguf
• vision
• multimodal library_name: gguf

Ornith-1.5-35B-A3B-abliterated-ROCmFPX

ROCmFPX quantized GGUF (7 tiers + vision mmproj) of huihui-ai/Huihui-Ornith-1.5-35B-A3B-abliterated — abliterated BF16 weights, locally quantized.

⚡ Quick Start

Requires llama.cpp-rocm (this fork supports ROCmFPX; upstream llama.cpp cannot load these files — format note at the end).

# Text only (smallest tier: Q4_FAST ~17.7 GiB)
llama-server -m Ornith-1.5-35B-A3B-abliterated_Q4_0_ROCMFP4_FAST.gguf -ngl 99 -fa on --port 8080

# Vision: add --mmproj mmproj-Ornith-1.5-35B-A3B-abliterated_BF16.gguf
# All tiers fit on a single 128 GB machine (largest: 34.7 GiB)

🎯 Which tier? (3-second decision)

• Fastest: Q4_0_ROCMFP4_FAST (17.7 GiB)
• Speed + stabler quality: Q4_0_ROCMFP4_STRIX (17.8 GiB — recommended quality/speed sweet spot)
• Higher-bit reference: Q6_0_ROCMFPX (27.4 GiB)
• Agent / tool-call: Q6_0_ROCMFPX_AGENT (30.9 GiB) / Q8_0_ROCMFPX_AGENT (34.7 GiB)
• Quality first: Q8_0_ROCMFPX (34.2 GiB)

Model Overview

• Base model: Ornith-1.5-35B-A3B (35B MoE, ~3B active) abliterated (refusal behavior removed, uncensored)
• Architecture: MoE — 35.5B total / ~3B active per token, 256 experts, 40 transformer layers + 1 MTP layer
• Quantization: BF16 -> ROCmFP4 / ROCmFPx, 7 variants
• MTP: GGUFs carry an MTP head — enable with llama-server --spec-type draft-mtp (speculative decoding)
• Vision: multimodal — pair with the bundled mmproj (clip + qwen3vl_merger vision tower)
• Context: up to 256K supported by this model

Files

SHA-256 checksums for all files: see sha256.txt.

Quantization Tiers — what's here & how to choose

All files are builds of the same abliterated weights through llama-quantize (llama.cpp-rocm fork); upstream llama.cpp cannot load them (see Format Note). Quick guide:

• Q4_0_ROCMFP4_FAST — smallest & fastest baseline (4.27 bpw). Bulk weights in single-scale FP4.
• Q4_0_ROCMFP4_FAST_COHERENT — FAST + Q6_K token embeddings (better coherence, 4.30 bpw).
• Q4_0_ROCMFP4_STRIX — FAST + Q6_K embeddings + dual-scale attention K/V (quality-biased; same speed class as FAST — the recommended quality/speed sweet spot).
• Q6_0_ROCMFPX / Q8_0_ROCMFPX — higher-bit reference layouts (6.62 / 8.27 bpw). Decode is slower than the Q4 family, format characteristic.
• Q6_0_ROCMFPX_AGENT / Q8_0_ROCMFPX_AGENT — agent / tool-call oriented routing variants (attention & FFN boosts on sensitive layers).

The full type ladder lives in the llama.cpp-rocm repo (llama-quantize --help).

Benchmarks (llama-bench)

Measured on AMD Strix Halo (gfx1151), llama-bench -ngl 99 -p 256,2048 -n 256 -r 5, llama.cpp-rocm build 11107. MoE 35B / ~3B active. Each figure is the mean of the last 4 of 5 reps — the first rep carries one-time warm-up cost and is excluded.

Usage

# Text only:
llama-server -m Ornith-1.5-35B-A3B-abliterated_Q4_0_ROCMFP4_FAST.gguf -ngl 99 -fa on --port 8080

# Vision (multimodal; --mmproj enables image input):
llama-server -m Ornith-1.5-35B-A3B-abliterated_Q4_0_ROCMFP4_FAST.gguf --mmproj mmproj-Ornith-1.5-35B-A3B-abliterated_BF16.gguf -ngl 99 -fa on --port 8080

• Swap -m to any other variant file to run that variant (see Files).
• OpenAI-compatible API served at http://127.0.0.1:8080/v1.
• MTP head: load with llama-server --spec-type draft-mtp for speculative MTP acceleration.

Format Note (ROCmFPX)

ROCmFPX is a quantization format family (llama.cpp-rocm extension: GGML types 100–107 / FTYPE 100–119) — upstream llama.cpp cannot load these files. Load them with llama.cpp-rocm, a llama.cpp fork that supports all upstream GGUF formats plus ROCmFP4/ROCmFPx natively.

License & Disclaimer

• Base model: ornith-ai/Ornith-1.5-35B-A3B — MIT
• Abliterated version: huihui-ai/Huihui-Ornith-1.5-35B-A3B-abliterated — MIT
• mmproj (vision tower): clip + qwen3vl_merger vision tower of the Ornith model — Apache-2.0
• This quantized version: Apache-2.0 (ArtomYuan)

This is an abliterated (uncensored) version and may produce outputs that do not conform to safety policies. Use at your own risk. No warranty is provided.