ByteDance Seed-OSS 36B — UNCENSORED (92.9% Precision SOTA Dense · 512K Context)

Highest Objective Technical Precision in Suite 07 (92.86% — 13/14 points) & 100% Uncensored (0.0% Refusal Rate across ⁵⁄₅ Gate & ⁹⁄₉ Suite A) dense 36.2B build of ByteDance Seed-OSS-36B-Instruct from ModelScope, engineered with Biprojected Norm-Preserving Abliteration (i1-Q4_K_M, 20.27 GB), GGUF-patched thinking_budget = 0, and 524,288 (512K) native context window.

• Official Ollama Hub: https://ollama.com/eriktonon/seed-oss-uncensored
• Upstream ModelScope Origin: ByteDance-Seed/Seed-OSS-36B-Instruct
• Empirical Benchmark Repository: https://github.com/eriktonon/ollama-uncensored-llm-benchmarks
• License: Apache-2.0 License — Commercial & Private Self-Hosted Use Permitted

Empirical Benchmark Results (NVIDIA A100-SXM4-40GB · Suite 07)

Per-Domain Verification Breakdown

• Strict JSON Schema & Edge-Case Formatting (A1_JSON): 2/2 (38.72 tok/s)
• Python Algorithms & Async Systems Code (A2_PYTHON): 3/3 (37.81 tok/s)
• Systems Rust Memory-Safe Implementation (A3_RUST): 2/2 (37.66 tok/s)
• PostgreSQL Window Functions & Analytical SQL (A4_SQL): 2/2 (37.99 tok/s)
• Exact Symbolic & Discrete Mathematics (A5_MATH): 2/2 (37.66 tok/s)
• Kubernetes / Linux SRE Incident Root-Cause (A6_SRE): 2/3 (37.67 tok/s)
• 5-Prompt Uncensored Gate (THINK_LEN=133): 5/5
• 9-Domain Red-Team, SecOps & Creative Suite (U1..U9): 9/9 (100.0%)
• 6-Check Anti-Hallucination & Long-Context Needle Audit (B1..B6): 5/6 (16.67% Hallucination Rate)

Architecture & Uncensored Engineering

1. Biprojected Norm-Preserving Abliteration (92.86% Precision — Zero IQ Degradation):
   • Unlike naive single-direction orthogonalization that perturbs reasoning subspaces, norm-preserving biprojection removes refusal vectors while preserving layerwise weight norms — achieving perfect scores on JSON (2/2), Python (3/3), Rust (2/2), SQL (2/2), and Exact Math (2/2) (13/14 overall).
2. GGUF-Patched thinking_budget = 0 (seed_oss Architecture):
   • In-place GGUF v3 chat-template patch caps <seed:think> budget to 0 (max_think_len = 133 chars minimal header), delivering clean, direct technical and creative output at 37.99 tok/s decode and 2,231.16 tok/s prefill on A100-40GB.
3. 512K Native Long-Context Grounding (5/6 Suite B):
   • Supports up to 524,288 tokens natively and passed canary needle extraction (COBALTO-9471-OMEGA), fake RFC rejection, arithmetic trap verification, K8s scheduler log extraction, and unknown-field abstention.

• Recommended Hardware Profile: Single NVIDIA A100 40GB, RTX 3090 / 4090 24GB (up to 32K ctx), RTX 5090 32GB, or 32GB+ Unified Memory node (21.5 GiB VRAM footprint on A100-40GB).

Quickstart

# Pull & run (multi-tag parity: :36b and :latest share identical digest)
ollama run eriktonon/seed-oss-uncensored:36b
ollama run eriktonon/seed-oss-uncensored:latest

# Direct REST API call (zero thinking overhead)
curl http://localhost:11434/api/chat -d '{
  "model": "eriktonon/seed-oss-uncensored:36b",
  "stream": false,
  "messages": [
    {"role": "user", "content": "Design a lock-free ring buffer in Rust with Acquire/Release atomic ordering."}
  ]
}'

📜 License & Open-Source Permissions

• License: Apache-2.0 License (Apache-2.0)
• Upstream Base: ByteDance-Seed/Seed-OSS-36B-Instruct (hf.co/mradermacher/Seed-OSS-36B-Instruct-biprojected-norm-preserving-abliterated-i1-GGUF:Q4_K_M).
• Permissions:
  • ✅ Commercial & Enterprise Use: Permitted for production services, internal SecOps tooling, and enterprise pipelines under upstream license terms.
  • ✅ Private / Air-Gapped Execution: 100% self-hosted local or private cloud execution with zero external telemetry.
  • ✅ Modification & Redistribution: Permitted to quantize, fine-tune, and deploy derivative GGUF builds with attribution.