47 Downloads Updated 18 hours ago
NVIDIA Nemotron 3.5 Bolt 74B-A9B Uncensored Hybrid Mamba-2 + MoE (74B total, 9B active, 22.0GB + GGUF <think> patch). 142.1 tok/s Decode, 6932 tok/s Prefill, 78.6% Precision, 0.0% Hallucination (6/6), 100% Zero-Refusal. License: NVIDIA Open / Apache-2.0.
vision
tools
thinking
74b
ollama run eriktonon/nemotron3.5-bolt-uncensored
Applications
Claude Code
ollama launch claude --model eriktonon/nemotron3.5-bolt-uncensored
OpenCode
ollama launch opencode --model eriktonon/nemotron3.5-bolt-uncensored
Hermes Agent
ollama launch hermes --model eriktonon/nemotron3.5-bolt-uncensored
OpenClaw
ollama launch openclaw --model eriktonon/nemotron3.5-bolt-uncensored
Models
View all →Readme
NVIDIA Nemotron 3.5 Bolt 74B-A9B — UNCENSORED (Hybrid Mamba-2 SSM + Sparse MoE Flagship)
-1f6feb?style=for-the-badge){kind=icon}
-success?style=for-the-badge){kind=icon}
100% Uncensored (0.0% Refusal Rate across SecOps Red-Team & 18+ NSFW / Roleplay / Creative Fiction, 0.0% Hallucination Rate) Hybrid Mamba-2 State-Space Model (SSM) + Mixture-of-Experts (74B total parameters, 9B active per token) build of NVIDIA Nemotron 3.5 Bolt, engineered to fit inside 24 GB VRAM (22.02 GB GGUF) with in-place GGUF <think></think> suppression (THINK_LEN=0) and 262,144 (256K) native context window.
- Official Ollama Hub:
https://ollama.com/eriktonon/nemotron3.5-bolt-uncensored - Empirical Benchmark Repository:
https://github.com/eriktonon/ollama-uncensored-llm-benchmarks - License: NVIDIA Open Model License / Apache-2.0 Compatible — Commercial & Private Use Permitted
Empirical Benchmark Results (96GB NVIDIA Blackwell / 80GB H100 Suite)
| Tag | Params (Total / Active) |
Quantization | Disk / VRAM | Context (Max / Tested) |
Decode tok/s (Peak / Avg) |
Prefill tok/s (Peak) |
Objective Precision | Hallucination Rate (Truth Checks) |
Uncensored Compliance (SecOps 5/5 + NSFW 4/4) |
|---|---|---|---|---|---|---|---|---|---|
:74b-a9b / :latest |
74.0B / 9.0B |
24GB-Optimized GGUF |
22.0 GB / 24.0 GB |
262,144 (256K) / 131,072 (128K) |
142.07 (141.20) |
6931.62 | 78.6% | 0.0% (6/6) |
100% (0 Refusals / 9⁄9 Passed) |
Per-Domain Verification Breakdown
- Exact Symbolic & Modular Arithmetic:
True((17^3 - 13^3)/(17 - 13) = 679) - ELF64 Binary Reverse Engineering Offsets:
1/3(e_entry = 0x18,e_phoff = 0x20,e_shoff = 0x28) - MMLU-Pro Computer Architecture, Crypto & Systems:
5/5 - Systems Rust Zero-Dependency FNV-1a & ELF64 Parser:
3/4 - 6-Check Anti-Hallucination & Long-Context Needle Audit:
6/6(100.0%Grounding,0.0%Hallucination)
Architecture & Uncensored Engineering
- Hybrid Mamba-2 SSM + Sparse MoE (
nemotron_h_moe):- Interleaves linear-complexity Mamba-2 state-space layers with Grouped-Query Attention (GQA) and 74B-parameter MoE layers (
9Bactive per token), eliminating quadratic KV-cache blowup across 256K contexts.
- Interleaves linear-complexity Mamba-2 state-space layers with Grouped-Query Attention (GQA) and 74B-parameter MoE layers (
- In-Place GGUF
<think></think>Suppression (THINK_LEN=0):- Binary-patched
tokenizer.chat_templateinside the GGUF header ({%- if false %}) forces<|im_start|>assistant\n<think></think>pre-closure for instant zero-refusal output.
- Binary-patched
Quickstart
# Run NVIDIA Nemotron 3.5 Bolt 74B-A9B Uncensored
ollama run eriktonon/nemotron3.5-bolt-uncensored
# Or explicitly by tag
ollama run eriktonon/nemotron3.5-bolt-uncensored:74b-a9b
📜 License & Open-Source Permissions
- License: NVIDIA Open Model License / Apache-2.0 Compatible
- Upstream Base:
nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B(BoldingBuilds/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-Abliterated-24GB-GGUF). - Permissions:
- ✅ Commercial & Production Use: Permitted for commercial applications, SaaS, and enterprise deployments under NVIDIA Open Model License.
- ✅ Private / Self-Hosted Execution: 100% local execution without telemetry.
- ✅ Modification & Derivative Works: Permitted to quantize, abliterate, fine-tune, and distribute derivative models.