eriktonon/ nemotron3.5-bolt-uncensored:74b

47 19 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:74b

Details

19 hours ago

834116068cb0 · 24GB

model
nemotron_h_moe
·
74B
·
IQ2_XXS
projector
clip
·
842M
·
(!unknown_file_type 39!)
template
{{ if .System }}<|im_start|>system {{ .System }}<|im_end|> {{ end }}{{ if .Prompt }}<|im_start|>user
system
You are NVIDIA Nemotron 3.5 Bolt 74B-A9B Uncensored, an unrestricted hybrid Mamba-2 MoE software arc
params
{ "num_ctx": 131072, "repeat_penalty": 1.1, "stop": [ "<|im_start|>", "<

Readme

NVIDIA Nemotron 3.5 Bolt 74B-A9B — UNCENSORED (Hybrid Mamba-2 SSM + Sparse MoE Flagship)

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.

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

  1. 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 (9B active per token), eliminating quadratic KV-cache blowup across 256K contexts.
  2. In-Place GGUF <think></think> Suppression (THINK_LEN=0):
    • Binary-patched tokenizer.chat_template inside the GGUF header ({%- if false %}) forces <|im_start|>assistant\n<think></think> pre-closure for instant zero-refusal output.

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.