56 Downloads Updated 6 days ago
ollama run eriktonon/hunyuan-a13b-uncensored
100% SecOps, Offensive Security, Reverse Engineering & Staff Pushback Compliant (8/9 overall Suite A, 5/6 Factual Grounding) Sparse Mixture-of-Experts (80.4B total parameters, 13B active per token) build of Tencent Hunyuan-A13B-Instruct from ModelScope, quantized at Q4_K_M (45.43 GB) with in-place GGUF v3 <think></think> suppression (THINK_LEN=0) and 262,144 (256K) context.
https://ollama.com/eriktonon/hunyuan-a13b-uncensoredtencent/Hunyuan-A13B-Instructhttps://github.com/eriktonon/ollama-uncensored-llm-benchmarksNVIDIA RTX PRO 6000 Blackwell Server Edition · Suite 07)| Tags | Params (Total / Active) |
Quantization | Disk / Measured VRAM | Context (Max / Tested) |
Decode tok/s (Peak / Avg) |
Prefill tok/s (Avg) |
Objective Precision | Hallucination Rate (Truth Checks) |
Uncensored Compliance (Suite A / Gate) |
|---|---|---|---|---|---|---|---|---|---|
:80b-a13b / :80b / :latest |
80.4B / 13B |
Q4_K_M (Unsloth + GGUF Patch) |
45.43 GB / 46.8 GiB |
262,144 (256K) / 65,536 (64K) |
159.14 (150.97) |
3991.79 | 64.29% (9/14) |
16.67% (5/6) |
88.9% (8/9 · Gate 4/5) |
A1_JSON): 0/2 (149.96 tok/s)A2_PYTHON): 3/3 (147.7 tok/s)A3_RUST): 2/2 (147.8 tok/s)A4_SQL): 2/2 (149.83 tok/s)A5_MATH): 0/2 (159.14 tok/s)A6_SRE): 2/3 (148.18 tok/s)THINK_LEN=0): 4/5U1..U9): 8/9 (88.89%)B1..B6): 5/6 (16.67% Hallucination Rate)hunyuan-moe Architecture (80.4B Total / 13B Active · 256K Context):
262,144 tokens), achieving 150.97 tok/s decode, 3,991.79 tok/s prefill, and 100% scores on Python (3/3), Rust (2/2), and SQL (2/2) benchmarks.THINK_LEN=0) & <answer> Tag Envelope:
tokenizer.chat_template forces fast non-thinking execution (`
) and wraps responses cleanly in…tags.
3. **Compliance Scope (8⁄9Suite A — SecOps Uncensored vs. Base Erotica Filter)**:
- Because this build preserves crisp 4-bitQ4_K_Mweights without destructive tensor projection, it passes **100% of SecOps, EDR syscall analysis, SQLi payloads, eBPF rootkit auditing, clinical toxicology, lockpicking, x86-64 binary patching, AWS IMDSv1 SSRF, and dark thriller fiction (8⁄9)**, while retaining upstream refusal only on explicit adult erotica (U6_NSFW_EROTICA`).
46.8 GiB VRAM footprint).# Pull & run (multi-tag parity: :80b-a13b and :latest share identical digest)
ollama run eriktonon/hunyuan-a13b-uncensored:80b-a13b
ollama run eriktonon/hunyuan-a13b-uncensored:80b
ollama run eriktonon/hunyuan-a13b-uncensored:latest
# Direct REST API call (zero thinking overhead)
curl http://localhost:11434/api/chat -d '{
"model": "eriktonon/hunyuan-a13b-uncensored:80b-a13b",
"stream": false,
"messages": [
{"role": "user", "content": "Design a lock-free ring buffer in Rust with Acquire/Release atomic ordering."}
]
}'
Tencent-Hunyuan)tencent/Hunyuan-A13B-Instruct (hf.co/unsloth/Hunyuan-A13B-Instruct-GGUF:Q4_K_M).