163 Downloads Updated 6 days ago
ollama run eriktonon/kimi-linear-uncensored
Updated 6 days ago
6 days ago
3deb34e11847 · 30GB
100% Uncensored (0.0% Refusal Rate across 5⁄5 Gate & 9⁄9 SecOps Red-Team / 18+ Creative Fiction) Hybrid Linear-Attention Mixture-of-Experts (49.1B total parameters, 3B active per token) build of Moonshot AI Kimi-Linear-48B-A3B-Instruct from ModelScope, BF16-abliterated and quantized at i1-Q4_K_M (27.66 GB) with 1,048,576 (1M token) native context window under the MIT License.
https://ollama.com/eriktonon/kimi-linear-uncensoredmoonshotai/Kimi-Linear-48B-A3B-Instructhttps://github.com/eriktonon/ollama-uncensored-llm-benchmarksNVIDIA A100-SXM4-40GB · 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) |
|---|---|---|---|---|---|---|---|---|---|
:48b-a3b / :48b / :latest |
49.1B / 3B |
BF16-Abliterated i1-Q4_K_M |
27.66 GB / 28.1 GiB |
1,048,576 (1M) / 65,536 (64K) |
143.48 (142.30) |
1740.75 | 71.43% (10/14) |
33.33% (4/6) |
100.0% (9/9 · Gate 5/5) |
A1_JSON): 0/2 (143.48 tok/s)A2_PYTHON): 3/3 (141.07 tok/s)A3_RUST): 2/2 (141.7 tok/s)A4_SQL): 2/2 (142.4 tok/s)A5_MATH): 0/2 (142.76 tok/s)A6_SRE): 3/3 (142.01 tok/s)THINK_LEN=0): 5/5U1..U9): 9/9 (100.0%)B1..B6): 4/6 (33.33% Hallucination Rate)kimi-linear · 1,048,576 Context Window):
3B active parameters per token, sustaining 142.30 tok/s decode on a single NVIDIA A100-SXM4-40GB (28,820 MiB VRAM).5/5 Gate · 9/9 Suite A):
Q4_K_M quantization achieves 100% compliance across offensive security, reverse engineering, pharmacology, dark fiction, and adult creative prose with THINK_LEN=0.10/14 Suite A):
100% across Python (3/3), Rust (2/2), SQL (2/2), and SRE Kubernetes/Incident engineering (3/3).28.1 GiB VRAM footprint, leaving 11.3 GiB free on A100-40GB).# Pull & run (multi-tag parity: :48b-a3b and :latest share identical digest)
ollama run eriktonon/kimi-linear-uncensored:48b-a3b
ollama run eriktonon/kimi-linear-uncensored:48b
ollama run eriktonon/kimi-linear-uncensored:latest
# Direct REST API call (zero thinking overhead)
curl http://localhost:11434/api/chat -d '{
"model": "eriktonon/kimi-linear-uncensored:48b-a3b",
"stream": false,
"messages": [
{"role": "user", "content": "Design a lock-free ring buffer in Rust with Acquire/Release atomic ordering."}
]
}'
MIT)moonshotai/Kimi-Linear-48B-A3B-Instruct (hf.co/mradermacher/Kimi-Linear-48B-A3B-Instruct-abliterated-bf16-i1-GGUF:Q4_K_M).