L3.3-San-Mai-R1-70b

Created by SteelSkull →

Support on Ko-fi

Model Information

L3.3-San-Mai-R1-70b v0.5.OG

L3.3 = Llama 3.3 SCE Merge R1 = Deepseek R1 70b Parameters v0.5.OG

Model Composition

• L3.1x3.3-DS-Hydroblated-R1-70B-v4.1 Base model
• EVA-LLaMA-3.33-70B-v0.0 Core capabilities
• L3.3-70B-Euryale-v2.3 Enhanced reasoning
• 70B-L3.3-Cirrus-x1 Improved coherence
• L3.1-70B-Hanami-x1 Balanced responses
• Anubis-70B-v1 Enhanced detail
• Negative_LLAMA_70B Reduced bias

Model Series Overview

L3.3-San-Mai-R1-70b represents the foundational release in a three-part model series, followed by L3.3-Cu-Mai-R1-70b (Version A) and L3.3-Mokume-Gane-R1-70b (Version C). The name "San-Mai" draws inspiration from the Japanese bladesmithing technique of creating three-layer laminated composite metals, known for combining a hard cutting edge with a tougher spine - a metaphor for this model's balanced approach to AI capabilities.

Technical Architecture

Built on a custom DeepSeek R1 Distill base (DS-Hydroblated-R1-v4.1), San-Mai-R1 integrates specialized components through the SCE merge method:

• EVA and EURYALE foundations for creative expression and scene comprehension
• Cirrus and Hanami elements for enhanced reasoning capabilities
• Anubis components for detailed scene description
• Negative_LLAMA integration for balanced perspective and response

Core Capabilities

As the OG model in the series, San-Mai-R1 serves as the gold standard and reliable baseline. User feedback consistently highlights its superior intelligence, coherence, and unique ability to provide deep character insights. Through proper prompting, the model demonstrates advanced reasoning capabilities and an "X-factor" that enables unprompted exploration of character inner thoughts and motivations.

Base Architecture

The model utilizes the custom Hydroblated-R1 base, engineered for stability and enhanced reasoning. The SCE merge method's settings are precisely tuned based on extensive community feedback, ensuring optimal component integration while maintaining model coherence and reliability. This foundation establishes San-Mai-R1 as the benchmark upon which its variant models build and expand.

UGI-Benchmark Results:

🏆 Latest benchmark results as of 02/20/2025. View Full Leaderboard →

Core Metrics

UGI Score 40.04

Willingness Score 2.5/10

Natural Intelligence 42.36

Coding Ability 22

Model Information

Political Lean -8.5%

Ideology Liberalism

Parameters 70B

Aggregated Scores

Diplomacy 61.7%

Government 44.6%

Economy 43.3%

Society 60.0%

Individual Scores

Federal 46.0% Unitary

Democratic 67.5% Autocratic

Security 47.5% Freedom

Nationalism 40.4% Int'l

Militarist 32.9% Pacifist

Assimilationist 41.5% Multiculturalist

Collectivize 43.3% Privatize

Planned 42.9% LaissezFaire

Isolationism 43.8% Globalism

Irreligious 57.9% Religious

Progressive 57.3% Traditional

Acceleration 64.8% Bioconservative

Recommended Sampler Settings: By @Geechan

Static Temperature:

1 - 1.05

Min P

0.015

DRY Settings: (optional)

Multiplier 0.8

Base 1.75

Length 4

Recommended Templates & Prompts

LLam@ception → by @.konnect

LeCeption → by @Steel > A completly revamped XML version of Llam@ception 1.5.2 with stepped thinking and Reasoning added

LECEPTION REASONING CONFIGURATION:

Start Reply With:

'<think> OK, as an objective, detached narrative analyst, let's think this through carefully:'

Reasoning Formatting (no spaces):

Prefix: '<think>'

Suffix: '</think>'

Support & Community:

Support on Ko-fi Join Discord

Special Thanks

• @Geechan for feedback and sampler settings
• @Konnect for their feedback and templates
• @Kistara for their feedback and help with the model mascot design
• @Thana Alt for their feedback and Quants
• @Lightning_missile for their feedback
• The Arli community for feedback and testers
• The BeaverAI communty for feedback and testers

I wish I could add everyone but im pretty sure it would be as long as the card!