Steelskull/L3.3-Electra-R1-70b
Hugging Face
TEXT GENERATIONConcurrency Cost:4Model Size:70BQuant:FP8Ctx Length:32kPublished:Mar 2, 2025License:eva-llama3.3Architecture:Transformer0.1K Warm

Steelskull/L3.3-Electra-R1-70b is a 70 billion parameter language model built on a custom DeepSeek R1 Distill base, specifically TheSkullery/L3.1x3.3-Hydroblated-R1-70B-v4.4. This model integrates specialized components through the SCE merge method, utilizing float32 processing and bfloat16 output for optimized performance. It is the sixth iteration in the "Unnamed" series, developed with extensive user feedback to enhance intelligence, coherence, and deep character insights. Electra-R1 excels at advanced reasoning and unprompted exploration of character motivations, making it ideal for complex narrative generation and roleplay scenarios.

Loading preview...

L3.3-Electra-R1-70b: An Advanced 70B Language Model

L3.3-Electra-R1-70b is the latest iteration in the "Unnamed" series by SteelSkull, a 70 billion parameter model built upon a custom DeepSeek R1 Distill base, specifically TheSkullery/L3.1x3.3-Hydroblated-R1-70B-v4.4. This model leverages the SCE merge method to integrate various specialized components, ensuring a robust and coherent architecture. It processes data using float32 and outputs in bfloat16 for optimized performance.

Key Capabilities & Differentiators

  • Enhanced Intelligence & Coherence: User feedback consistently highlights Electra-R1's superior intelligence and coherence, making it a new gold standard and baseline for the series.
  • Deep Character Insights: The model demonstrates a unique ability to provide deep character insights and unprompted exploration of inner thoughts and motivations, particularly valuable for narrative and roleplay applications.
  • Advanced Reasoning: Through proper prompting, Electra-R1 exhibits advanced reasoning capabilities.
  • Custom Base Architecture: Built on the Hydroblated-R1 base, known for stability and enhanced reasoning, with SCE merge settings precisely tuned based on extensive community feedback from over 10 different models.
  • Specialized Component Integration: Incorporates models like EVA-LLaMA-3.33-70B-v0.0 for core capabilities, Wayfarer-Large-70B-Llama-3.3 for storytelling and roleplay, L3.3-70B-Euryale-v2.3 as an all-rounder RP model, 70B-L3.3-Cirrus-x1 for improved coherence, L3.1-70B-Hanami-x1 for balanced responses, and Anubis-70B-v1 for enhanced detail. It also includes Negative_LLAMA_70B and Fallen-Llama-3.3-R1-70B-v1 for reduced bias.

Recommended Use Cases

Electra-R1 is particularly well-suited for:

  • Complex Narrative Generation: Its ability to provide deep character insights and coherent storytelling makes it excellent for creative writing.
  • Advanced Roleplay: Excels in scenarios requiring nuanced character interactions and exploration of motivations.
  • Reasoning-Intensive Tasks: Benefits from its enhanced reasoning capabilities for more analytical prompts.

Recommended sampler settings by @Geechan are provided, including a static temperature of 1.0 (or dynamic 0.8-1.05), Min P of 0.025-0.03, and specific DRY settings (Multiplier: 0.8, Base: 1.74, Length: 4-6). The model also supports advanced reasoning configurations using XML tags like <think> for structured thought processes.

Popular Sampler Settings

Top 3 parameter combinations used by Featherless users for this model. Click a tab to see each config.

temperature
top_p
top_k
frequency_penalty
presence_penalty
repetition_penalty
min_p