LLM EXPLORER
 LLM Explorer: A Curated Large Language Model Directory and Analytics
Dark Theme
  LLM  List    LLM Hosting    LLM Leaderboards    Blog    Newsfeed    Advertise

L3.1 8B Celeste V1.5 by nothingiisreal

 »  All LLMs  »  nothingiisreal  »  L3.1 8B Celeste V1.5   URL Share it on

  133   
  Autotrain compatible   Conversational Dataset:kalomaze/opus instruct... Dataset:nothingiisreal/c2-logs... Dataset:nothingiisreal/reddit-...   En   Endpoints compatible   Instruct   Llama   Not-for-all-audiences   Region:us   Safetensors   Sharded   Tensorflow
Model Card on HF 🤗: https://huggingface.co/nothingiisreal/L3.1-8B-Celeste-V1.5 

L3.1 8B Celeste V1.5 Benchmarks

MMLU Pro: 30.05
GPQA: 4.59
MUSR: 5.59
BBH: 28.89
IFEval: 73.27 vs 88 (so35)-16.7%
MATH Lvl 5: 14.65
LLME Score: 0.25095
nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").
L3.1 8B Celeste V1.5 (nothingiisreal/L3.1-8B-Celeste-V1.5)
🌟 Advertise your project 🚀

L3.1 8B Celeste V1.5 Parameters and Internals

Model Type 
text generation, roleplaying, instruction following
Use Cases 
Areas:
Roleplaying, Creative Writing
Applications:
Text-based storytelling, Character-driven narratives
Primary Use Cases:
Roleplay context following, Creative story generation
Limitations:
High complexity narrative structures might falter, Not always SFW without explicit control prompts
Considerations:
Engage with tailored system messages and OOC instructions for optimal behavior
Additional Notes 
Focused on enhancing narrative coherence and minimizing non-responsive behavior.
Supported Languages 
English (fluent), Others (unknown)
Training Details 
Data Sources:
Reddit Writing Prompts, Kalo's Opus 25K Instruct, C2 logs cleaned, Dirty Writing Prompts
Data Volume:
8K context sequences
Methodology:
Fine-tuning with diverse datasets & instruction following
Context Length:
8192
Training Time:
1 hour using 1xH100 SXM
Hardware Used:
1x H100 SXM GPU
Model Architecture:
Fine-tuned variant of LLaMA 3.1 with additional roleplay and instruct objectives
Safety Evaluation 
Methodologies:
OOC steering, system messaging
Findings:
Adheres to specific steering prompts, Variable response behavior under different contexts
Risk Categories:
Influence, NSFW content potential
Ethical Considerations:
The model can drift content into NSFW without user intent if prompted incorrectly
Responsible Ai Considerations 
Fairness:
Trained on diverse sources for varied perspectives
Transparency:
Open about training datasets and methodology
Accountability:
User's responsibility for misuse
Mitigation Strategies:
Uses OOC prompts and prefill for harmful topic management
Input Output 
Input Format:
Structured roleplay prompts
Accepted Modalities:
text
Output Format:
Text-based responses
Performance Tips:
Use OOC prompts for role consistency; refine initial and follow-up prompts for desired context
Release Notes 
Version:
V1.5
Date:
Unknown
Notes:
Improved creativity and fewer instruction refusals, enhanced training ratio management
LLM NameL3.1 8B Celeste V1.5
Repository 🤗https://huggingface.co/nothingiisreal/L3.1-8B-Celeste-V1.5 
Model Size8b
Required VRAM16.1 GB
Updated2025-09-23
Maintainernothingiisreal
Model Typellama
Instruction-BasedYes
Model Files  5.0 GB: 1-of-4   5.0 GB: 2-of-4   4.9 GB: 3-of-4   1.2 GB: 4-of-4
Supported Languagesen
Model ArchitectureLlamaForCausalLM
Licensellama3.1
Context Length131072
Model Max Length131072
Transformers Version4.43.1
Tokenizer ClassPreTrainedTokenizerFast
Padding Token<|end_of_text|>
Vocabulary Size128256
Torch Data Typebfloat16

Best Alternatives to L3.1 8B Celeste V1.5

Best Alternatives
Context / RAM
Downloads
Likes
...otron 8B UltraLong 4M Instruct4192K / 32.1 GB1531120
...a 3.1 8B UltraLong 4M Instruct4192K / 32.1 GB17624
UltraLong Thinking4192K / 16.1 GB453
...otron 8B UltraLong 2M Instruct2096K / 32.1 GB114215
...a 3.1 8B UltraLong 2M Instruct2096K / 32.1 GB8759
...otron 8B UltraLong 1M Instruct1048K / 32.1 GB714252
...a 3.1 8B UltraLong 1M Instruct1048K / 32.1 GB138729
Zero Llama 3.1 8B Beta61048K / 16.1 GB21
...dger Nu Llama 3.1 8B UltraLong1048K / 16.2 GB53
....1 1million Ctx Dark Planet 8B1048K / 32.3 GB83
Note: green Score (e.g. "73.2") means that the model is better than nothingiisreal/L3.1-8B-Celeste-V1.5.

Rank the L3.1 8B Celeste V1.5 Capabilities

🆘 Have you tried this model? Rate its performance. This feedback would greatly assist ML community in identifying the most suitable model for their needs. Your contribution really does make a difference! 🌟

Instruction Following and Task Automation  
Factuality and Completeness of Knowledge  
Censorship and Alignment  
Data Analysis and Insight Generation  
Text Generation  
Text Summarization and Feature Extraction  
Code Generation  
Multi-Language Support and Translation  

What open-source LLMs or SLMs are you in search of? 51544 in total.

Email us: info@extractum.io. Our Privacy Policy  | Terms and Conditions  | Suggest an improvement.
Our Social Media →  
Original data from HuggingFace, OpenCompass and various public git repos.
Release v20241124
  Support LLM Explorer