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

Cosmosage V2 by Tijmen2

 »  All LLMs  »  Tijmen2  »  Cosmosage V2   URL Share it on

  3   
  4bit   Autotrain compatible Base model:finetune:mistralai/... Base model:mistralai/mistral-7...   Conversational   Cosmology   Dataset:teknium/openhermes-2.5   En   Endpoints compatible   Gptq   Mistral   Physics   Pytorch   Quantized   Region:us   Safetensors
Model Card on HF 🤗: https://huggingface.co/Tijmen2/cosmosage_v2 

Cosmosage V2 Benchmarks

ARC: 59.73 vs 96.7 (so35)-38.2%
HellaSwag: 80.9 vs 95.3 (gpt4)-15.1%
MMLU: 59.57 vs 88.3 (so35)-32.5%
TruthfulQA: 50.98 vs 59 (gpt4)-13.6%
WinoGrande: 75.93 vs 87.5 (gpt4)-13.2%
GSM8K: 36.85 vs 96.4 (so35)-61.8%
LLME Score: 0.16034
nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").
Cosmosage V2 (Tijmen2/cosmosage_v2)
🌟 Advertise your project 🚀

Cosmosage V2 Parameters and Internals

Model Type 
text generation
Additional Notes 
The model is designed as a natural-language cosmology assistant but continues to have reliability issues in ensuring factually accurate responses.
Training Details 
Data Sources:
thousands of papers and textbooks
Methodology:
first underwent continued pretraining on papers and textbooks, then fine-tuned on synthetically-generated question-answer pairs
Hardware Used:
4xA100 (80 GB) at the Center for Computational Astrophysics (CfCA), National Astronomical Observatory of Japan (NAOJ)
Input Output 
Input Format:
inst chat template with U+2581 Lower One Eighth Block Unicode Character to separate sections
Output Format:
text in Q&A format
LLM NameCosmosage V2
Repository 🤗https://huggingface.co/Tijmen2/cosmosage_v2 
Base Model(s)  mistralai/Mistral-7B-v0.1   mistralai/Mistral-7B-v0.1
Model Size7b
Required VRAM4.2 GB
Updated2025-08-31
MaintainerTijmen2
Model Typemistral
Model Files  4.2 GB   7.7 GB   14.5 GB   14.5 GB
Supported Languagesen
GPTQ QuantizationYes
Quantization Typegptq|4bit|8bit
Model ArchitectureMistralForCausalLM
Licensemit
Context Length32768
Model Max Length32768
Transformers Version4.38.0.dev0
Tokenizer ClassLlamaTokenizer
Padding Token</s>
Vocabulary Size32000
Torch Data Typebfloat16

Best Alternatives to Cosmosage V2

Best Alternatives
Context / RAM
Downloads
Likes
...enHermes 2.5 Mistral 7B Marlin32K / 4.1 GB5052
Zephyr 7B Beta Marlin32K / 4.1 GB1170
...ral 7B Instruct V0.3 GPTQ 4bit32K / 4.2 GB7600320
...ephyr 7B Beta Channelwise Gptq32K / 4 GB290560
...ral 7B Instruct V0.3 GPTQ 4bit32K / 4.2 GB294618
Mistral 7B Instruct V0.3 GPTQ32K / 4.2 GB5551
...istral 7B Pruned50 GPTQ Marlin32K / 4 GB80
Mistral 7B Unsloth Gptq 8bit32K / 7.7 GB50
...phyr 7B Beta Assistant V1 Gptq32K / 4.2 GB21
...l Neural Chat 7B V3.8 Bit Gptq32K / 7.7 GB70
Note: green Score (e.g. "73.2") means that the model is better than Tijmen2/cosmosage_v2.

Rank the Cosmosage V2 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? 51022 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