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

Cosmo 1B by HuggingFaceTB

 »  All LLMs  »  HuggingFaceTB  »  Cosmo 1B   URL Share it on

  912   
  Autotrain compatible Dataset:huggingfacetb/cosmoped...   En   Endpoints compatible   Llama   Region:us   Safetensors   Sharded   Tensorflow
Model Card on HF 🤗: https://huggingface.co/HuggingFaceTB/cosmo-1b 

Cosmo 1B Benchmarks

ARC: 38.57 vs 96.7 (so35)-60.1%
HellaSwag: 55.08 vs 95.3 (gpt4)-42.2%
MMLU: 26.5 vs 88.3 (so35)-70%
TruthfulQA: 38.26 vs 59 (gpt4)-35.2%
WinoGrande: 55.33 vs 87.5 (gpt4)-36.8%
GSM8K: 5.38 vs 96.4 (so35)-94.4%
LLME Score: 0.21195
nn.n% — How the model compares to the reference models: Anthropic Sonnet 3.5 ("so35"), GPT-4o ("gpt4o") or GPT-4 ("gpt4").
Cosmo 1B (HuggingFaceTB/cosmo-1b)
🌟 Advertise your project 🚀

Cosmo 1B Parameters and Internals

Model Type 
text generation, chat
Use Cases 
Areas:
text generation, educational
Applications:
story generation, Q&A generation
Primary Use Cases:
chatbot applications
Limitations:
may hallucinate or provide incomplete/incorrect answers due to synthetic training data
Supported Languages 
en (full)
Training Details 
Data Sources:
Cosmopedia, non-synthetic sources like code-python-0.60-to-1.00 and web-0.50-to-1.00 subsets of AutoMathText, The Stack's Jupyter Notebooks, ultrachat
Data Volume:
180 billion tokens
Methodology:
Used UltraChat data formatted for LlaMa models, to avoid post pre-training instruction-tuning, and upsampled data from seed sources.
Context Length:
2048
Training Time:
15 hours
Hardware Used:
160 H100 GPUs
Model Architecture:
Llama-2
Input Output 
Input Format:
Chat and regular text prompts
Accepted Modalities:
text
Output Format:
Generated text sequence
Performance Tips:
Use chat format for better instructional adherence.
LLM NameCosmo 1B
Repository 🤗https://huggingface.co/HuggingFaceTB/cosmo-1b 
Model Size1b
Required VRAM7 GB
Updated2025-07-24
MaintainerHuggingFaceTB
Model Typellama
Model Files  5.0 GB: 1-of-2   2.0 GB: 2-of-2
Supported Languagesen
Model ArchitectureLlamaForCausalLM
Licenseapache-2.0
Context Length2048
Model Max Length2048
Transformers Version4.37.2
Tokenizer ClassLlamaTokenizer
Vocabulary Size32000
Torch Data Typebfloat16

Best Alternatives to Cosmo 1B

Best Alternatives
Context / RAM
Downloads
Likes
ISA 02 Nano Llama 3.2 1B1024K / 2.5 GB323
LWM Text Chat 1M1024K / 13.5 GB1363174
LWM Text 1M1024K / 13.5 GB12129
JOSIE 1M Base1024K / 13.5 GB121
JOSIE 1M Base1024K / 13.5 GB61
Llama 3.2 1B Instruct128K / 2.5 GB34164541004
Llama 3.2 1B128K / 2.5 GB35070201995
Llama 3.2 1B Instruct128K / 2.5 GB7990176
Llasa 1B Multilingual128K / 3.5 GB6273738
Llama 3.2 1B128K / 2.5 GB28120
Note: green Score (e.g. "73.2") means that the model is better than HuggingFaceTB/cosmo-1b.

Rank the Cosmo 1B 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? 50035 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