LLM EXPLORER 59,420 MODELS INDEXED

KULLM RLHF by Trofish

By Trofish · 11 downloads
HF Repo ↗ Copy URL

KULLM RLHF is an open-source language model by Trofish. Features: LLM, VRAM: 25.8GB, Context: 2K, LLM Explorer Score: 0.08.

  Arxiv:2303.16634   Endpoints compatible   Gpt neox   Pytorch   Region:us
Model Card on HF 🤗: https://huggingface.co/Trofish/KULLM-RLHF 

KULLM RLHF Parameters and Internals

Model Type 
Conversational AI, Chatbot
Use Cases 
Areas:
Research, Chatbot Development
Applications:
Korean language conversational AI
Primary Use Cases:
Friendly and harmless everyday conversations in Korean
Additional Notes 
The final model, implemented with RLHF and DeepSpeedChat, aims to produce high-quality conversational responses that are user-friendly and ethical.
Supported Languages 
Korean (high)
Training Details 
Data Sources:
Self-Instruct using GPT-4, RLHF with human feedback, DeepSpeed optimization
Methodology:
Reinforcement Learning from Human Feedback, Self-Instruct Data Augmentation, DeepSpeed for large-scale distributed deep learning
Hardware Used:
Google Colab A100 40GB GPU
Model Architecture:
Used KULLM as baseline model trained with RLHF and SFT (Supervised Fine-tuning) techniques
LLM NameKULLM RLHF
Repository 🤗https://huggingface.co/Trofish/KULLM-RLHF 
Required VRAM25.8 GB
Updated2026-07-28
MaintainerTrofish
Model Typegpt_neox
Model Files  25.8 GB
Model ArchitectureGPTNeoXForCausalLM
Context Length2048
Model Max Length2048
Transformers Version4.31.0
Tokenizer ClassPreTrainedTokenizerFast
Padding Token<|endoftext|>
Vocabulary Size30008
Torch Data Typefloat16

Best Alternatives to KULLM RLHF

Best Alternatives
Context / RAM
Downloads
Likes
Catlm8K / 7.8 GB144
...Prover 14final Checkpoint 58304K / 14.9 GB60
Neox Musenet Untrained4K / 7.3 GB60
Stabillm Instruct De4K / 31.8 GB70
MusePy 1 12K / 0.7 GB251
Open Calm Large2K / 1.8 GB134610
MonoCoder OMP2K / 3.6 GB130
ProofGPT V0.12K / 2.9 GB763
Open Calm Small2K / 0.4 GB221420
Step3 Mk72K / 25.8 GB70
Note: green Score (e.g. "73.2") means that the model is better than Trofish/KULLM-RLHF.