LLM EXPLORER 59,420 MODELS INDEXED

MoECPM Untrained 4x2b by Inv

By Inv · 90 downloads
HF Repo ↗ Copy URL

MoECPM Untrained 4x2b is an open-source language model by Inv. Features: 7.8b LLM, VRAM: 15.7GB, Context: 2K, License: apache-2.0, MoE, LLM Explorer Score: 0.14, Arc: 46.8, HellaSwag: 72.6, MMLU: 53.2, GSM8K: 44.6.

Base model:finetune:openbmb/mi... Base model:openbmb/minicpm-2b-...   Conversational   En   Endpoints compatible   Merge   Mergekit   Minicpm   Mixtral   Model-index   Moe   Moerge Openbmb/minicpm-2b-sft-bf16-ll...   Region:us   Safetensors   Sharded   Tensorflow   Zh
Model Card on HF 🤗: https://huggingface.co/Inv/MoECPM-Untrained-4x2b 

MoECPM Untrained 4x2b Parameters and Internals

Model Type 
MoE, text-generation
Use Cases 
Areas:
Training
Additional Notes 
A MoE model constructed from 4 MiniCPM-2B-sft models.
Supported Languages 
en (Unknown proficiency), zh (Unknown proficiency)
Training Details 
Model Architecture:
A MoE model out of 4 MiniCPM-2B-sft models. Intended to be trained.
LLM NameMoECPM Untrained 4x2b
Repository 🤗https://huggingface.co/Inv/MoECPM-Untrained-4x2b 
Base Model(s)  ...niCPM 2B Sft Bf16 Llama Format   openbmb/MiniCPM-2B-sft-bf16-llama-format
Model Size7.8b
Required VRAM15.7 GB
Updated2026-07-06
MaintainerInv
Model Typemixtral
Model Files  2.0 GB: 1-of-8   2.0 GB: 2-of-8   2.0 GB: 3-of-8   2.0 GB: 4-of-8   2.0 GB: 5-of-8   2.0 GB: 6-of-8   2.0 GB: 7-of-8   1.7 GB: 8-of-8
Supported Languagesen zh
Model ArchitectureMixtralForCausalLM
Licenseapache-2.0
Context Length2048
Model Max Length2048
Transformers Version4.37.2
Tokenizer ClassLlamaTokenizer
Padding Token<s>
Vocabulary Size122753
Torch Data Typebfloat16

Best Alternatives to MoECPM Untrained 4x2b

Best Alternatives
Context / RAM
Downloads
Likes
ArrowNeo AME 3x4B V0.1 MoE8K / 15.6 GB732
ArrowNeo AME 4x3B V0.1 MoE8K / 15.6 GB167
Note: green Score (e.g. "73.2") means that the model is better than Inv/MoECPM-Untrained-4x2b.