AI & ML interests

We create a Dront AI. Мы создаем Dront ИИ.

Recent Activity

MishaGGG  updated a Space about 6 hours ago
VDC-team/README
MishaGGG  updated a collection 1 day ago
VDrontV3
MishaGGG  published a model 1 day ago
VDC-team/VDrontV3-Mini
View all activity

Organization Card

English:

DAC – Dront AI Command

We are DAC (Dront AI Command). We create datasets and neural networks of the Dront family.

Russian:

DAC – Dront AI Command

Мы — DAC (Dront AI Command). Мы занимаемся созданием датасетов и нейронных сетей семейства Dront.

Email (temporary) : mirnahlysta@gmail.com VDront versions:

Versions:

Version Description Architecture
SmallDront The very first version—and the stupidest one. GPT-Like
SmallDront 2 Improved SmallDront 1. GPT-Like
VDrontGPT The first models trained on FineWeb-Edu, which knew more facts than SmallDront. GPT-Like
V1 Based on knowledge obtained from VDrontGPT, a 100-million-parameter model was created that outperformed VDrontGPT. GPT-Like
V1.5 Improved V1, with two versions: Base (150m) and Flash (60m). GPT-Like
V2 A higher-quality version of VDront, featuring a 400M model, Instruct 100m, and Title 100m capabilities. Have a .gguf and .dq(dront quant). GPT-Like
V3 Our architecture, similar to MoE, unlike V2 can simultaneously perform the functions of the Base, SmallTalk, and Chat models. Custom MoE/GPT-Like

Models (V1-V3):

VDront Model Architecture Tokenizer Size (params) Base model Quantization
V1 VDrontV1 GPT-Like GPT2 100m -
V1.5 VDrontV1.5 GPT-Like GPT2 150m VDrontV1
V1.5 VDrontV1.5-Flash GPT-Like GPT2 60m -
V2 VDrontV2-0.4b GPT-Like GPT2 400m VDrontV1 ✅ .gguf
V2 VDrontV2-0.1b GPT-Like GPT2 100m VDrontV1 ✅ .gguf .dq
V2 VDrontV2-0.1b-Instruct GPT-Like GPT2 100m VDrontV2-0.1b ✅ .gguf
V2 VDrontV2-0.1b-Title GPT-Like GPT2 100m VDrontV2-0.1b
V3 VDrontV3-Mini GPT-Like/MoE-custom GPT2 200m(100m per token) VDrontV2-0.1b

Most models are released under CC0-1.0 or MIT; this is indicated on the model card.