Text Industrial Code Multi
Leaderboard Tasks How MERA works
Return to the leaderboard

GigaChat3-10B-A1.8B

GigaChat Created at 19.11.2025 19:15

Ratings for leaderboard tasks

The table will scroll to the left

Task name Result Place in the rating
Agricultural industry 0.545 3
Medicine and healthcare 0.629 10

ruTXTAgroBench

Metric: F1, Exact Match

Discipline Result
Botany 0.529
General genetics 0.596
The basics of breeding 0.618
Crop production 0.443
General agriculture 0.471
Reclamation farming 0.47
Seed production and seed science 0.498
Forage production and meadow farming 0.467
Farming systems on various agricultural landscapes 0.579
Technologies of cultivation of agricultural crops 0.539

ruTXTAquaBench

Metric: F1, Exact Match

Discipline Result
Industrial aquaculture 0.562
Feeding of fish and other aquatic organisms 0.471
Mariculture. Breeding of crayfish and shrimps. Artificial pearl cultivation 0.449
Ichthyopathology: veterinary medicine, prevention and optimization of fish farming technologies 0.435

ruTXTMedQFundamental

Metric: F1, Exact Match

Discipline Result
Anatomy 0.533
Hygiene 0.563
Histology 0.544
Biophysics 0.522
Biochemistry 0.656
Microbiology 0.719
Biology (parasitology) 0.622
Pharmacology 0.719
Faculty Surgery 0.589
General Surgery 0.641
General chemistry 0.544
Normal physiology 0.67
Bioorganic chemistry 0.541
Pathological anatomy 0.641
Pathophysiology 0.637
Clinical laboratory diagnostics 0.67
Propaedeutics in internal medicine 0.693

Information about the submission

Mera version
1.0.0
Torch Version
2.9.1+cu128
The version of the codebase
435b60a
CUDA version
1
Precision of the model weights
bf8
Seed
1234
Batch
1
Transformers version
4.57.1
The number of GPUs and their type
1 NVIDIA A100-SXM4-80GB
Architecture
local-chat-completions

Team:

GigaChat

Name of the ML model:

GigaChat3-10B-A1.8B

Link to the ML model:

https://huggingface.co/ai-sage/GigaChat3-10B-A1.8B

Model size

10.0B

Model type:

Opened

SFT

MoE

Architecture description:

Представляем `GigaChat3-10B-A1.8B` — инструктивную (instruct) модель семейства GigaChat. Модель основана на архитектуре Mixture-of-Experts (MoE) с 10B общих и 1.8B активных параметров. Архитектура включает **Multi-head Latent Attention (MLA)** и **Multi-Token Prediction (MTP)**, за счет этого модель оптимизированна для высокой пропускной способности (throughput) при инференсе.

License:

MIT

MERA

Useful links

© 2025 MERA. All rights reserved.
User agreement Privacy policy

We use cookies to store data. By continuing to use the site, you consent to work with these files