README.md

metadata

license: mit
datasets:
  - Vikhrmodels/GrandMaster-PRO-MAX
language:
  - en
  - ru
tags:
  - mistral
  - chat
  - conversational
  - transformers
inference:
  parameters:
    temperature: 0
pipeline_tag: text-generation
base_model:
  - mistralai/Mistral-Small-3.1-24B-Instruct-2503
  - anthracite-core/Mistral-Small-3.1-24B-Instruct-2503-HF
library_name: vllm

Zero-Mistral-Small-3.1-24B-Instruct-2503-beta

Zero-Mistral-Small-3.1 is an improved TEXT-ONLY version of mistralai/Mistral-Small-3.1-24B-Instruct-2503 based on NO-VISION anthracite-core/Mistral-Small-3.1-24B-Instruct-2503-HF, primarily adapted for Russian and English languages. The training involved SFT stage on GrandMaster-PRO-MAX dataset.

This is a beta version. Benchmarks and some more fine-tuning coming soon.

Current status:

Trained with lm_head
Train loss: 0.564200
Eval loss: 0.638504