Mellum2 Base Pretrain

[!Note] Use this checkpoint as a starting point for research on long-context extension or for 8K-context continued pretraining and fine-tuning. For downstream applications use Base, Instruct, or Thinking instead.

Mellum2 Base Highlights

Mellum2 Base is a pretrained causal language model trained by JetBrains.

The model uses a Mixture-of-Experts architecture with 64 experts and activates 8 experts per token. It uses a combination of sliding-window and full attention layers, with a context length of 8,192 tokens.

This is a checkpoint before long-context extension.

Mellum2 Model Family

This repository contains one checkpoint from the Mellum2 family.

Model Overview

Mellum2 Base has the following features:

• Number of Layers: 28
• Hidden Size: 2304
• Intermediate Size: 7168
• MoE Intermediate Size: 896
• Number of Experts: 64
• Number of Activated Experts: 8
• Number of Attention Heads (GQA): 32 for Q and 4 for KV
• Context Length: 8,192
• Sliding Window: 1,024
• Vocabulary Size: 98,304
• Precision: bfloat16
• License: Apache 2.0

Serving with vLLM

This checkpoint has an 8K context length (long-context extension is applied in Base).

vllm serve JetBrains/Mellum2-12B-A2.5B-Base-Pretrain --max-model-len 8192

Quickstart

Text-Only Input (base model — use the completions endpoint, not chat)

from openai import OpenAI
# Configured by environment variables
client = OpenAI()

completion = client.completions.create(
    model="JetBrains/Mellum2-12B-A2.5B-Base-Pretrain",
    prompt="def fibonacci(n):\n    ",
    max_tokens=4096,
    temperature=0.6,
    top_p=0.95,
    extra_body={
        "top_k": 20,
    },
)
print("Completion:", completion)

Evaluation

Evaluation results are available in the model card. All values are self-reported by JetBrains.

For more details, see the Mellum2 Technical Report.

License

Released under the Apache 2.0 license.