Skip to content
Other

Mesosfer d24 SFT

custom GPT model for instructions, code, and agent tasks

2 min read

A compact language model based on a custom PyTorch GPT architecture (d24), trained via SFT (Supervised Fine-Tuning on prepared example dialogues and tasks). The authors trained it on a combined d24-tulu-nemotron-mix dataset, merging Tülu 3 instruction-following and code-writing datasets with NVIDIA Nemotron-SFT-Agentic-v4 agent data. As a result, the model can chat, write code, and execute multi-step instructions. This solution is well-suited for developers and hobbyists looking for a lightweight local model with a ready-to-run inference pipeline independent of heavy standard frameworks.

Key features

  • Trained on a mixture of Tülu 3 Instruction Following, Tülu 3 Code, and NVIDIA Nemotron-SFT-Agentic-v4 datasets for code, instructions, and agent scenarios.
  • Built on a custom 24-layer GPT architecture (n_layer=24, n_head=12, n_embd=1536) with a 2048-token context length.
  • Uses a custom RustBPE / tiktoken-based tokenizer with a vocab_size=65536 vocabulary.
  • Delivered in BF16 precision with a file weight of about 5.4 GB, along with a ready-to-use inference.py script for a quick start.

What you can do

  • Run a local console chat or single response generation using the built-in inference.py script.
  • Use it for generating and reviewing basic Python code, such as writing helper functions.
  • Test it in prototypes of simple text assistants and agent chains by adjusting temperature and repetition-penalty settings.

Example

python ~/.cache/mesosfer/inference.py --prompt "Write a Python function to check if a string is a palindrome."
python ~/.cache/mesosfer/inference.py --temperature 0.2 --repetition-penalty 1.08
Read moreDetails
Checkpoint: d24-tulu-v1 (step 6431, val_bpb = 0.3940)
Architecture: 24 layers, 12 attention heads (n_kv_head=12), embedding size 1536, context 2048, window_pattern="L"
File weight: ~5.4 GB in BF16 format, running via Mesosfer with CUDA or ROCm is supported
Language: English
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 5.3 GB · ready files6.4 GB
  • FP8 / INT8recommendedweights 2.7 GB3.4 GB
  • 4-bitweights 1.5 GB2.2 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 5060 Ti 8GB

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # Dummy9898/mesosfer-d24-sft-nemotron
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download Dummy9898/mesosfer-d24-sft-nemotron --local-dir ~/models/mesosfer-d24-sft-nemotron
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of mesosfer-d24-sft-nemotron (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

Dummy9898/mesosfer-d24-sft-nemotron
Author
Dummy9898
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

00

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments