Skip to content
Other

OmniSRE

A LoRA adapter for DevOps and SRE tasks based on Qwen3.8

2 min readupdated

OmniSRE is an open-source systems reliability engineering (SRE) agent designed for automated triage, localization, root cause analysis, and incident remediation. The model is built as a rank-64 LoRA (a small fine-tuning add-on) for the Qwen/Qwen3.8-27B language model. The agent follows a core engineering safety principle: "never write without reading first." This solution is useful for developers, DevOps engineers, and support teams to automate failure analysis.

Key features

  • Rank-64 LoRA (a small add-on) for Qwen/Qwen3.8-27B with 466.9 million trainable parameters.
  • Trained on real incident traces, SRE logs, and troubleshooting instructions (runbooks).
  • Strict approach of reading data before executing any remediation actions.
  • Support for the base model architecture with a frozen vision tower and MTP head.

What you can do

  • Apply the adapter for automated incident analysis and log parsing in DevOps pipelines.
  • Use it as a smart on-call assistant for diagnosing system failures.
  • Build your own chatbot for working with Kubernetes and infrastructure alerts.
  • Integrate into monitoring systems for preliminary localization and root cause analysis of issues.

Example

import torch
from peft import PeftModel
from transformers import AutoModelForImageTextToText, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("Qwen/Qwen3.8-27B", trust_remote_code=True)
base = AutoModelForImageTextToText.from_pretrained(
    "Qwen/Qwen3.8-27B", dtype=torch.bfloat16, device_map="auto", trust_remote_code=True
)
model = PeftModel.from_pretrained(base, "w3joe/OmniSRE")
Read moreDetails
Base model: Qwen/Qwen3.8-27B (64 layers, 262K context)
Adapter: LoRA r = 64, 466.9M parameters (1.68%)
Training: Stage 1a (5,038 traces, loss down to 0.175), Phase 1b (98.5% token accuracy)
Hardware: 4 × H100 BF16 for training · 1 × H100 / L40S FP8 for serving
License: Apache 2.0
Examples
Discuss Telegram 0

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 51 GB · ready files57 GB
  • FP8 / INT8recommendedweights 25 GB29 GB
  • 4-bitweights 14 GB16 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.

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.

    # w3joe/OmniSRE
    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 w3joe/OmniSRE --local-dir ~/models/OmniSRE
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to OmniSRE
  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 Qwen3.8-27B (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 2All on Hugging Face

About the model

w3joe/OmniSRE
Author
w3joe
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

Language models

OnyX-9B

OnyX-9B