Skip to content
Computer vision

Looped Qwen3.5-0.8B Visual SRE

Hybrid Multimodal Model for Incident Triage

2 min read

The model tackles Grafana telemetry chart analysis and strict JSON tool calling. Built on the Qwen/Qwen3.5-0.8B architecture (0.85 billion parameters) with added recurrent cyclic layers (Encode-Think-Decode), it was trained on an NVIDIA RTX 4060 consumer graphics card. Its main distinction from the base version is the ability to perform chain-of-thought reasoning directly over the pixels of monitoring charts and bypass text-visual traps. It helps developers and engineers automate the analysis of IT incidents using dashboard screenshots.

Key features

  • Encode-Think-Decode (ETD) architecture with a two-pass run of middle layers through isolated hybrid caches.
  • Trained on 180 multimodal trajectories featuring Grafana dark and light themes, as well as conflicting trap scenarios.
  • Support for KV cache quantization (TurboQuant 4-bit) with visual token protection and the flash-linear-attention (fla) accelerator.
  • Low system requirements: full operation and inference on a single RTX 4060 graphics card with 8 GB VRAM.

What you can do

  • Automatically diagnose outages and failures from Grafana monitoring dashboard screenshots.
  • Set up strict JSON tool calling for integration with incident response systems.
  • Test resilience against visual-textual conflicts in incident management pipelines.

Example

import torch
from transformers import AutoProcessor, Qwen3_5ForConditionalGeneration
from huggingface_hub import snapshot_download

repo_dir = snapshot_download("kaii07/looped-qwen35-0.8b-visual-sre-v2.1")
base = Qwen3_5ForConditionalGeneration.from_pretrained(repo_dir, dtype=torch.bfloat16, device_map="cuda")
print("Модель успешно загружена")
Read moreDetails
Benchmarks: 84.0% accuracy on Hard OOD SRE (vs. 0.0% for base) and 68.0% in BFCL v3.
Hardware: tested on an NVIDIA RTX 4060 Laptop GPU (8.00 GB VRAM), peak training consumption 3.37 GB.
Limitations: requires a specific loading architecture via the author's repository instead of standard AutoModel.
Discuss Telegram 0

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 1.6 GB · ready files2.2 GB
  • FP8 / INT8recommendedweights 0.8 GB1.4 GB
  • 4-bitweights 0.4 GB1.0 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 2060

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.

    # kaii07/looped-qwen35-0.8b-visual-sre-v2.1
    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 kaii07/looped-qwen35-0.8b-visual-sre-v2.1 --local-dir ~/models/looped-qwen35-0.8b-visual-sre-v2.1
    # 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

  1. Qwen3.5-0.8B-Basefamily root
  2. Qwen3.5-0.8B
  3. looped-qwen35-0.8b-visual-sre-v2.1

New posts about versions of Qwen3.5-0.8B-Base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 9All on Hugging Face
Merges 1All on Hugging Face

About the model

kaii07/looped-qwen35-0.8b-visual-sre-v2.1
Author
kaii07
Category
Computer vision
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