Skip to content
Other

eCity-TS-313M

прогноз для водоканалов с интервалами

2 min read

eCity-TS-313M — модель прогнозирования многомерных временных рядов для сценариев водоканалов от eCityAI-Sup, построенная на базе Toto 2.0. Это decoder-only patch-based Transformer на 313M параметров: внимание чередует ось времени (causal) и ось переменных (fully connected), а quantile head выдаёт интервалы. На входе — история до 4096 шагов по 15 переменным, на выходе — 96 шагов и 9 квантилей.

Ключевые особенности

  • Архитектура: decoder-only patch-based Transformer; внимание чередует time axis (causal) и variate axis (fully connected).
  • Quantile head: 9 квантилей (0.1–0.9); медиана — индекс 4, P10/P90 — индексы 0 и 8.
  • Инференс через ONNX Runtime и TensorRT; в репозитории есть готовые артефакты и demo-скрипты.
  • Сценарии водоканала: объём подачи, поток в распределительной сети, давление и уровень резервуаров.
  • Батч до 8 площадок за один проход; B — единственная динамическая размерность.

Пример

import numpy as np
import onnxruntime as ort
ort.preload_dlls()
session = ort.InferenceSession("artifacts/ecity_ts_313m_15x4096.onnx", providers=["CUDAExecutionProvider"])
target = np.asarray(x, dtype=np.float32)[None]  # [1, 15, 4096]
target_mask = np.isfinite(x)[None]
series_ids = np.arange(15, dtype=np.int64)[None]
quantiles = session.run(["quantiles"], {"target": target, "target_mask": target_mask, "series_ids": series_ids})[0]
ПодробнееНа датасете Xi'an Xingfu Lindai: MAE 38.04 против 229.56 у TimesFM и 87.35 у Moirai; RMSE 78.48 против 371.75 и 153.07.
TensorRT на H200, batch=4: P50 142 мс, throughput 28.4 samples/s; ошибки ONNX/TensorRT около 4e-4 max_abs.
ONNX-веса ~1.3 ГБ; TensorRT engine ~1216 МБ и привязан к GPU/CUDA/TensorRT. Нужны CUDA 12.x и cuDNN 9.
Ограничения: фиксированные V=15, T=4096, H=96; только fp32; batch 1–8; текущий артефакт считает все точки наблюдаемыми.
База — Toto 2.0; репозиторий покрывает только инференс, без обучения.
Discuss Telegram 0

Will it run?

Working out the memory needs

The bot is still collecting this model's weight sizes; the estimate will appear here in a few minutes. Meanwhile you can rent a GPU with headroom.

Massed ComputeOpen Massed Compute

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.

    # eCityAI-Sup/eCity-TS
    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 eCityAI-Sup/eCity-TS --local-dir ~/models/eCity-TS
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to eCity-TS
  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 eCity-TS (quants, finetunes, LoRAs) will show up under the bell.

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

About the model

eCityAI-Sup/eCity-TS
Author
eCityAI-Sup
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

Other

FarmSAM

FarmSAM