[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-2504":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":10,"language":11,"languages":10,"totalLinesOfCode":10,"stars":12,"forks":13,"watchers":14,"openIssues":15,"contributorsCount":16,"subscribersCount":16,"size":16,"stars1d":17,"stars7d":18,"stars30d":19,"stars90d":16,"forks30d":16,"starsTrendScore":20,"compositeScore":21,"rankGlobal":10,"rankLanguage":10,"license":22,"archived":23,"fork":23,"defaultBranch":24,"hasWiki":23,"hasPages":23,"topics":25,"createdAt":10,"pushedAt":10,"updatedAt":26,"readmeContent":27,"aiSummary":28,"trendingCount":16,"starSnapshotCount":16,"syncStatus":29,"lastSyncTime":30,"discoverSource":31},2504,"timesfm","google-research\u002Ftimesfm","google-research","TimesFM (Time Series Foundation Model) is a pretrained time-series foundation model developed by Google Research for time-series forecasting.","https:\u002F\u002Fresearch.google\u002Fblog\u002Fa-decoder-only-foundation-model-for-time-series-forecasting\u002F",null,"Python",20631,2005,148,143,0,31,153,951,129,119.91,"Apache License 2.0",false,"master",[],"2026-06-12 04:00:14","# TimesFM\n\nTimesFM (Time Series Foundation Model) is a pretrained time-series foundation\nmodel developed by Google Research for time-series forecasting.\n\n*   Paper:\n    [A decoder-only foundation model for time-series forecasting](https:\u002F\u002Farxiv.org\u002Fabs\u002F2310.10688),\n    ICML 2024.\n*   All checkpoints:\n    [TimesFM Hugging Face Collection](https:\u002F\u002Fhuggingface.co\u002Fcollections\u002Fgoogle\u002Ftimesfm-release-66e4be5fdb56e960c1e482a6).\n*   [Google Research blog](https:\u002F\u002Fresearch.google\u002Fblog\u002Fa-decoder-only-foundation-model-for-time-series-forecasting\u002F).\n*   TimesFM in Google 1P Products:\n    *   [BigQuery ML](https:\u002F\u002Fcloud.google.com\u002Fbigquery\u002Fdocs\u002Ftimesfm-model): Enterprise level SQL queries for scalability and reliability.\n    *   [Google Sheets](https:\u002F\u002Fworkspaceupdates.googleblog.com\u002F2026\u002F02\u002Fforecast-data-in-connected-sheets-BigQueryML-TimesFM.html): For your daily spreadsheet. \n    *   [Vertex Model Garden](https:\u002F\u002Fpantheon.corp.google.com\u002Fvertex-ai\u002Fpublishers\u002Fgoogle\u002Fmodel-garden\u002Ftimesfm): Dockerized endpoint for agentic calling.\n\nThis open version is not an officially supported Google product.\n\n**Latest Model Version:** TimesFM 2.5\n\n**Archived Model Versions:**\n\n-   1.0 and 2.0: relevant code archived in the sub directory `v1`. You can `pip\n    install timesfm==1.3.0` to install an older version of this package to load\n    them.\n\n## Update - Apr. 9, 2026\n\nAdded fine-tuning example using HuggingFace Transformers + PEFT (LoRA) — see\n[`timesfm-forecasting\u002Fexamples\u002Ffinetuning\u002F`](timesfm-forecasting\u002Fexamples\u002Ffinetuning\u002F).\nAlso added unit tests (`tests\u002F`) and incorporated several community fixes.\n\nShoutout to [@kashif](https:\u002F\u002Fgithub.com\u002Fkashif) and [@darkpowerxo](https:\u002F\u002Fgithub.com\u002Fdarkpowerxo). \n\n## Update - Mar. 19, 2026\n\nHuge shoutout to [@borealBytes](https:\u002F\u002Fgithub.com\u002FborealBytes) for adding the support for [AGENTS](https:\u002F\u002Fgithub.com\u002Fgoogle-research\u002Ftimesfm\u002Fblob\u002Fmaster\u002FAGENTS.md)! TimesFM [SKILL.md](https:\u002F\u002Fgithub.com\u002Fgoogle-research\u002Ftimesfm\u002Ftree\u002Fmaster\u002Ftimesfm-forecasting) is out.\n\n## Update - Oct. 29, 2025\n\nAdded back the covariate support through XReg for TimesFM 2.5.\n\n\n## Update - Sept. 15, 2025\n\nTimesFM 2.5 is out!\n\nComparing to TimesFM 2.0, this new 2.5 model:\n\n-   uses 200M parameters, down from 500M.\n-   supports up to 16k context length, up from 2048.\n-   supports continuous quantile forecast up to 1k horizon via an optional 30M\n    quantile head.\n-   gets rid of the `frequency` indicator.\n-   has a couple of new forecasting flags.\n\nSince the Sept. 2025 launch, the following improvements have been completed:\n\n1.  ✅ Flax version of the model for faster inference.\n2.  ✅ Covariate support via XReg (see Oct. 2025 update).\n3.  ✅ Documentation, examples, and agent skill (see `timesfm-forecasting\u002F`).\n4.  ✅ Fine-tuning example with LoRA via HuggingFace Transformers + PEFT (see `timesfm-forecasting\u002Fexamples\u002Ffinetuning\u002F`).\n5.  ✅ Unit tests for core layers, configs, and utilities (see `tests\u002F`).\n\n### Install\n\n1.  Clone the repository:\n    ```shell\n    git clone https:\u002F\u002Fgithub.com\u002Fgoogle-research\u002Ftimesfm.git\n    cd timesfm\n    ```\n\n2.  Create a virtual environment and install dependencies using `uv`:\n    ```shell\n    # Create a virtual environment\n    uv venv\n    \n    # Activate the environment\n    source .venv\u002Fbin\u002Factivate\n    \n    # Install the package in editable mode with torch\n    uv pip install -e .[torch]\n    # Or with flax\n    uv pip install -e .[flax]\n    # Or XReg is needed\n    uv pip install -e .[xreg]\n    ```\n\n3. [Optional] Install your preferred `torch` \u002F `jax` backend based on your OS and accelerators\n(CPU, GPU, TPU or Apple Silicon).:\n\n-   [Install PyTorch](https:\u002F\u002Fpytorch.org\u002Fget-started\u002Flocally\u002F).\n-   [Install Jax](https:\u002F\u002Fdocs.jax.dev\u002Fen\u002Flatest\u002Finstallation.html#installation)\n    for Flax.\n\n### Code Example\n\n```python\nimport torch\nimport numpy as np\nimport timesfm\n\ntorch.set_float32_matmul_precision(\"high\")\n\nmodel = timesfm.TimesFM_2p5_200M_torch.from_pretrained(\"google\u002Ftimesfm-2.5-200m-pytorch\")\n\nmodel.compile(\n    timesfm.ForecastConfig(\n        max_context=1024,\n        max_horizon=256,\n        normalize_inputs=True,\n        use_continuous_quantile_head=True,\n        force_flip_invariance=True,\n        infer_is_positive=True,\n        fix_quantile_crossing=True,\n    )\n)\npoint_forecast, quantile_forecast = model.forecast(\n    horizon=12,\n    inputs=[\n        np.linspace(0, 1, 100),\n        np.sin(np.linspace(0, 20, 67)),\n    ],  # Two dummy inputs\n)\npoint_forecast.shape  # (2, 12)\nquantile_forecast.shape  # (2, 12, 10): mean, then 10th to 90th quantiles.\n```\n","TimesFM 是由 Google Research 开发的一个预训练时间序列基础模型，主要用于时间序列预测。该模型采用解码器-only架构，具有200M参数规模，支持长达16k的上下文长度和高达1k的地平线连续分位数预测。TimesFM 提供了多种检查点，并支持通过HuggingFace Transformers进行微调，同时具备单元测试以确保代码质量。它适用于需要高精度时间序列预测的各种场景，如企业级SQL查询、日常电子表格分析等。此外，TimesFM还被集成到Google的一些产品中，比如BigQuery ML、Google Sheets以及Vertex Model Garden，为用户提供便捷的时间序列预测服务。",2,"2026-06-11 02:50:08","top_language"]