{"id":"534ebd6c-72f7-4169-b46c-5739c9132ee5","kind":"products","code":"UB-D6899","name":"River","nameBasis":"original-brand","summary":"River AI provides a managed API from training to deployment for developers and teams who want to have their own models; the platform runs LoRA fine-tuning and reinforcement learning based on open source large models. Users can bring in their own data and reward functions to let the model learn specific tasks. The weights and checkpoints after training can be saved, iterated, and services provided through an OpenAI-compatible interface. [Core Function] LoRA low-rank fine-tuning: Update model weights through configurable adapters, train on modules such as attention, MLP, and MoE experts to quickly shape models for specific tasks. Reinforcement learning training: supports the RL process based on custom rewards, completes sampling, scoring, updating and other steps to help the model continue to improve against actual goals. Various open source bases: Provide open source models such as Qwen, Kimi, GLM, DeepSeek, Nemotron, etc., covering parameter scales from approximately 35B to 1T. The specific available models depend on account permissions. Python client and deployment: Sampling, training and saving checkpoints are unified through river-client. The training results can be directly connected to existing applications using OpenAI compatible endpoints.","description":"River AI provides a managed API from training to deployment for developers and teams who want to have their own models; the platform runs LoRA fine-tuning and reinforcement learning based on open source large models. Users can bring in their own data and reward functions to let the model learn specific tasks. The weights and checkpoints after training can be saved, iterated, and services provided through an OpenAI-compatible interface. [Core Function] LoRA low-rank fine-tuning: Update model weights through configurable adapters, train on modules such as attention, MLP, and MoE experts to quickly shape models for specific tasks. Reinforcement learning training: supports the RL process based on custom rewards, completes sampling, scoring, updating and other steps to help the model continue to improve against actual goals. Various open source bases: Provide open source models such as Qwen, Kimi, GLM, DeepSeek, Nemotron, etc., covering parameter scales from approximately 35B to 1T. The specific available models depend on account permissions. Python client and deployment: Sampling, training and saving checkpoints are unified through river-client. The training results can be directly connected to existing applications using OpenAI compatible endpoints.","country":null,"category":"Coding assistants","originalName":"River","eventDate":null,"website":null,"sourceUpdatedAt":"2026-10-27T14:00:18.813Z","englishUpdatedAt":"2026-10-28T00:56:07.026Z","reviewedAt":"2026-10-28T00:56:07.026Z","snapshotAt":"2026-10-28T00:47:51.550Z","sourceHash":"b0422205d8809849066ba88bcffdcfbe2de45ec20ef884436d0793a9cbff43e6","companyId":"09c044ef-44bb-4efb-a354-f28438ca1028","foundedYear":null,"logoUrl":null,"originalText":"River AI 面向希望拥有自有模型的开发者和团队，提供从训练到部署的托管 API；平台基于开源大模型运行 LoRA 微调与强化学习，用户可带入自己的数据和奖励函数，让模型学习特定任务，训练后的权重与检查点可保存、迭代，并通过兼容 OpenAI 的接口提供服务。【核心功能】LoRA 低秩微调：通过可配置的适配器更新模型权重，在注意力、MLP 及 MoE 专家等模块上训练，快速塑造面向具体任务的模型。强化学习训练：支持基于自定义奖励的 RL 流程，完成采样、评分、更新等步骤，帮助模型针对实际目标持续改进。多种开源基座：提供 Qwen、Kimi、GLM、DeepSeek、Nemotron 等开源模型，覆盖约 35B 至 1T 参数规模，具体可用模型取决于账号权限。Python 客户端与部署：通过 river-client 统一完成采样、训练和保存检查点，训练结果可使用 OpenAI 兼容端点直接接入现有应用。","sources":[],"related":[{"id":"09c044ef-44bb-4efb-a354-f28438ca1028","kind":"companies","code":"UB-C12855","name":"River","nameBasis":"source-english","summary":"Your own frontier AI lab. In one API.","description":"Develop frontier language models and agents. Training, reinforcement learning, and inference in one system, on River Cloud or your own GPU cluster.","country":null,"category":null,"originalName":"River","eventDate":null,"website":"https://river.ai/","sourceUpdatedAt":"2026-10-26T00:00:00.000Z","englishUpdatedAt":"2026-11-01T02:03:52.283Z","reviewedAt":"2026-11-01T02:03:52.283Z","snapshotAt":"2026-11-01T00:40:31.077Z","sourceHash":"74d1183b4623ca55137b99e781c7143b56dd0b992e7b639d4b3cfed6a1054a08","companyId":null,"foundedYear":null,"logoUrl":null,"path":"/companies/river"}],"path":"/products/river","structured":{"city":null,"country":null,"location":null,"role":null,"category":"Coding assistants","company":{"name":"River","path":"/companies/river"}},"definition":"River is a product in the Coding assistants category from River.","summaryBasis":"translated","thin":false,"canonical":"https://uniquebase.ai/products/river"}