[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"product:river":3},{"id":4,"name":5,"slug":6,"slogan":7,"description":8,"organization":5,"images":9,"avatar_url":10,"image_url":11,"website_url":12,"access_links":13,"status":18,"build_by":19,"categories":20,"stats":27,"upload_relationship":30,"create_at":31,"update_at":32,"owner_id":28,"upload_user_id":33,"reviewer_id":34,"video_info":35,"prerelease_cover_url":36,"scheduled_publish_at":35,"user_scheduled":37,"emoji_reacts":38,"tag":35,"pass_review":35,"user_star":37},3110,"River","river","提供托管 API 的基于开源大模型运行与微调的 AI 工具","{\"type\": \"doc\", \"content\": [{\"type\": \"paragraph\", \"content\": [{\"text\": \"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。\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"【核心功能】\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"LoRA 低秩微调：\", \"type\": \"text\", \"marks\": [{\"type\": \"bold\"}]}, {\"text\": \"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。\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"强化学习训练：\", \"type\": \"text\", \"marks\": [{\"type\": \"bold\"}]}, {\"text\": \"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。\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"多种开源基座：\", \"type\": \"text\", \"marks\": [{\"type\": \"bold\"}]}, {\"text\": \"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。\", \"type\": \"text\"}]}, {\"type\": \"paragraph\", \"content\": [{\"text\": \"Python 客户端与部署：\", \"type\": \"text\", \"marks\": [{\"type\": \"bold\"}]}, {\"text\": \"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。\", \"type\": \"text\"}]}]}","https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10005546_1788087300_b0f1d4a7-32f7-4b79-b018-9fc4d0825e2d.png;https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10005546_1788087300_0549153f-b8c6-443d-ba64-5cb379b83b83.png;https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10005546_1788087300_880be2c8-2f9d-41af-a009-98ae4a2346a5.png;https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10005546_1788087300_254fe703-2851-42ba-8ed1-eb1d4380cde9.png;https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10005546_1788087300_f3a3c46d-aec2-4181-b28a-c045eb556228.png","https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10005546_1788087185_e1f5ea16-fb3d-4166-9a76-e101def44b51.png","https:\u002F\u002Fwatcha.tos-cn-beijing.volces.com\u002Fprod\u002Fuser\u002Fuploads\u002F10005546_1788087282_466fc1ee-ae0f-45ce-9b76-029820406f87.jpg","https:\u002F\u002Friver.ai\u002F",{"items":14},[15],{"platform":16,"url":12,"is_primary":17},"WEB",true,"PUBLISHED",[],[21,24],{"id":22,"name":23},6,"编程开发",{"id":25,"name":26},14,"效率工具",{"upvotes":28,"stars":28,"review_count":28,"reply_count":28,"update_at":29},0,"2026-11-12T15:39:54.474Z","THIRD_PARTY","2026-08-30T10:55:09.46Z","2026-08-30T12:26:37.736Z",10005546,10004627,null,"",false,{"reacts":39},[]]