{"id":30609334,"url":"https://github.com/ByteDance-Seed/seed-oss","last_synced_at":"2025-08-30T04:05:00.242Z","repository":{"id":311000162,"uuid":"1034166942","full_name":"ByteDance-Seed/seed-oss","owner":"ByteDance-Seed","description":null,"archived":false,"fork":false,"pushed_at":"2025-08-21T14:36:20.000Z","size":172,"stargazers_count":323,"open_issues_count":1,"forks_count":11,"subscribers_count":3,"default_branch":"master","last_synced_at":"2025-08-21T14:54:20.260Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"apache-2.0","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/ByteDance-Seed.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null}},"created_at":"2025-08-08T00:48:59.000Z","updated_at":"2025-08-21T14:46:07.000Z","dependencies_parsed_at":"2025-08-21T15:04:58.574Z","dependency_job_id":null,"html_url":"https://github.com/ByteDance-Seed/seed-oss","commit_stats":null,"previous_names":["bytedance-seed/seed-oss"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/ByteDance-Seed/seed-oss","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ByteDance-Seed%2Fseed-oss","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ByteDance-Seed%2Fseed-oss/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ByteDance-Seed%2Fseed-oss/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ByteDance-Seed%2Fseed-oss/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/ByteDance-Seed","download_url":"https://codeload.github.com/ByteDance-Seed/seed-oss/tar.gz/refs/heads/master","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/ByteDance-Seed%2Fseed-oss/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":272800965,"owners_count":24995185,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","status":"online","status_checked_at":"2025-08-30T02:00:09.474Z","response_time":77,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2025-08-30T04:02:01.675Z","updated_at":"2025-08-30T04:05:00.237Z","avatar_url":"https://github.com/ByteDance-Seed.png","language":"Python","funding_links":[],"categories":["NLP","Project List","Python"],"sub_categories":["3. Pretraining","\u003cspan id=\"tool\"\u003eLLM (LLM \u0026 Tool)\u003c/span\u003e"],"readme":"\u003cdiv align=\"center\"\u003e\n 👋 Hi, everyone!\n    \u003cbr\u003e\n    We are \u003cb\u003eByteDance Seed Team.\u003c/b\u003e\n\u003c/div\u003e\n\n\u003cp align=\"center\"\u003e\n  You can get to know us better through the following channels👇\n  \u003cbr\u003e\n  \u003ca href=\"https://seed.bytedance.com/\"\u003e\n    \u003cimg src=\"https://img.shields.io/badge/Website-%231e37ff?style=for-the-badge\u0026logo=bytedance\u0026logoColor=white\"\u003e\u003c/a\u003e\n\u003c/p\u003e\n\n![seed logo](https://github.com/user-attachments/assets/c42e675e-497c-4508-8bb9-093ad4d1f216)\n\n\n# Seed-OSS Open-Source Models\n\u003cp align=\"center\"\u003e\n  \u003ca href=\"https://github.com/ByteDance-Seed/seed-oss\"\u003e\n    \u003cimg src=\"https://img.shields.io/badge/Seed-Project Page-yellow\"\u003e\u003c/a\u003e\n  \u003ca href=\"https://github.com/ByteDance-Seed/seed-oss\"\u003e\n    \u003cimg src=\"https://img.shields.io/badge/Seed-Tech Report Coming Soon-red\"\u003e\u003c/a\u003e\n  \u003ca href=\"https://huggingface.co/collections/ByteDance-Seed/seed-oss-68a609f4201e788db05b5dcd\"\u003e\n    \u003cimg src=\"https://img.shields.io/badge/Seed-Hugging Face-orange\"\u003e\u003c/a\u003e\n  \u003cbr\u003e\n  \u003ca href=\"./LICENSE\"\u003e\n    \u003cimg src=\"https://img.shields.io/badge/License-Apache2.0-blue\"\u003e\u003c/a\u003e\n\u003c/p\u003e\n\n## News\n- [2025/08/20]🔥We release `Seed-OSS-36B-Base` (both with and without synthetic data versions) and `Seed-OSS-36B-Instruct`.\n\n## Introduction\nSeed-OSS is a series of open-source large language models developed by ByteDance's Seed Team, designed for powerful long-context, reasoning, agent and general capabilities, and versatile developer-friendly features. Although trained with only 12T tokens, Seed-OSS achieves excellent performance on several popular open benchmarks.\n\nWe release this series of models to the open-source community under the Apache-2.0 license.\n\n\u003e [!NOTE]\n\u003e Seed-OSS is primarily optimized for international (i18n) use cases.\n\n### Key Features\n- **Flexible Control of Thinking Budget**: Allowing users to flexibly adjust the reasoning length as needed. This capability of dynamically controlling the reasoning length enhances inference efficiency in practical application scenarios.\n- **Enhanced Reasoning Capability**: Specifically optimized for reasoning tasks while maintaining balanced and excellent general capabilities.\n- **Agentic Intelligence**: Performs exceptionally well in agentic tasks such as tool-using and issue resolving.\n- **Research-Friendly**: Given that the inclusion of synthetic instruction data in pre-training may affect the post-training research, we released pre-trained models both with and without instruction data, providing the research community with more diverse options.\n- **Native Long Context**: Trained with up-to-512K long context natively.\n\n### Model Summary\n\nSeed-OSS adopts the popular causal language model architecture with RoPE, GQA attention, RMSNorm and SwiGLU activation.\n\n\u003cdiv align=\"center\"\u003e\n\n| | |\n|:---:|:---:|\n| | **Seed-OSS-36B** |\n| **Parameters** | 36B |\n| **Attention** | GQA |\n| **Activation Function** | SwiGLU |\n| **Number of Layers** | 64 |\n| **Number of QKV Heads** | 80 / 8 / 8 |\n| **Head Size** | 128 |\n| **Hidden Size** | 5120 |\n| **Vocabulary Size** | 155K |\n| **Context Length** | 512K |\n| **RoPE Base Frequency** | 1e7 |\n\n\u003c/div\u003e\n\n\n## Evaluation Results\n\n### Seed-OSS-36B-Base\n\nIncorporating synthetic instruction data into pretraining leads to improved performance on most benchmarks. We adopt the version augmented with synthetic instruction data (i.e., *w/ syn.*) as `Seed-OSS-36B-Base`. We also release `Seed-OSS-36B-Base-woSyn` trained without such data (i.e., *w/o syn.*), offering the community a high-performance foundation model unaffected by synthetic instruction data.\n\n\u003cdiv align=\"center\"\u003e\n\u003ctable\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"center\"\u003eBenchmark\u003c/th\u003e\n\u003cth align=\"center\"\u003e\u003csup\u003e\u003ca href=\"https://seed.bytedance.com/en/seed1_6\"\u003eSeed1.6-Base\u003c/a\u003e\u003c/sup\u003e\u003c/th\u003e\n\u003cth align=\"center\"\u003e\u003csup\u003eQwen3-30B-A3B-Base-2507*\u003c/sup\u003e\u003c/th\u003e\n\u003cth align=\"center\"\u003e\u003csup\u003eQwen2.5-32B-Base*\u003c/sup\u003e\u003c/th\u003e\n\u003cth align=\"center\"\u003e\u003csup\u003eSeed-OSS-36B-Base\u003cbr\u003e(\u003ci\u003ew/ syn.\u003c/i\u003e)\u003c/sup\u003e\u003c/th\u003e\n\u003cth align=\"center\"\u003e\u003csup\u003eSeed-OSS-36B-Base-woSyn\u003cbr\u003e(\u003ci\u003ew/o syn.\u003c/i\u003e)\u003c/sup\u003e\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=6\u003e\u003cstrong\u003eKnowledge\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eMMLU-Pro\u003c/td\u003e\n\u003ctd align=\"center\"\u003e70\u003c/td\u003e\n\u003ctd align=\"center\"\u003e59.8\u003c/td\u003e\n\u003ctd align=\"center\"\u003e58.5 (55.1)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e65.1\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e60.4\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eMMLU\u003c/td\u003e\n\u003ctd align=\"center\"\u003e88.8\u003c/td\u003e\n\u003ctd align=\"center\"\u003e82.7\u003c/td\u003e\n\u003ctd align=\"center\"\u003e84 (83.3)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e84.9\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e84.8\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eTriviaQA\u003c/td\u003e\n\u003ctd align=\"center\"\u003e91\u003c/td\u003e\n\u003ctd align=\"center\"\u003e76.2\u003c/td\u003e\n\u003ctd align=\"center\"\u003e76\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e82.1\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e81.9\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eGPQA-D\u003c/td\u003e\n\u003ctd align=\"center\"\u003e43.4\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e37\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e29.3\u003c/td\u003e\n\u003ctd align=\"center\"\u003e31.7\u003c/td\u003e\n\u003ctd align=\"center\"\u003e35.2\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eSimpleQA\u003c/td\u003e\n\u003ctd align=\"center\"\u003e17.1\u003c/td\u003e\n\u003ctd align=\"center\"\u003e7.2\u003c/td\u003e\n\u003ctd align=\"center\"\u003e6.1\u003c/td\u003e\n\u003ctd align=\"center\"\u003e5.8\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e7.4\u003c/b\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=6\u003e\u003cstrong\u003eReasoning\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eBBH\u003c/td\u003e\n\u003ctd align=\"center\"\u003e92.1\u003c/td\u003e\n\u003ctd align=\"center\"\u003e81.4\u003c/td\u003e\n\u003ctd align=\"center\"\u003e79.1 (84.5)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e87.7\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e87.2\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eAGIEval-en\u003c/td\u003e\n\u003ctd align=\"center\"\u003e78\u003c/td\u003e\n\u003ctd align=\"center\"\u003e66.4\u003c/td\u003e\n\u003ctd align=\"center\"\u003e65.6\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e70.7\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e70.1\u003c/td\u003e\n\u003c/tr\u003e\n\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=6\u003e\u003cstrong\u003eMath\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eGSM8K\u003c/td\u003e\n\u003ctd align=\"center\"\u003e93.1\u003c/td\u003e\n\u003ctd align=\"center\"\u003e87\u003c/td\u003e\n\u003ctd align=\"center\"\u003e87.5 (92.9)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e90.8\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e90.3\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eMATH\u003c/td\u003e\n\u003ctd align=\"center\"\u003e72.9\u003c/td\u003e\n\u003ctd align=\"center\"\u003e61.1\u003c/td\u003e\n\u003ctd align=\"center\"\u003e63.5 (57.7)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e81.7\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e61.3\u003c/td\u003e\n\u003c/tr\u003e\n\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=6\u003e\u003cstrong\u003eCoding\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eMBPP\u003c/td\u003e\n\u003ctd align=\"center\"\u003e83.6\u003c/td\u003e\n\u003ctd align=\"center\"\u003e78.8\u003c/td\u003e\n\u003ctd align=\"center\"\u003e77.8 (84.5)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e80.6\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e74.6\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eHumanEval\u003c/td\u003e\n\u003ctd align=\"center\"\u003e78\u003c/td\u003e\n\u003ctd align=\"center\"\u003e70.7\u003c/td\u003e\n\u003ctd align=\"center\"\u003e47.6 (58.5)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e76.8\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e75.6\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\n\u003csup\u003e\n- \u003cb\u003eBold\u003c/b\u003e denotes open-source SOTA.\n\u003c/sup\u003e\u003cbr/\u003e\u003csup\u003e\n- \"*\" indicates that the results in this column are presented in the format of \"reproduced_results (reported_results_if_any)\".\n\u003c/sup\u003e\n\n### Seed-OSS-36B-Instruct\n\n\u003cdiv align=\"center\"\u003e\n\u003ctable\u003e\n\u003cthead\u003e\n\u003ctr\u003e\n\u003cth align=\"center\"\u003eBenchmark\u003c/th\u003e\n\u003cth align=\"center\"\u003e\u003csup\u003e\u003ca href=\"https://console.volcengine.com/ark/region:ark+cn-beijing/model/detail?Id=doubao-seed-1-6-thinking\"\u003eSeed1.6-Thinking-0715\u003c/a\u003e\u003c/sup\u003e\u003c/th\u003e\n\u003cth align=\"center\"\u003e\u003csup\u003eOAI-OSS-20B*\u003c/sup\u003e\u003c/th\u003e\n\u003cth align=\"center\"\u003e\u003csup\u003eQwen3-30B-A3B-Thinking-2507*\u003c/sup\u003e\u003c/th\u003e\n\u003cth align=\"center\"\u003e\u003csup\u003eQwen3-32B*\u003c/sup\u003e\u003c/th\u003e\n\u003cth align=\"center\"\u003e\u003csup\u003eGemma3-27B\u003c/sup\u003e\u003c/th\u003e\n\u003cth align=\"center\"\u003e\u003csup\u003eSeed-OSS-36B-Instruct\u003c/sup\u003e\u003c/th\u003e\n\u003c/tr\u003e\n\u003c/thead\u003e\n\u003ctbody\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=7\u003e\u003cstrong\u003eKnowledge\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eMMLU-Pro\u003c/td\u003e\n\u003ctd align=\"center\"\u003e86.6\u003c/td\u003e\n\u003ctd align=\"center\"\u003e76.2\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e81.9\u003c/ins\u003e (80.9)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e81.8\u003c/td\u003e\n\u003ctd align=\"center\"\u003e67.5\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e82.7\u003c/b\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eMMLU\u003c/td\u003e\n\u003ctd align=\"center\"\u003e90.6\u003c/td\u003e\n\u003ctd align=\"center\"\u003e81.7 (85.3)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e86.9\u003c/ins\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e86.2\u003c/td\u003e\n\u003ctd align=\"center\"\u003e76.9\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e87.4\u003c/b\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eGPQA-D\u003c/td\u003e\n\u003ctd align=\"center\"\u003e80.7\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e72.2\u003c/b\u003e (71.5)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e71.4\u003c/ins\u003e (73.4)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e66.7 (68.4)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e42.4\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e71.4\u003c/ins\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eSuperGPQA\u003c/td\u003e\n\u003ctd align=\"center\"\u003e63.4\u003c/td\u003e\n\u003ctd align=\"center\"\u003e50.1\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e57.3\u003c/b\u003e (56.8)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e49.3\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e55.7\u003c/ins\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eSimpleQA\u003c/td\u003e\n\u003ctd align=\"center\"\u003e23.7\u003c/td\u003e\n\u003ctd align=\"center\"\u003e6.7\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e23.6\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e8.6\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e10\u003c/ins\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e9.7\u003c/td\u003e\n\u003c/tr\u003e\n\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=7\u003e\u003cstrong\u003eMath\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eAIME24\u003c/td\u003e\n\u003ctd align=\"center\"\u003e90.3\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e92.7\u003c/b\u003e (92.1)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e87.7\u003c/td\u003e\n\u003ctd align=\"center\"\u003e82.7 (81.4)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e91.7\u003c/ins\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eAIME25\u003c/td\u003e\n\u003ctd align=\"center\"\u003e86\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e90.3\u003c/b\u003e (91.7)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e81.3 (85)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e73.3 (72.9)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e84.7\u003c/ins\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eBeyondAIME\u003c/td\u003e\n\u003ctd align=\"center\"\u003e60\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e69\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e56\u003c/td\u003e\n\u003ctd align=\"center\"\u003e29\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e65\u003c/ins\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=7\u003e\u003cstrong\u003eReasoning\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eArcAGI V2\u003c/td\u003e\n\u003ctd align=\"center\"\u003e1.16\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e1.74\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e0.87\u003c/td\u003e\n\u003ctd align=\"center\"\u003e0\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e1.45\u003c/ins\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eKORBench\u003c/td\u003e\n\u003ctd align=\"center\"\u003e74.8\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e72.3\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e70.2\u003c/td\u003e\n\u003ctd align=\"center\"\u003e65.4\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e70.6\u003c/ins\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eHLE\u003c/td\u003e\n\u003ctd align=\"center\"\u003e13.9\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e12.7\u003c/b\u003e (10.9)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e8.7\u003c/td\u003e\n\u003ctd align=\"center\"\u003e6.9\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e10.1\u003c/ins\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=7\u003e\u003cstrong\u003eCoding\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eLiveCodeBench v6\u003cbr/\u003e\u003csup\u003e(02/2025-05/2025)\u003c/sup\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e66.8\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e63.8\u003c/ins\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e60.3 (66)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e53.4\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e67.4\u003c/b\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=7\u003e\u003cstrong\u003eInstruction Following\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eIFEval\u003c/td\u003e\n\u003ctd align=\"center\"\u003e86.3\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e92.8\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e88 (88.9)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e88.4 (85)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e90.4\u003c/ins\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e85.8\u003c/td\u003e\n\u003c/tr\u003e\n\n\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=7\u003e\u003cstrong\u003eAgent\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eTAU1-Retail\u003c/td\u003e\n\u003ctd align=\"center\"\u003e63\u003c/td\u003e\n\u003ctd align=\"center\"\u003e(54.8)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e58.7\u003c/ins\u003e (67.8)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e40.9\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e70.4\u003c/b\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eTAU1-Airline\u003c/td\u003e\n\u003ctd align=\"center\"\u003e49\u003c/td\u003e\n\u003ctd align=\"center\"\u003e(38)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e47\u003c/b\u003e (48)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e38\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e46\u003c/ins\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eSWE-Bench Verified\u003cbr/\u003e\u003csup\u003e(OpenHands)\u003c/sup\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e41.8\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e(60.7)\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e31\u003c/td\u003e\n\u003ctd align=\"center\"\u003e23.4\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e56\u003c/ins\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eSWE-Bench Verified\u003cbr/\u003e\u003csup\u003e(AgentLess 4*10)\u003c/sup\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e48.4\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e33.5\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e39.7\u003c/ins\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e47\u003c/b\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eMulti-SWE-Bench\u003c/td\u003e\n\u003ctd align=\"center\"\u003e17.7\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e9.5\u003c/ins\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e7.7\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e17\u003c/b\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=7\u003e\u003cstrong\u003eMultilingualism\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eMMMLU\u003c/td\u003e\n\u003ctd align=\"center\"\u003e84.3\u003c/td\u003e\n\u003ctd align=\"center\"\u003e77.4 (75.7)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e79\u003c/b\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e79\u003c/b\u003e (80.6)\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e78.4\u003c/ins\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=7\u003e\u003cstrong\u003eLong Context\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eRULER\u003cbr/\u003e\u003csup\u003e(128K)\u003c/sup\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e94.5\u003c/td\u003e\n\u003ctd align=\"center\"\u003e78.7\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cins\u003e94.5\u003c/ins\u003e\u003c/td\u003e\n\u003ctd align=\"center\"\u003e77.5\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e\u003cb\u003e94.6\u003c/b\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\n\u003ctr\u003e\n\u003ctd align=\"center\" colspan=7\u003e\u003cstrong\u003eSafety\u003c/strong\u003e\u003c/td\u003e\n\u003c/tr\u003e\n\u003ctr\u003e\n\u003ctd align=\"center\"\u003eAIR-Bench\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e-\u003c/td\u003e\n\u003ctd align=\"center\"\u003e75.6\u003c/td\u003e\n\u003c/tr\u003e\n\u003c/tbody\u003e\n\u003c/table\u003e\n\u003c/div\u003e\n\n\u003csup\u003e\n- \u003cb\u003eBold\u003c/b\u003e denotes open-source SOTA. \u003cins\u003eUnderlined\u003c/ins\u003e indicates the second place in the open-source model.\n\u003c/sup\u003e\u003cbr/\u003e\u003csup\u003e\n- \"*\" indicates that the results in this column are presented in the format of \"reproduced_results (reported_results_if_any)\". Some results have been omitted due to the failure of the evaluation run.\n\u003c/sup\u003e\u003cbr/\u003e\u003csup\u003e\n- The results of Gemma3-27B are sourced directly from its technical report.\n\u003c/sup\u003e\u003cbr/\u003e\u003csup\u003e\n- The results of ArcAGI-V2 were measured on the official evaluation set, which was not involved in the training process.\n\u003c/sup\u003e\u003cbr/\u003e\u003csup\u003e\n- Generation configs for Seed-OSS-36B-Instruct: temperature=1.1, top_p=0.95. Specifically, for Taubench, temperature=1, top_p=0.7.\n\u003c/sup\u003e\u003cbr/\u003e\u003csup\u003e\n\u003c/sup\u003e\n\n\u003e [!NOTE]\n\u003e We recommend sampling with `temperature=1.1` and `top_p=0.95`.\n\n### Thinking Budget\n\nUsers can flexibly specify the model's thinking budget. The figure below shows the performance curves across different tasks as the thinking budget varies. For simpler tasks (such as IFEval), the model's chain of thought (CoT) is shorter, and the score exhibits fluctuations as the thinking budget increases. For more challenging tasks (such as AIME and LiveCodeBench), the model's CoT is longer, and the score improves with an increase in the thinking budget.\n\n![thinking_budget](./figures/thinking_budget.png)\n\nHere is an example with a thinking budget set to 512: during the reasoning process, the model periodically triggers self-reflection to estimate the consumed and remaining budget, and delivers the final response once the budget is exhausted or the reasoning concludes.\n```\n\u003cseed:think\u003e\nGot it, let's try to solve this problem step by step. The problem says ... ...\n\u003cseed:cot_budget_reflect\u003eI have used 129 tokens, and there are 383 tokens remaining for use.\u003c/seed:cot_budget_reflect\u003e\nUsing the power rule, ... ...\n\u003cseed:cot_budget_reflect\u003eI have used 258 tokens, and there are 254 tokens remaining for use.\u003c/seed:cot_budget_reflect\u003e\nAlternatively, remember that ... ...\n\u003cseed:cot_budget_reflect\u003eI have used 393 tokens, and there are 119 tokens remaining for use.\u003c/seed:cot_budget_reflect\u003e\nBecause if ... ...\n\u003cseed:cot_budget_reflect\u003eI have exhausted my token budget, and now I will start answering the question.\u003c/seed:cot_budget_reflect\u003e\n\u003c/seed:think\u003e\nTo solve the problem, we start by using the properties of logarithms to simplify the given equations: (full answer omitted).\n```\n\nIf no thinking budget is set (default mode), Seed-OSS will initiate thinking with unlimited length. If a thinking budget is specified, users are advised to prioritize values that are integer multiples of 512 (e.g., 512, 1K, 2K, 4K, 8K, or 16K), as the model has been extensively trained on these intervals. Models are instructed to output a direct response when the thinking budget is 0, and we recommend setting any budget below 512 to this value.\n\n## Quick Start\n```shell\npip install git+https://github.com/huggingface/transformers.git@56d68c6706ee052b445e1e476056ed92ac5eb383\n```\n\n```python\nfrom transformers import AutoModelForCausalLM, AutoTokenizer\nimport os\nimport re\n\nmodel_name_or_path = \"ByteDance-Seed/Seed-OSS-36B-Instruct\"\n\ntokenizer = AutoTokenizer.from_pretrained(model_name_or_path)\nmodel = AutoModelForCausalLM.from_pretrained(model_name_or_path, device_map=\"auto\")  # You may want to use bfloat16 and/or move to GPU here\nmessages = [\n    {\"role\": \"user\", \"content\": \"How to make pasta?\"},\n]\ntokenized_chat = tokenizer.apply_chat_template(\n  messages, \n  tokenize=True, \n  add_generation_prompt=True, \n  return_tensors=\"pt\", \n  thinking_budget=512 # control the thinking budget\n)\n\noutputs = model.generate(tokenized_chat.to(model.device), max_new_tokens=2048)\n\noutput_text = tokenizer.decode(outputs[0])\n```\n\n## Inference\n\n### Download Model\n\nDownload Seed-OSS checkpoint to `./Seed-OSS-36B-Instruct`\n\n### Transformers\nThe `generate.py` script provides a simple interface for model inference with configurable options.\n\n#### Basic Usage\n```shell\ncd inference\npython3 generate.py --model_path /path/to/model\n```\n\n#### Key Parameters\n| Parameter | Description |\n|-----------|-------------|\n| `--model_path` | Path to the pretrained model directory (required) |\n| `--prompts` | Input prompts (default: sample cooking/code questions) |\n| `--max_new_tokens` | Maximum tokens to generate (default: 4096) |\n| `--attn_implementation` | Attention mechanism: `flash_attention_2` (default) or `eager` |\n| `--load_in_4bit/8bit` | Enable 4-bit/8-bit quantization (reduces memory usage) |\n| `--thinking_budget` | Thinking budget in tokens (default: -1 for unlimited budget) |\n\n#### Quantization Examples\n```shell\n# 8-bit quantization\npython3 generate.py --model_path /path/to/model --load_in_8bit True\n\n# 4-bit quantization\npython3 generate.py --model_path /path/to/model --load_in_4bit True\n```\n\n#### Custom Prompts\n```shell\npython3 generate.py --model_path /path/to/model --prompts \"['What is machine learning?', 'Explain quantum computing']\"\n```\n\n### vLLM\nUse vllm \u003e= 0.10.0 or higher for inference.\n\n- First install vLLM with Seed-OSS support version:\n```shell\nVLLM_USE_PRECOMPILED=1 VLLM_TEST_USE_PRECOMPILED_NIGHTLY_WHEEL=1 pip install git+https://github.com/vllm-project/vllm.git\n```\n\n- Start vLLM API server:\n```shell\npython3 -m vllm.entrypoints.openai.api_server \\\n    --host localhost \\\n    --port 4321 \\\n    --enable-auto-tool-choice \\\n    --tool-call-parser seed_oss \\\n    --trust-remote-code \\\n    --model ./Seed-OSS-36B-Instruct \\\n    --chat-template ./Seed-OSS-36B-Instruct/chat_template.jinja \\\n    --tensor-parallel-size 8 \\\n    --dtype bfloat16 \\\n    --served-model-name seed_oss\n```\n\n- Test with OpenAI client:\n\nChat\n\n```shell\n# no stream\npython3 inference/vllm_chat.py --max_new_tokens 4096 --thinking_budget -1\n# stream\npython3 inference/vllm_chat.py --max_new_tokens 4096 --thinking_budget -1 --stream\n```\n\nTool Call\n```shell\n# no stream\npython3 inference/vllm_tool_call.py --max_new_tokens 4096 --thinking_budget -1\n# stream\npython3 inference/vllm_tool_call.py --max_new_tokens 4096 --thinking_budget -1 --stream\n```\n\n\n## Model Card\nSee [MODEL_CARD](./MODEL_CARD.md).\n\n## License\nThis project is licensed under Apache-2.0. See the [LICENSE](./LICENSE) flie for details.\n\n## Citation\n\n```bibtex\n@misc{seed2025seed-oss,\n  author={ByteDance Seed Team},\n  title={Seed-OSS Open-Source Models},\n  year={2025},\n  howpublished={\\url{https://github.com/ByteDance-Seed/seed-oss}}\n}\n```\n\n## About [ByteDance Seed Team](https://seed.bytedance.com/)\n\nFounded in 2023, ByteDance Seed Team is dedicated to crafting the industry's most advanced AI foundation models. The team aspires to become a world-class research team and make significant contributions to the advancement of science and society.\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FByteDance-Seed%2Fseed-oss","html_url":"https://awesome.ecosyste.ms/projects/github.com%2FByteDance-Seed%2Fseed-oss","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2FByteDance-Seed%2Fseed-oss/lists"}