做網(wǎng)站 就上寶華建站,百度排行榜風(fēng)云榜小說,做風(fēng)險代理案源的網(wǎng)站,網(wǎng)站建設(shè)系統(tǒng)改版MiniMax-M2 為開發(fā)者提供了一個引人注目的解決方案#xff0c;它通過一個擁有 2300 億參數(shù)但僅激活 100 億參數(shù)的專家混合模型#xff0c;來提供編碼和智能體能力。該模型在保持與 Claude Sonnet 4.5 和 GPT-5 等尖端模型相媲美的性能的同時#xff0c;僅需其一小部分計算開…MiniMax-M2 為開發(fā)者提供了一個引人注目的解決方案它通過一個擁有 2300 億參數(shù)但僅激活 100 億參數(shù)的專家混合模型來提供編碼和智能體能力。該模型在保持與 Claude Sonnet 4.5 和 GPT-5 等尖端模型相媲美的性能的同時僅需其一小部分計算開銷因此尤其適合那些對成本控制和低延遲有嚴(yán)格要求的部署場景。模型概覽核心能力面向開發(fā)者的核心價值關(guān)鍵指標(biāo)/詳情智能體性能MiniMax-M2 使用…標(biāo)簽將其推理過程與最終輸出分離。這使模型能夠在多輪交互中保持連貫的思維鏈。擅長需要規(guī)劃、執(zhí)行與調(diào)整的復(fù)雜長程任務(wù)是構(gòu)建自主智能體的理想選擇。在 BrowseComp44.0 分和 ArtifactsBench66.8 分上表現(xiàn)出色超越多個規(guī)模更大的模型。高級編碼專為端到端的開發(fā)者工作流設(shè)計支持包含“編碼-運(yùn)行-修復(fù)”的迭代循環(huán)以及多文件編輯。在 Terminal-Bench46.3 分和 SWE-bench Verified69.4 分基準(zhǔn)測試中極具競爭力。工具調(diào)用能力為復(fù)雜工具集成Shell、瀏覽器、搜索而構(gòu)建在與外部數(shù)據(jù)或系統(tǒng)交互時表現(xiàn)穩(wěn)健可靠。提供專門的工具調(diào)用指南。在 HLE使用工具及其他工具增強(qiáng)基準(zhǔn)測試中表現(xiàn)強(qiáng)勁。卓越的通用智能在通用知識和推理方面保持競爭力確保即使在核心編碼任務(wù)之外也能可靠工作。綜合 AA 智能得分達(dá) 61 分在開源模型中名列前茅。部署指南官方文檔給出了多種運(yùn)行 MiniMax-M2 的方式。以下為官方文檔中推薦的配置實際需求請根據(jù)具體用例調(diào)整4×96 GB GPU支持最長 400 K token 的上下文8×144 GB GPU支持最長 3 M token 的上下文由于我們這次用的是數(shù)據(jù)量比較大的模型所以我們直接用 8×H200 的集群來運(yùn)行它。我們在這里使用的是 DigitalOcean 的 GPU Droplet 云服務(wù)器。目前 DigitalOcean 可以提供 H200單卡或 8 卡、H100單卡或 8 卡等一系列 GPU 服務(wù)器機(jī)型而且支持按需實例和裸金屬。相對于 AWS、GCP 等云平臺DigitalOcean 提供的 GPU 服務(wù)器總體成本更低而且使用簡單無學(xué)習(xí)成本。DigitalOcean 還將在明年年初正式推出基于 NVIDIA B300 的 GPU Droplet 服務(wù)器詳情可直接咨詢 DigitalOcean 中國區(qū)獨(dú)家戰(zhàn)略合作伙伴卓普云 AI Droplet。1. 在 Web 控制臺里apt install python3.10-venvv pip install triton-kernels githttps://github.com/triton-lang/triton.gitv3.5.0#subdirectorypython/triton_kernels vllm --extra-index-url https://wheels.vllm.ai/nightly --prereleaseallow啟動服務(wù)SAFETENSORS_FAST_GPU1 vllm serve MiniMaxAI/MiniMax-M2 --trust-remote-code --tensor-parallel-size 4 --enable-auto-tool-choice --tool-call-parser minimax_m2 --reasoning-parser minimax_m2_append_think安裝 vllm 和 fla-corepip install vllm fla-core然后我們發(fā)送一段請求。curl http://localhost:8000/v1/chat/completions -H Content-Type: application/json -d { model: MiniMaxAI/MiniMax-M2, messages: [ {role: system, content: [{type: text, text: You are a helpful assistant.}]}, {role: user, content: [{type: text, text: Who won the world series in 2020?}]} ] }我們在 Web 控制臺里看到了模型的輸出。我們看下模型給我們輸出了什么content: thinkOkay, the user is asking who won the World Series in 2020. Let me start by recalling the correct information. The Los Angeles Dodgers defeated the Tampa Bay Rays. That was a significant event because it was during the pandemic, so the season had unique adjustments./thinkThe user might be a baseball fan wanting a quick fact, or perhaps someone who has heard about the series and wants confirmation. They didnt mention any specific context, so keeping it straightforward is best. But maybe theyre interested in the historical aspect of that win, especially since it was the Dodgers first title since 1988. Including that detail adds value.I should also note the unusual circumstances of the 2020 season. The COVID-19 pandemic affected everything, with shortened seasons, limited crowds, and other changes. Mentioning these factors could address deeper curiosity about how the event stood out. However, since the user didnt ask for that, I should balance conciseness with relevant context.\n\nThe user possibly a casual viewer looking for a quick answer? Or a student working on a project? They might need the answer for trivia or an assignment. Either way, providing the teams and the year, along with the significance of the Dodgers win, covers the essentials without overcomplicating. No need for excessive details unless they follow up. Keeping the response clear and accurate is key here.\n\n|im-start|The **Los Angeles Dodgers** won the **2020 World Series**. \n\nThey defeated the **Tampa Bay Rays** in 6 games (4-2). This was the Dodgers first World Series championship since 1988. The 2020 World Series was held in Arlington, Texas, at Globe Life Field (the neutral site) due to the COVID-19 pandemic and its impact on the MLB season. \n\nThe decisive game was game 6, played on October 27, 2020, where the Dodgers won 3-1.|im-end|這段輸出展示了 MiniMax-M2 的核心特性交錯思考格式使用think標(biāo)簽將內(nèi)部推理與最終答案分開。高質(zhì)量輸出給出準(zhǔn)確、簡潔且格式規(guī)范的答案既包含關(guān)鍵事實道奇擊敗光芒也補(bǔ)充了相關(guān)背景疫情環(huán)境、中立球場、歷史意義體現(xiàn)了前沿級別的事實檢索與總結(jié)能力。如果你正在構(gòu)建智能體系統(tǒng)、編程工具或者任何既需要高智能又追求高效率的應(yīng)用不妨試用一下這個模型。6. 常見問題QMiniMax-M2 是什么A總參 230 B 的 MoE 模型專為代碼與 Agent 場景設(shè)計每 token 僅激活 10 B兼顧性能與成本。Q支持工具調(diào)用嗎A支持。采用“工具優(yōu)先”設(shè)計可自動判斷何時調(diào)用外部工具。Q什么是“交錯思考”A模型用 … 把中間推理與最終答案分開方便多輪對話中保持連貫的邏輯鏈。Q有哪些 Agent 基準(zhǔn)表現(xiàn)A在 Terminal-Bench 得 46.3 %在 BrowseComp 得 44 %超過很多更大的通用模型。