国产小尺寸代表，2B 规模适合低显存与端侧。A representative Chinese small-size model; the 2B scale suits low VRAM and on-device use.
端侧On-device
小尺寸Small sizes
中文强项Strong Chinese
先说清楚：这个站的数据怎么来的First, where this site's data comes from
✅ 实测字段Measured fields
直接从 Hugging Face 公开 API 读取的客观数据：文件体积（字节）、上下文长度、架构、许可、下载量、更新时间。Objective data read straight from the Hugging Face public API: file sizes in bytes, context length, architecture, licence, downloads, last updated.
✅ 推导Derived
由采集字段按公开规则算出：每权重比特数（依 llama.cpp 命名标准）、档位归属、显存需求区间。推导规则在Computed from the collected fields by published rules: bits per weight (per the llama.cpp naming convention), tier assignment, VRAM requirement range. The rules are fully published inMETHODOLOGY.md 完整公开，任何人可复算。 — anyone can recompute them.
❌ 我们不说的What we do not claim
哪个量化质量更高、哪个更快、哪个模型更强——这些需要实测，本站不跑 benchmark，所以不装作知道。凡我们给出的主观判断，一律显式标注「我们的口径」。Which quantisation is higher quality, which is faster, which model is stronger — these need measurement. This site does not run benchmarks, so it does not pretend to know. Every subjective judgement we do offer is explicitly marked as our own reading.
本站This site does不下载、不托管任何模型权重文件not download or host any model weight files。所有数字都来自公开元数据，点击即可回到发布页核对。. Every figure comes from public metadata, and clicking it takes you back to the release page to verify.采集时间Collected: 2026-10-30（数据快照，非实时） (a snapshot, not live)
出品属地Origin国产domestic出品方为面壁智能 · OpenBMB（北京）。Made by ModelBest · OpenBMB (Beijing).
国内下载Direct access国内可直连direct OKModelScope 官方镜像直连可达（2026-10-20 实测 200）；Hugging Face 直连不可达。Official ModelScope mirror reachable (measured 200); Hugging Face direct unreachable. · 国内镜像入口domestic mirror
备案状态CAC filing已备案Filed：Beijing-MiniCPM-202503180058公示名称listed as「MiniCPM」 · 备案时间filed 2025-04-10 · 2025-06 批次附件第 351 行；此前「面壁露卡 LUCA」也在首批（Beijing-LuKa-20231016）。Row 351 of the 2025-06 batch attachment; ModelBest’s “LuCa” was also in the first batch (Beijing-LuKa-20231016).
备案信息以网信办《生成式人工智能服务已备案信息》公示名单为准（按批次滚动更新，本站只抄编号不解读）；「国内可直连」是本站探测环境中的可达性记录，不是厂商承诺。Filing status is quoted from the CAC public registry (updated batch by batch; this site copies the number and does not interpret it). "Direct access" is this site's own reachability measurement from a mainland network, not a vendor promise.
网信办《生成式人工智能服务已备案信息》滚动公告页（13 个批次附件 PDF，2026-10-20 全量下载检索）：CAC · filed generative-AI service information (rolling announcement page, 13 batch attachment PDFs, full download & search on 2026-10-20): https://www.cac.gov.cn/2024-04/02/c_1713729983803145.htm
✅ 公开下载 无需申请✅ Public download no application needed
规模Scale
官网标注Vendor states
2.5B 密集2.5B dense
官方 API 申报Official API
—
本站实测反推This site, measured
—
权重精度构成Precision breakdown
—
架构Architecture
llama
上下文长度Context length
131,072 tokens
基座Base model
本仓即基座（不是微调版）this repo is the base model, not a fine-tune
任务类型Task type
—
输入模态Input modality
✗ 不支持✗ Not supported出处 ↗source ↗模型卡model card · 2026-10-28 HF 任务标签 text-generation；模型卡未提图像输入。HF task tag text-generation; the card does not mention image input.
工具调用Tool calling
✓ 支持✓ Supported出处 ↗source ↗模型卡model card · 2026-10-28 模型卡「Tool Calling」节：XML 风格工具调用，SGLang 内置 minicpm5 解析器转 OpenAI tool_calls。Card has a "Tool Calling" section: XML-style tool calls converted to OpenAI tool_calls by SGLang's built-in minicpm5 parser.
B该选哪个Which one to pick按你的显存倒推work backwards from your VRAM
下面按发布文件体积 × 1.15 做示意筛选。文件大小来自发布页列表，不是运行内存实测；15% 只是固定练习假设，不能代替随上下文、架构和并发变化的缓存预算。先筛候选，再用容量工具核对。An illustrative filter uses published file size × 1.15. Sizes come from release listings, not measured resident memory. The fixed 15% teaching assumption does not replace a context-, architecture- and concurrency-dependent cache budget. Filter candidates, then check capacity.
系统内存与多设备：单独核对System RAM and multiple devices: check separately
分层加载、CPU 卸载和多设备分配取决于运行器、模型与后端版本。合计容量不等于单进程可用显存；不能仅凭表格保证能运行或推断速度。Layer loading, CPU offload and multi-device allocation depend on runtime, model and backend version. Aggregate capacity is not process-available VRAM; this table establishes neither execution nor speed.
「每权重比特」是按 llama.cpp 公开命名标准从文件名推导的，同一档位在不同模型上的真实值会略有差异。这是Bits per weight is derived from the filename by llama.cpp’s published naming convention; the real value varies slightly between models at the same tier. This isB 级推导grade-B derived，不是实测值。 — not a measured value.
卡片左侧色条 = 推荐档位Colour bar on the left = recommended tier近似无损Near lossless保守Conservative平衡Balanced长上下文Long context极限压缩Extreme compression
「建议档」是本站按体积最小且落在平衡档的规则自动选出的，不是质量排名。官方量化排在最前；带 imatrix 标记的仓库带重要性矩阵数据。The recommended tier is picked by rule — smallest size that still lands in the balanced tier — not by a quality ranking. Official quantisations are listed first; repos marked imatrix ship an importance matrix.
D每个变体详解Every variant in detail七字段seven fields
每个变体按七个字段展开：它是什么 → 量化者做了什么 → 与同系列其他变体的区别 → 存在的意义 → 优势 → 为什么选它 → 已知取舍。最后一项是本站唯一会写主观判断的地方，且都标注了「我们的口径」。Each variant expands along seven fields: what it is → what the quantiser did → how it differs from other variants in this series → why it exists → strengths → why you would pick it → known trade-offs.The last field is the only place this site offers a subjective judgement, and every item is marked as our own reading.
它是什么What it isopenbmb/MiniCPM5-2B-GGUF ↗，提供 3 个 GGUF 量化档，合计 8.6 GB。这是模型方自己发布的，不是第三方。openbmb/MiniCPM5-2B-GGUF ↗, providing 3 GGUF tiers totalling 8.6 GB. Published by the model authors themselves, not a third party.
量化者做了什么What the quantiser did模型作者自行量化，只有覆盖不到的档位才需要考虑第三方Quantised by the model author; third parties are only worth considering for tiers this does not cover；未提供重要性矩阵（imatrix），走的是纯权重量化路线No importance matrix (imatrix); this is pure weight quantisation；声明架构 llama，需核对运行器版本支持Declares llama; check support in your runtime version；声明上下文 131,072 tokensDeclared context 131,072 tokens；权重总字节 9.3 GB（发布页 gguf.total 字段）Total weight bytes 9.3 GB (the release page's gguf.total field)。
与同系列其他变体的区别How it differs from other variants in this series同系列另有 9 个量化来源（bartowski 20 档tiers、Mungert 25 档tiers、NANI-Nithin 30 档tiers、eaddario 8 档tiers 等）。The series has 9 other quantisation sources (bartowski 20 tiers, Mungert 25 tiers, NANI-Nithin 30 tiers, eaddario 8 tiers, among others). 本仓 3 档，覆盖最少。 This repo has 3 tiers — the narrowest coverage in this series.
存在的意义Why it exists由模型发布方提供，便于回查发布说明与来源。官方身份不保证每个任务的质量、速度或兼容性。Published by the model author, allowing its release notes and provenance to be checked. Official status does not establish quality, speed or compatibility for every task.
优势Strengths· 发布来源为模型方，具体支持范围以发布说明为准Published by the model author; consult its release notes for support scope · 规则候选为 Q4_K_M（1.6 GB），不构成性能结论Rule-based candidate: Q4_K_M (1.6 GB), not a performance finding
为什么选它 / 目的Why you would pick it / its purpose文件体积候选为 Q4_K_M（1.6 GB，平衡档）。先用容量预算核对缓存、并发与系统开销，再查发布说明中的版本兼容性。File-size candidate: Q4_K_M (1.6 GB, Balanced). Check cache, concurrency and system overhead with the capacity budget, then verify version compatibility in the release notes.
已知取舍Known trade-offs· 快照仅识别 3 个档位；不合预算时需再查其他发布源The snapshot identifies only 3 tiers; inspect other sources if none fits your budget
量化者做了什么What the quantiser did未提供重要性矩阵（imatrix），走的是纯权重量化路线No importance matrix (imatrix); this is pure weight quantisation；声明架构 llama，需核对运行器版本支持Declares llama; check support in your runtime version；声明上下文 131,072 tokensDeclared context 131,072 tokens；发布页声明溯源到 openbmb/MiniCPM5-2BThe release page traces this back to openbmb/MiniCPM5-2B；权重总字节 36.3 GB（发布页 gguf.total 字段）Total weight bytes 36.3 GB (the release page's gguf.total field)。
与同系列其他变体的区别How it differs from other variants in this series同系列另有 9 个量化来源（openbmb 3 档tiers、Mungert 25 档tiers、NANI-Nithin 30 档tiers、eaddario 8 档tiers 等）。The series has 9 other quantisation sources (openbmb 3 tiers, Mungert 25 tiers, NANI-Nithin 30 tiers, eaddario 8 tiers, among others). 本仓 20 档，覆盖中等。 This repo has 20 tiers — middling coverage in this series.
存在的意义Why it exists提供 Q4_K_L 等量化候选。请核对基座版本、量化说明、校验值与运行器支持；第三方身份本身不说明质量高低。Provides quantisation candidates such as Q4_K_L. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 快照识别 20 个档位，可按体积与运行器支持筛选The snapshot identifies 20 tiers; filter by size and runtime support · 规则候选为 Q4_K_L（1.7 GB），不构成性能结论Rule-based candidate: Q4_K_L (1.7 GB), not a performance finding
为什么选它 / 目的Why you would pick it / its purpose文件体积候选为 Q4_K_L（1.7 GB，平衡档）。先用容量预算核对缓存、并发与系统开销，再查发布说明中的版本兼容性。File-size candidate: Q4_K_L (1.7 GB, Balanced). Check cache, concurrency and system overhead with the capacity budget, then verify version compatibility in the release notes.
已知取舍Known trade-offs· 未记录 imatrix，需另查发布说明；该字段缺失不能单独判定质量No imatrix recorded; inspect release notes. This missing field alone does not determine quality
量化者做了什么What the quantiser did未提供重要性矩阵（imatrix），走的是纯权重量化路线No importance matrix (imatrix); this is pure weight quantisation；声明架构 llama，需核对运行器版本支持Declares llama; check support in your runtime version；声明上下文 131,072 tokensDeclared context 131,072 tokens；权重总字节 61.8 GB（发布页 gguf.total 字段）Total weight bytes 61.8 GB (the release page's gguf.total field)。
与同系列其他变体的区别How it differs from other variants in this series同系列另有 9 个量化来源（openbmb 3 档tiers、bartowski 20 档tiers、NANI-Nithin 30 档tiers、eaddario 8 档tiers 等）。The series has 9 other quantisation sources (openbmb 3 tiers, bartowski 20 tiers, NANI-Nithin 30 tiers, eaddario 8 tiers, among others). 本仓 25 档，覆盖中等。 This repo has 25 tiers — middling coverage in this series.
存在的意义Why it exists提供 Q4_1_L 等量化候选。请核对基座版本、量化说明、校验值与运行器支持；第三方身份本身不说明质量高低。Provides quantisation candidates such as Q4_1_L. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 快照识别 25 个档位，可按体积与运行器支持筛选The snapshot identifies 25 tiers; filter by size and runtime support · 规则候选为 Q4_1_L（1.8 GB），不构成性能结论Rule-based candidate: Q4_1_L (1.8 GB), not a performance finding
为什么选它 / 目的Why you would pick it / its purpose文件体积候选为 Q4_1_L（1.8 GB，平衡档）。先用容量预算核对缓存、并发与系统开销，再查发布说明中的版本兼容性。File-size candidate: Q4_1_L (1.8 GB, Balanced). Check cache, concurrency and system overhead with the capacity budget, then verify version compatibility in the release notes.
已知取舍Known trade-offs· 未记录 imatrix，需另查发布说明；该字段缺失不能单独判定质量No imatrix recorded; inspect release notes. This missing field alone does not determine quality
量化者做了什么What the quantiser did未提供重要性矩阵（imatrix），走的是纯权重量化路线No importance matrix (imatrix); this is pure weight quantisation；声明架构 llama，需核对运行器版本支持Declares llama; check support in your runtime version；声明上下文 131,072 tokensDeclared context 131,072 tokens；发布页声明溯源到 openbmb/MiniCPM5-2BThe release page traces this back to openbmb/MiniCPM5-2B；权重总字节 43.1 GB（发布页 gguf.total 字段）Total weight bytes 43.1 GB (the release page's gguf.total field)。
与同系列其他变体的区别How it differs from other variants in this series同系列另有 9 个量化来源（openbmb 3 档tiers、bartowski 20 档tiers、Mungert 25 档tiers、eaddario 8 档tiers 等）。The series has 9 other quantisation sources (openbmb 3 tiers, bartowski 20 tiers, Mungert 25 tiers, eaddario 8 tiers, among others). 本仓 30 档，覆盖最全。 This repo has 30 tiers — the broadest coverage coverage in this series.
存在的意义Why it exists提供 Q4_1 等量化候选。请核对基座版本、量化说明、校验值与运行器支持；第三方身份本身不说明质量高低。Provides quantisation candidates such as Q4_1. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 快照识别 30 个档位，可按体积与运行器支持筛选The snapshot identifies 30 tiers; filter by size and runtime support · 规则候选为 Q4_1（1.6 GB），不构成性能结论Rule-based candidate: Q4_1 (1.6 GB), not a performance finding
为什么选它 / 目的Why you would pick it / its purpose文件体积候选为 Q4_1（1.6 GB，平衡档）。先用容量预算核对缓存、并发与系统开销，再查发布说明中的版本兼容性。File-size candidate: Q4_1 (1.6 GB, Balanced). Check cache, concurrency and system overhead with the capacity budget, then verify version compatibility in the release notes.
已知取舍Known trade-offs· 未记录 imatrix，需另查发布说明；该字段缺失不能单独判定质量No imatrix recorded; inspect release notes. This missing field alone does not determine quality
量化者做了什么What the quantiser did未提供重要性矩阵（imatrix），走的是纯权重量化路线No importance matrix (imatrix); this is pure weight quantisation；声明架构 llama，需核对运行器版本支持Declares llama; check support in your runtime version；声明上下文 131,072 tokensDeclared context 131,072 tokens；发布页声明溯源到 openbmb/MiniCPM5-2BThe release page traces this back to openbmb/MiniCPM5-2B；权重总字节 17.7 GB（发布页 gguf.total 字段）Total weight bytes 17.7 GB (the release page's gguf.total field)。
与同系列其他变体的区别How it differs from other variants in this series同系列另有 9 个量化来源（openbmb 3 档tiers、bartowski 20 档tiers、Mungert 25 档tiers、NANI-Nithin 30 档tiers 等）。The series has 9 other quantisation sources (openbmb 3 tiers, bartowski 20 tiers, Mungert 25 tiers, NANI-Nithin 30 tiers, among others). 本仓 8 档，覆盖中等。 This repo has 8 tiers — middling coverage in this series.
存在的意义Why it exists提供 Q4_K 等量化候选。请核对基座版本、量化说明、校验值与运行器支持；第三方身份本身不说明质量高低。Provides quantisation candidates such as Q4_K. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 规则候选为 Q4_K（1.4 GB），不构成性能结论Rule-based candidate: Q4_K (1.4 GB), not a performance finding
为什么选它 / 目的Why you would pick it / its purpose文件体积候选为 Q4_K（1.4 GB，平衡档）。先用容量预算核对缓存、并发与系统开销，再查发布说明中的版本兼容性。File-size candidate: Q4_K (1.4 GB, Balanced). Check cache, concurrency and system overhead with the capacity budget, then verify version compatibility in the release notes.
已知取舍Known trade-offs· 未记录 imatrix，需另查发布说明；该字段缺失不能单独判定质量No imatrix recorded; inspect release notes. This missing field alone does not determine quality
量化者做了什么What the quantiser did带重要性矩阵（imatrix：/root/workspace/HF/calibration/minicpm5-2b/minicpm5_2b_combined_gpu0.imatrix.gguf），用激活数据统计各层重要性后再量化，通常比纯 RTN 稳Ships an importance matrix (imatrix: /root/workspace/HF/calibration/minicpm5-2b/minicpm5_2b_combined_gpu0.imatrix.gguf) — quantises after scoring each layer by activation statistics, usually steadier than plain RTN；声明架构 llama，需核对运行器版本支持Declares llama; check support in your runtime version；声明上下文 131,072 tokensDeclared context 131,072 tokens；发布页声明溯源到 openbmb/MiniCPM5-2BThe release page traces this back to openbmb/MiniCPM5-2B；权重总字节 23.2 GB（发布页 gguf.total 字段）Total weight bytes 23.2 GB (the release page's gguf.total field)。
与同系列其他变体的区别How it differs from other variants in this series同系列另有 9 个量化来源（openbmb 3 档tiers、bartowski 20 档tiers、Mungert 25 档tiers、NANI-Nithin 30 档tiers 等）。The series has 9 other quantisation sources (openbmb 3 tiers, bartowski 20 tiers, Mungert 25 tiers, NANI-Nithin 30 tiers, among others). 本仓 16 档，覆盖中等。 This repo has 16 tiers — middling coverage in this series.
存在的意义Why it exists提供 Q4_K_M 等量化候选。请核对基座版本、量化说明、校验值与运行器支持；第三方身份本身不说明质量高低。Provides quantisation candidates such as Q4_K_M. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 元数据记录了 imatrix；质量仍需结合校准资料与任务评估Metadata records an imatrix; quality still requires calibration details and task evaluation · 规则候选为 Q4_K_M（1.6 GB），不构成性能结论Rule-based candidate: Q4_K_M (1.6 GB), not a performance finding
为什么选它 / 目的Why you would pick it / its purpose文件体积候选为 Q4_K_M（1.6 GB，平衡档）。先用容量预算核对缓存、并发与系统开销，再查发布说明中的版本兼容性。File-size candidate: Q4_K_M (1.6 GB, Balanced). Check cache, concurrency and system overhead with the capacity budget, then verify version compatibility in the release notes.
已知取舍Known trade-offs· 本站未评估该版本的质量、吞吐与全部兼容性；元数据完整不等于适合你的任务This build’s quality, throughput and complete compatibility have not been evaluated here; complete metadata does not establish suitability for your task
量化者做了什么What the quantiser did带重要性矩阵（imatrix：models/MiniCPM5-2B-imatrix.gguf），用激活数据统计各层重要性后再量化，通常比纯 RTN 稳Ships an importance matrix (imatrix: models/MiniCPM5-2B-imatrix.gguf) — quantises after scoring each layer by activation statistics, usually steadier than plain RTN；声明架构 llama，需核对运行器版本支持Declares llama; check support in your runtime version；声明上下文 131,072 tokensDeclared context 131,072 tokens；发布页声明溯源到 openbmb/MiniCPM5-2BThe release page traces this back to openbmb/MiniCPM5-2B；权重总字节 17.1 GB（发布页 gguf.total 字段）Total weight bytes 17.1 GB (the release page's gguf.total field)。
与同系列其他变体的区别How it differs from other variants in this series同系列另有 9 个量化来源（openbmb 3 档tiers、bartowski 20 档tiers、Mungert 25 档tiers、NANI-Nithin 30 档tiers 等）。The series has 9 other quantisation sources (openbmb 3 tiers, bartowski 20 tiers, Mungert 25 tiers, NANI-Nithin 30 tiers, among others). 本仓 11 档，覆盖中等。 This repo has 11 tiers — middling coverage in this series.
存在的意义Why it exists提供 Q4_K_M 等量化候选。请核对基座版本、量化说明、校验值与运行器支持；第三方身份本身不说明质量高低。Provides quantisation candidates such as Q4_K_M. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 元数据记录了 imatrix；质量仍需结合校准资料与任务评估Metadata records an imatrix; quality still requires calibration details and task evaluation · 规则候选为 Q4_K_M（1.6 GB），不构成性能结论Rule-based candidate: Q4_K_M (1.6 GB), not a performance finding
为什么选它 / 目的Why you would pick it / its purpose文件体积候选为 Q4_K_M（1.6 GB，平衡档）。先用容量预算核对缓存、并发与系统开销，再查发布说明中的版本兼容性。File-size candidate: Q4_K_M (1.6 GB, Balanced). Check cache, concurrency and system overhead with the capacity budget, then verify version compatibility in the release notes.
已知取舍Known trade-offs· 本站未评估该版本的质量、吞吐与全部兼容性；元数据完整不等于适合你的任务This build’s quality, throughput and complete compatibility have not been evaluated here; complete metadata does not establish suitability for your task
量化者做了什么What the quantiser did未提供重要性矩阵（imatrix），走的是纯权重量化路线No importance matrix (imatrix); this is pure weight quantisation；声明架构 llama，需核对运行器版本支持Declares llama; check support in your runtime version；声明上下文 131,072 tokensDeclared context 131,072 tokens；发布页声明溯源到 openbmb/MiniCPM5-2BThe release page traces this back to openbmb/MiniCPM5-2B；权重总字节 17.9 GB（发布页 gguf.total 字段）Total weight bytes 17.9 GB (the release page's gguf.total field)。
与同系列其他变体的区别How it differs from other variants in this series同系列另有 9 个量化来源（openbmb 3 档tiers、bartowski 20 档tiers、Mungert 25 档tiers、NANI-Nithin 30 档tiers 等）。The series has 9 other quantisation sources (openbmb 3 tiers, bartowski 20 tiers, Mungert 25 tiers, NANI-Nithin 30 tiers, among others). 本仓 8 档，覆盖中等。 This repo has 8 tiers — middling coverage in this series.
存在的意义Why it exists提供 Q4_K_M 等量化候选。请核对基座版本、量化说明、校验值与运行器支持；第三方身份本身不说明质量高低。Provides quantisation candidates such as Q4_K_M. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 规则候选为 Q4_K_M（1.6 GB），不构成性能结论Rule-based candidate: Q4_K_M (1.6 GB), not a performance finding
为什么选它 / 目的Why you would pick it / its purpose文件体积候选为 Q4_K_M（1.6 GB，平衡档）。先用容量预算核对缓存、并发与系统开销，再查发布说明中的版本兼容性。File-size candidate: Q4_K_M (1.6 GB, Balanced). Check cache, concurrency and system overhead with the capacity budget, then verify version compatibility in the release notes.
已知取舍Known trade-offs· 未记录 imatrix，需另查发布说明；该字段缺失不能单独判定质量No imatrix recorded; inspect release notes. This missing field alone does not determine quality
量化者做了什么What the quantiser did未提供重要性矩阵（imatrix），走的是纯权重量化路线No importance matrix (imatrix); this is pure weight quantisation；声明架构 llama，需核对运行器版本支持Declares llama; check support in your runtime version；声明上下文 131,072 tokensDeclared context 131,072 tokens；发布页声明溯源到 openbmb/MiniCPM5-2BThe release page traces this back to openbmb/MiniCPM5-2B；权重总字节 10.9 GB（发布页 gguf.total 字段）Total weight bytes 10.9 GB (the release page's gguf.total field)。
与同系列其他变体的区别How it differs from other variants in this series同系列另有 9 个量化来源（openbmb 3 档tiers、bartowski 20 档tiers、Mungert 25 档tiers、NANI-Nithin 30 档tiers 等）。The series has 9 other quantisation sources (openbmb 3 tiers, bartowski 20 tiers, Mungert 25 tiers, NANI-Nithin 30 tiers, among others). 本仓 6 档，覆盖中等。 This repo has 6 tiers — middling coverage in this series.
存在的意义Why it exists提供 Q4_K_M 等量化候选。请核对基座版本、量化说明、校验值与运行器支持；第三方身份本身不说明质量高低。Provides quantisation candidates such as Q4_K_M. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 规则候选为 Q4_K_M（1.6 GB），不构成性能结论Rule-based candidate: Q4_K_M (1.6 GB), not a performance finding
为什么选它 / 目的Why you would pick it / its purpose文件体积候选为 Q4_K_M（1.6 GB，平衡档）。先用容量预算核对缓存、并发与系统开销，再查发布说明中的版本兼容性。File-size candidate: Q4_K_M (1.6 GB, Balanced). Check cache, concurrency and system overhead with the capacity budget, then verify version compatibility in the release notes.
已知取舍Known trade-offs· 未记录 imatrix，需另查发布说明；该字段缺失不能单独判定质量No imatrix recorded; inspect release notes. This missing field alone does not determine quality
量化者做了什么What the quantiser did未提供重要性矩阵（imatrix），走的是纯权重量化路线No importance matrix (imatrix); this is pure weight quantisation；声明架构 llama，需核对运行器版本支持Declares llama; check support in your runtime version；声明上下文 131,072 tokensDeclared context 131,072 tokens；发布页声明溯源到 openbmb/MiniCPM5-2BThe release page traces this back to openbmb/MiniCPM5-2B；权重总字节 22.8 GB（发布页 gguf.total 字段）Total weight bytes 22.8 GB (the release page's gguf.total field)。
与同系列其他变体的区别How it differs from other variants in this series同系列另有 9 个量化来源（openbmb 3 档tiers、bartowski 20 档tiers、Mungert 25 档tiers、NANI-Nithin 30 档tiers 等）。The series has 9 other quantisation sources (openbmb 3 tiers, bartowski 20 tiers, Mungert 25 tiers, NANI-Nithin 30 tiers, among others). 本仓 12 档，覆盖中等。 This repo has 12 tiers — middling coverage in this series.
存在的意义Why it exists提供 Q4_K_M 等量化候选。请核对基座版本、量化说明、校验值与运行器支持；第三方身份本身不说明质量高低。Provides quantisation candidates such as Q4_K_M. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 规则候选为 Q4_K_M（1.6 GB），不构成性能结论Rule-based candidate: Q4_K_M (1.6 GB), not a performance finding
为什么选它 / 目的Why you would pick it / its purpose文件体积候选为 Q4_K_M（1.6 GB，平衡档）。先用容量预算核对缓存、并发与系统开销，再查发布说明中的版本兼容性。File-size candidate: Q4_K_M (1.6 GB, Balanced). Check cache, concurrency and system overhead with the capacity budget, then verify version compatibility in the release notes.
已知取舍Known trade-offs· 未记录 imatrix，需另查发布说明；该字段缺失不能单独判定质量No imatrix recorded; inspect release notes. This missing field alone does not determine quality
E量化者档案Quantiser profiles选量化者意味着什么what choosing a quantiser means
量化者决定了这个包的档位覆盖、命名规范、是否带 imatrix、什么时候跟进新模型。这一区块说明「选这个量化者的量化意味着什么」。The quantiser decides this build's tier coverage, naming convention, whether it ships imatrix, and how fast it follows new models. This section spells out what choosing that quantiser means.
openbmb模型方发布的 GGUF，仍需核对版本与运行器支持Published by the model authors themselves
在本系列：1 个仓 · 277k 下载In this series: 1 repos · 277k downloads · 含官方仓 · includes the official repo
<p>Published by the model authors themselves</p><p class="muted">This publisher ships no description. What follows is the repo’s own metadata (size, licence, file size, last updated) — no editorialising from us.</p>
Position: A long-standing quantiser with the most complete tier naming and documentation.
Characteristics:
Every tier states what scenario it suits, rather than just dumping files
Naming follows the llama.cpp convention — Q4_K_M is the community's Q4_K_M
The same model often ships both static and imatrix directories
How to choose their builds: read the tier-by-tier README in the repo — it is one of the few places that actually explains *why* one tier loses little quality and another loses more.
Note: repo names often carry a Qwen_ or Meta-Llama_ prefix (Hugging Face repo names must be unique, so quantisers prefix the organisation name to disambiguate) — do not miss the prefix when searching.
<p class="muted">This publisher ships no description. What follows is the repo’s own metadata (size, licence, file size, last updated) — no editorialising from us.</p>
<p class="muted">This publisher ships no description. What follows is the repo’s own metadata (size, licence, file size, last updated) — no editorialising from us.</p>
<p class="muted">This publisher ships no description. What follows is the repo’s own metadata (size, licence, file size, last updated) — no editorialising from us.</p>
<p class="muted">This publisher ships no description. What follows is the repo’s own metadata (size, licence, file size, last updated) — no editorialising from us.</p>
<p class="muted">This publisher ships no description. What follows is the repo’s own metadata (size, licence, file size, last updated) — no editorialising from us.</p>
<p class="muted">This publisher ships no description. What follows is the repo’s own metadata (size, licence, file size, last updated) — no editorialising from us.</p>
<p class="muted">This publisher ships no description. What follows is the repo’s own metadata (size, licence, file size, last updated) — no editorialising from us.</p>
Position: Author of the Mirostat sampling algorithm, long-running low-bit experimental quantisations.
Characteristics:
Willing to publish very low-bit tiers others dare not (IQ2_XXS, Q2_K and the like)
Focused on the quantisation method itself — tries different imatrix files and different sources for the importance matrix
File names carry suffixes like i1 or h2, marking which version of the importance matrix was used
Who it suits: experiments along the lines of "fit the largest possible model into the smallest possible VRAM", or comparisons of how different importance matrices affect quality.
Who it does not suit: everyday use. Quality loss at those very low bits is usually already noticeable — unless you genuinely have no alternative, do not make it your primary build.
按更新时间倒序。更新频繁说明该量化者在跟进新版本；长期不更新则基础模型升版后可能不同步。Sorted by most recent update. Frequent updates mean the quantiser is tracking new releases; a long silence means it may have fallen behind the base model.
本站只负责「该下哪一档」。跑起来是另一件事 —— 下面这几个工具自己的文档里写明了支持本地模型（Ollama / vLLM / llama.cpp / LM Studio），点进去看具体怎么接。This site only answers which tier to download. Running it is a different problem — the tools below document local-model support in their own docs (Ollama / vLLM / llama.cpp / LM Studio). Follow the link to see how.
这些条目的本地模型支持情况由它们自己的官方文档佐证；本站不实测。Local-model support for each entry is evidenced by that project official documentation. This site runs no measurements.
关于数据与许可Data and licensing
本站数据全部来自All data here comes from theHugging Face公开 API，采集时间public API, collected2026-10-30，为快照数据。; this is a snapshot.
本站This site does不下载、不托管任何模型权重文件not download or host any model weight files，只提供发布页链接。许可字段原样引自发布页， — it only links to release pages. Licence fields are quoted verbatim from them,本站不解读许可条款、不构成法律意见this site does not interpret licence terms or give legal advice，一切以 and everything is governed by发布页原文the legal page为准。.
模型名称均为各自权利人的商标，此处仅作描述性使用。本站与上述厂商无隶属或合作关系。Model names are trademarks of their respective owners and are used here only descriptively. This site is not affiliated with or endorsed by any of the vendors above.