智谱开放权重系列；不同规格的许可分别标注，部署前核对对应版本与容量需求。Zhipu's open-weight series. Licences are listed per variant; check the specific version and capacity requirements before deployment.
通用对话General chat
工具调用Tool use
MIT 许可MIT licence
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出品方为智谱 AI（北京）。Made by Zhipu AI (Beijing).
国内下载Direct access国内可直连direct OKModelScope 官方组织 ZhipuAI（「智谱.AI」）镜像直连可达（2026-10-20 实测 200）；HF 的 zai-org 在 ModelScope 未检索到。ModelScope official org ZhipuAI mirror reachable (measured 200); the zai-org handle itself not found on ModelScope. · 国内镜像入口domestic mirror
备案状态CAC filing已备案Filed：Beijing-ChatGLM-20230821公示名称listed as「智谱清言（ChatGLM）」 · 备案时间filed 2023-08-31 · 首批第 2 行；2025-06 起另有「智谱清言」登记 Beijing-ZhiPuQingYan-20250606S0055。Row 2 of the first batch; since 2025-06 there is also a “ZhiPu QingYan” filing, Beijing-ZhiPuQingYan-20250606S0055.
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
ModelScope 官方组织 ZhipuAI 的 GLM-5.3 镜像（2026-10-20 实测 200）：ModelScope mirror of GLM-5.3 under the official ZhipuAI org (200 verified on 2026-10-20): https://www.modelscope.cn/models/ZhipuAI/GLM-5.3
成员Member 1GLM 5.3（完整版）GLM 5.3 (Full)
754B 总参（MoE，激活数未公开）754B total params (MoE, activation count not disclosed) · 旗舰Flagship · 3 个量化来源3 quant sources
A基础模型卡Base model card规格 · 许可 · 访问门槛 · 参数量三来源交叉核对specs · licence · access · parameter count cross-checked across three sources
⚠ Non-permissive licence — the release page is authoritative. This site does not interpret licence terms or give legal advice.
访问门槛Access
✅ 公开下载 无需申请✅ Public download no application needed
规模Scale
官网标注Vendor states
754B 总参（MoE，激活数未公开）754B total params (MoE, activation count not disclosed)
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
glm-dsa
上下文长度Context length
1,048,576 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（同系列 GLM-5.3-Flash 才是多模态）。HF task tag text-generation; the card does not mention image input (the sibling GLM-5.3-Flash is the multimodal one).
工具调用Tool calling
✓ 支持✓ Supported出处 ↗source ↗模型卡model card · 2026-10-28 模型卡：评测在 Claude Code 等 agent 框架内跑 Terminal-Bench/ALE 等工具类基准，模型在循环中调用工具。Card: evaluated inside agent harnesses (Claude Code) on tool-heavy benchmarks such as Terminal-Bench and ALE, with the model driving tools in the loop.
定位Role
旗舰Flagship
标签Tags
—
官方下载量Downloads
2.4M点赞likes 2k
发布日Released
—
最近更新Last updated
2026-10-14
发布页体积Release-page size
—
许可为 other，须读发布页原文。Licence is "other"; you must read the release page text before using it.
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.
本成员最小发布文件已超过上面的 80 GB 示例容量。先核对可用容量、分配方式与运行器支持，不能从文件大小断言运行方式。The smallest published file exceeds the 80 GB example. Check available capacity, allocation and runtime support; file size does not establish an execution configuration.
此示意筛选没有候选No candidate under this illustrative filter（最小 216.7 GB） (smallest tier 216.7 GB) 示例容量档；不是硬件型号推荐，需另算缓存与运行时。Illustrative capacity; not a hardware recommendation. Budget cache/runtime separately.
12 GB 示例容量12 GB illustrative capacity
此示意筛选没有候选No candidate under this illustrative filter（最小 216.7 GB） (smallest tier 216.7 GB) 示例容量档；不是硬件型号推荐，需另算缓存与运行时。Illustrative capacity; not a hardware recommendation. Budget cache/runtime separately.
16 GB 示例容量16 GB illustrative capacity
此示意筛选没有候选No candidate under this illustrative filter（最小 216.7 GB） (smallest tier 216.7 GB) 示例容量档；不是硬件型号推荐，需另算缓存与运行时。Illustrative capacity; not a hardware recommendation. Budget cache/runtime separately.
24 GB 示例容量24 GB illustrative capacity
此示意筛选没有候选No candidate under this illustrative filter（最小 216.7 GB） (smallest tier 216.7 GB) 示例容量档；不是硬件型号推荐，需另算缓存与运行时。Illustrative capacity; not a hardware recommendation. Budget cache/runtime separately.
48 GB 示例容量48 GB illustrative capacity
此示意筛选没有候选No candidate under this illustrative filter（最小 216.7 GB） (smallest tier 216.7 GB) 示例容量档；不是硬件型号推荐，需另算缓存与运行时。Illustrative capacity; not a hardware recommendation. Budget cache/runtime separately.
80 GB 示例容量80 GB illustrative capacity
此示意筛选没有候选No candidate under this illustrative filter（最小 216.7 GB） (smallest tier 216.7 GB) 示例容量档；不是硬件型号推荐，需另算缓存与运行时。Illustrative capacity; not a hardware recommendation. Budget cache/runtime separately.
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.
此示意筛选没有候选No candidate under this illustrative filter（最小 216.7 GB） (smallest tier 216.7 GB) 示例合计容量；不是单进程可用量，多卡/系统内存不能直接等同一块显存。Illustrative aggregate capacity, not process-available memory. Multiple devices/RAM are not one VRAM pool.
128 GB 合计容量示例128 GB aggregate-capacity example
此示意筛选没有候选No candidate under this illustrative filter（最小 216.7 GB） (smallest tier 216.7 GB) 示例合计容量；不是单进程可用量，多卡/系统内存不能直接等同一块显存。Illustrative aggregate capacity, not process-available memory. Multiple devices/RAM are not one VRAM pool.
256 GB 合计容量示例256 GB aggregate-capacity example
UD-IQ1_SUD-IQ1_S（217 GB · 极限压缩Extreme compression）
示例合计容量；不是单进程可用量，多卡/系统内存不能直接等同一块显存。Illustrative aggregate capacity, not process-available memory. Multiple devices/RAM are not one VRAM pool.
640 GB 合计容量示例640 GB aggregate-capacity example
UD-Q4_K_XLUD-Q4_K_XL（467 GB · 保守档Conservative）
更省：Smaller: IQ4_XSIQ4_XS 367 GB / UD-IQ4_XSUD-IQ4_XS 365 GB 示例合计容量；不是单进程可用量，多卡/系统内存不能直接等同一块显存。Illustrative aggregate capacity, not process-available memory. Multiple devices/RAM are not one VRAM pool.
本系列档位体积跨度 217 GB → 1508 GB。
Tier sizes in this series span 217 GB → 1508 GB.
跨度过大说明同一系列的量化选择空间很宽，这正是量化选型的主要收益所在。
展开：发布文件体积Expand: published file sizes（17 档tiers）
量化档Quant
最小体积Min size
每权重比特bpw
档位Tier
这档意味着什么What this tier means
该档文件数 / 体积跨度Files in tier / size range
最小体积来源
UD-IQ1_SUD-IQ1_S
217 GB
1.70 bpw
极限压缩Extreme compression
能装下就装，质量明显下降Fit it in at the cost of visibly lower quality
「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.
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 glm-dsa，需核对运行器版本支持Declares glm-dsa; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3The release page traces this back to zai-org/GLM-5.3；权重总字节 5951.1 GB（发布页 gguf.total 字段）Total weight bytes 5951.1 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 2 个量化来源（AesSedai 5 档tiers、Coder40-95 12 档tiers）。The series has 2 other quantisation sources (AesSedai 5 tiers, Coder40-95 12 tiers). 本仓 12 档，覆盖最全。 This repo has 12 tiers — the broadest coverage coverage in this series.
存在的意义Why it exists提供 UD-IQ4_XS 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 UD-IQ4_XS. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 规则候选为 UD-IQ4_XS（365 GB），不构成性能结论Rule-based candidate: UD-IQ4_XS (365 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 UD-IQ4_XS（365 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: UD-IQ4_XS (365 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 · 全仓 5542 GB，硬盘紧张的话建议只下需要的 1~2 个档位5542 GB for the whole repo — if disk is tight, download only the one or two tiers you need · 许可是 other，必须自己读发布页原文再决定能不能用The licence is other — read the release page yourself before deciding whether you may use it
AesSedai
other 继承自基座 zai-org/GLM-5.3，量化仓未单独声明 Inherited from base zai-org/GLM-5.3 — the quant repo declares none of its own5 档5 tiers · 1809 GB · 614 下载614 downloads
量化者做了什么What the quantiser did带重要性矩阵（imatrix：/mnt/srv/snowdrift/fp16/GLM-5.3/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: /mnt/srv/snowdrift/fp16/GLM-5.3/imatrix.gguf) — quantises after scoring each layer by activation statistics, usually steadier than plain RTN；声明架构 glm-dsa，需核对运行器版本支持Declares glm-dsa; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3-BF16The release page traces this back to zai-org/GLM-5.3-BF16；权重总字节 1942.7 GB（发布页 gguf.total 字段）Total weight bytes 1942.7 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 2 个量化来源（unsloth 12 档tiers、Coder40-95 12 档tiers）。The series has 2 other quantisation sources (unsloth 12 tiers, Coder40-95 12 tiers). 本仓 5 档，覆盖最少。 This repo has 5 tiers — the narrowest 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（469 GB），不构成性能结论Rule-based candidate: Q4_K_M (469 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（469 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 (469 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· 全仓 1809 GB，硬盘紧张的话建议只下需要的 1~2 个档位1809 GB for the whole repo — if disk is tight, download only the one or two tiers you need · 许可是 other，必须自己读发布页原文再决定能不能用The licence is other — read the release page yourself before deciding whether you may use it · 量化仓未填写许可元数据字段。此处 other 来自基座 zai-org/GLM-5.3 的推断，不是量化者明确授权；请同时核对基座许可、量化仓文件与再分发条件The quant repo does not fill the licence metadata field. The displayed other is inferred from base zai-org/GLM-5.3, not an explicit grant by the quantiser. Check the base licence, quantisation-repo files and redistribution conditions together.
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 glm-dsa，需核对运行器版本支持Declares glm-dsa; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3The release page traces this back to zai-org/GLM-5.3；权重总字节 5951.1 GB（发布页 gguf.total 字段）Total weight bytes 5951.1 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 2 个量化来源（unsloth 12 档tiers、AesSedai 5 档tiers）。The series has 2 other quantisation sources (unsloth 12 tiers, AesSedai 5 tiers). 本仓 12 档，覆盖最全。 This repo has 12 tiers — the broadest coverage coverage in this series.
存在的意义Why it exists提供 UD-IQ4_XS 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 UD-IQ4_XS. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 规则候选为 UD-IQ4_XS（365 GB），不构成性能结论Rule-based candidate: UD-IQ4_XS (365 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 UD-IQ4_XS（365 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: UD-IQ4_XS (365 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 · 全仓 5542 GB，硬盘紧张的话建议只下需要的 1~2 个档位5542 GB for the whole repo — if disk is tight, download only the one or two tiers you need · 许可是 other，必须自己读发布页原文再决定能不能用The licence is other — read the release page yourself before deciding whether you may use it
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.
unsloth第三方量化发布者，具体覆盖与下载量见快照Third-party quantiser; see snapshot coverage and downloads
Position: The broadest quantisation coverage and the highest downloads among third-party quantisers.
What it is good at: publishes GGUF for almost every mainstream base model first, with the most complete tier range — from extreme compression to near lossless. Offers both static and dynamic (the UD- prefix) quantisation.
Why choose the unsloth build:
Broad coverage → the tier you want is usually there
Naming discipline → Q4_K_M means Q4_K_M, with no different name for the same tier
Ships an importance matrix (imatrix) → quality is steadier than plain RTN
The smoothest path to official tutorials and LM Studio / llama.cpp integration
Note: unsloth's dynamic quantisation (UD-Q4_K_XL and similar) is slightly larger and slightly slower at the same tier, in exchange for usually better quality. Whether that trade is worth it depends on whether your bottleneck is VRAM or speed — this is our reading, not a measurement.
<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>
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.
✅ 公开下载 无需申请✅ Public download no application needed
规模Scale
官网标注Vendor states
321B 总参（MoE，激活数未公开）321B total params (MoE, activation count not disclosed)
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
glm5-next / glm5next（各量化仓申报不一） (quant repos declare this differently)
上下文长度Context length
1,048,576 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
✓ 支持✓ Supported出处 ↗source ↗模型卡model card · 2026-10-28 模型卡：GLM-5 系列首个原生多模态；HF 任务标签 image-text-to-text。Card: the first natively multimodal model in the GLM-5 series; HF task tag image-text-to-text.
工具调用Tool calling
✓ 支持✓ Supported出处 ↗source ↗模型卡model card · 2026-10-28 模型卡：评测在 mini-swe-agent、Claude Code、AutomationBench 等 agent 框架内跑，模型在循环中调用工具。Card: evaluated in agent harnesses (mini-swe-agent, Claude Code, AutomationBench), with the model driving tools in the loop.
定位Role
轻量主力Lightweight main
标签Tags
—
官方下载量Downloads
6.4M点赞likes 3k
发布日Released
—
最近更新Last updated
2026-10-15
发布页体积Release-page size
—
模型卡标注 MIT 许可；部署前仍需核对对应版本的许可原文、依赖与使用范围。The model card lists MIT. Check the licence text for the specific version, dependencies and intended use before deployment.
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.
本成员最小发布文件已超过上面的 80 GB 示例容量。先核对可用容量、分配方式与运行器支持，不能从文件大小断言运行方式。The smallest published file exceeds the 80 GB example. Check available capacity, allocation and runtime support; file size does not establish an execution configuration.
此示意筛选没有候选No candidate under this illustrative filter（最小 93.1 GB） (smallest tier 93.1 GB) 示例容量档；不是硬件型号推荐，需另算缓存与运行时。Illustrative capacity; not a hardware recommendation. Budget cache/runtime separately.
12 GB 示例容量12 GB illustrative capacity
此示意筛选没有候选No candidate under this illustrative filter（最小 93.1 GB） (smallest tier 93.1 GB) 示例容量档；不是硬件型号推荐，需另算缓存与运行时。Illustrative capacity; not a hardware recommendation. Budget cache/runtime separately.
16 GB 示例容量16 GB illustrative capacity
此示意筛选没有候选No candidate under this illustrative filter（最小 93.1 GB） (smallest tier 93.1 GB) 示例容量档；不是硬件型号推荐，需另算缓存与运行时。Illustrative capacity; not a hardware recommendation. Budget cache/runtime separately.
24 GB 示例容量24 GB illustrative capacity
此示意筛选没有候选No candidate under this illustrative filter（最小 93.1 GB） (smallest tier 93.1 GB) 示例容量档；不是硬件型号推荐，需另算缓存与运行时。Illustrative capacity; not a hardware recommendation. Budget cache/runtime separately.
48 GB 示例容量48 GB illustrative capacity
此示意筛选没有候选No candidate under this illustrative filter（最小 93.1 GB） (smallest tier 93.1 GB) 示例容量档；不是硬件型号推荐，需另算缓存与运行时。Illustrative capacity; not a hardware recommendation. Budget cache/runtime separately.
80 GB 示例容量80 GB illustrative capacity
此示意筛选没有候选No candidate under this illustrative filter（最小 93.1 GB） (smallest tier 93.1 GB) 示例容量档；不是硬件型号推荐，需另算缓存与运行时。Illustrative capacity; not a hardware recommendation. Budget cache/runtime separately.
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.
此示意筛选没有候选No candidate under this illustrative filter（最小 93.1 GB） (smallest tier 93.1 GB) 示例合计容量；不是单进程可用量，多卡/系统内存不能直接等同一块显存。Illustrative aggregate capacity, not process-available memory. Multiple devices/RAM are not one VRAM pool.
128 GB 合计容量示例128 GB aggregate-capacity example
UD-Q2_K_XLUD-Q2_K_XL（109 GB · 长上下文优先Long context first）
更省：Smaller: UD-IQ2_XXSUD-IQ2_XXS 102 GB / UD-IQ1_MUD-IQ1_M 98 GB 示例合计容量；不是单进程可用量，多卡/系统内存不能直接等同一块显存。Illustrative aggregate capacity, not process-available memory. Multiple devices/RAM are not one VRAM pool.
256 GB 合计容量示例256 GB aggregate-capacity example
UD-Q4_K_XLUD-Q4_K_XL（200 GB · 保守档Conservative）
更省：Smaller: Q4_K_MQ4_K_M 189 GB / Q4_KQ4_K 181 GB 示例合计容量；不是单进程可用量，多卡/系统内存不能直接等同一块显存。Illustrative aggregate capacity, not process-available memory. Multiple devices/RAM are not one VRAM pool.
640 GB 合计容量示例640 GB aggregate-capacity example
Q8_0Q8_0（341 GB · 近似无损Near lossless）
更省：Smaller: UD-Q6_KUD-Q6_K 292 GB / Q5_K_MQ5_K_M 241 GB 示例合计容量；不是单进程可用量，多卡/系统内存不能直接等同一块显存。Illustrative aggregate capacity, not process-available memory. Multiple devices/RAM are not one VRAM pool.
本系列档位体积跨度 93 GB → 642 GB。
Tier sizes in this series span 93 GB → 642 GB.
跨度过大说明同一系列的量化选择空间很宽，这正是量化选型的主要收益所在。
展开：发布文件体积Expand: published file sizes（26 档tiers）
量化档Quant
最小体积Min size
每权重比特bpw
档位Tier
这档意味着什么What this tier means
该档文件数 / 体积跨度Files in tier / size range
最小体积来源
UD-IQ1_SUD-IQ1_S
93 GB
1.70 bpw
极限压缩Extreme compression
能装下就装，质量明显下降Fit it in at the cost of visibly lower quality
「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.
mit 继承自基座Inherited from basezai-org/GLM-5.3-Flash · 量化仓未单独声明许可 · the quant repo declares no licence of its own
1
130 GB
1,048,576
作者标注 3.49bpwauthor-stated 3.49bpw 140 GB
180
2026-09-12
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.
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 glm5-next，需核对运行器版本支持Declares glm5-next; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3-FlashThe release page traces this back to zai-org/GLM-5.3-Flash；权重总字节 2540.4 GB（发布页 gguf.total 字段）Total weight bytes 2540.4 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（vcruz305 2 档tiers、aj9o9 2 档tiers、ggml-org 3 档tiers、batiai 5 档tiers 等）。The series has 9 other quantisation sources (vcruz305 2 tiers, aj9o9 2 tiers, ggml-org 3 tiers, batiai 5 tiers, among others). 本仓 12 档，覆盖最全。 This repo has 12 tiers — the broadest coverage coverage in this series.
存在的意义Why it exists提供 UD-IQ4_XS 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 UD-IQ4_XS. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 规则候选为 UD-IQ4_XS（157 GB），不构成性能结论Rule-based candidate: UD-IQ4_XS (157 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 UD-IQ4_XS（157 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: UD-IQ4_XS (157 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 · 全仓 2366 GB，硬盘紧张的话建议只下需要的 1~2 个档位2366 GB for the whole repo — if disk is tight, download only the one or two tiers you need
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 glm5next，需核对运行器版本支持Declares glm5next; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3-Flash-BF16The release page traces this back to zai-org/GLM-5.3-Flash-BF16；权重总字节 309.6 GB（发布页 gguf.total 字段）Total weight bytes 309.6 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（unsloth 12 档tiers、aj9o9 2 档tiers、ggml-org 3 档tiers、batiai 5 档tiers 等）。The series has 9 other quantisation sources (unsloth 12 tiers, aj9o9 2 tiers, ggml-org 3 tiers, batiai 5 tiers, among others). 本仓 2 档，覆盖中等。 This repo has 2 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（193 GB），不构成性能结论Rule-based candidate: Q4_K_M (193 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（193 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 (193 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 2 个档位；不合预算时需再查其他发布源The snapshot identifies only 2 tiers; inspect other sources if none fits your budget · 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 · 全仓 288 GB，硬盘紧张的话建议只下需要的 1~2 个档位288 GB for the whole repo — if disk is tight, download only the one or two tiers you need
量化者做了什么What the quantiser did带重要性矩阵（imatrix：imatrix_unsloth.gguf_file），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: imatrix_unsloth.gguf_file) — quantises after scoring each layer by activation statistics, usually steadier than plain RTN；声明架构 glm5next，需核对运行器版本支持Declares glm5next; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3-FlashThe release page traces this back to zai-org/GLM-5.3-Flash；权重总字节 199.8 GB（发布页 gguf.total 字段）Total weight bytes 199.8 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（unsloth 12 档tiers、vcruz305 2 档tiers、ggml-org 3 档tiers、batiai 5 档tiers 等）。The series has 9 other quantisation sources (unsloth 12 tiers, vcruz305 2 tiers, ggml-org 3 tiers, batiai 5 tiers, among others). 本仓 2 档，覆盖中等。 This repo has 2 tiers — middling coverage in this series.
存在的意义Why it exists提供 IQ3_XXS 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 IQ3_XXS. 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 IQ3_XXS（112 GB），不构成性能结论Rule-based candidate: IQ3_XXS (112 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 IQ3_XXS（112 GB，长上下文优先）。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: IQ3_XXS (112 GB, Long context first). 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 2 个档位；不合预算时需再查其他发布源The snapshot identifies only 2 tiers; inspect other sources if none fits your budget
量化者做了什么What the quantiser did带重要性矩阵（imatrix：./upload-GLM_5.3_Flash/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: ./upload-GLM_5.3_Flash/imatrix.gguf) — quantises after scoring each layer by activation statistics, usually steadier than plain RTN；声明架构 glm5-next，需核对运行器版本支持Declares glm5-next; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3-Flash-BF16The release page traces this back to zai-org/GLM-5.3-Flash-BF16；权重总字节 425.2 GB（发布页 gguf.total 字段）Total weight bytes 425.2 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（unsloth 12 档tiers、vcruz305 2 档tiers、aj9o9 2 档tiers、batiai 5 档tiers 等）。The series has 9 other quantisation sources (unsloth 12 tiers, vcruz305 2 tiers, aj9o9 2 tiers, batiai 5 tiers, among others). 本仓 3 档，覆盖中等。 This repo has 3 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· 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（181 GB），不构成性能结论Rule-based candidate: Q4_K (181 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（181 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 (181 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 · 全仓 396 GB，硬盘紧张的话建议只下需要的 1~2 个档位396 GB for the whole repo — if disk is tight, download only the one or two tiers you need · 许可是 other，必须自己读发布页原文再决定能不能用The licence is other — read the release page yourself before deciding whether you may use it
量化者做了什么What the quantiser did带重要性矩阵（imatrix：/mnt/nas/batiai-models/GLM-5.3-Flash-workspace/imatrix.dat），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: /mnt/nas/batiai-models/GLM-5.3-Flash-workspace/imatrix.dat) — quantises after scoring each layer by activation statistics, usually steadier than plain RTN；声明架构 glm5next，需核对运行器版本支持Declares glm5next; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3-FlashThe release page traces this back to zai-org/GLM-5.3-Flash；权重总字节 654.8 GB（发布页 gguf.total 字段）Total weight bytes 654.8 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（unsloth 12 档tiers、vcruz305 2 档tiers、aj9o9 2 档tiers、ggml-org 3 档tiers 等）。The series has 9 other quantisation sources (unsloth 12 tiers, vcruz305 2 tiers, aj9o9 2 tiers, ggml-org 3 tiers, among others). 本仓 5 档，覆盖中等。 This repo has 5 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（189 GB），不构成性能结论Rule-based candidate: Q4_K_M (189 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（189 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 (189 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· 全仓 610 GB，硬盘紧张的话建议只下需要的 1~2 个档位610 GB for the whole repo — if disk is tight, download only the one or two tiers you need
AesSedai
mit 继承自基座 zai-org/GLM-5.3-Flash，量化仓未单独声明 Inherited from base zai-org/GLM-5.3-Flash — the quant repo declares none of its own5 档5 tiers · 783 GB · 2k 下载2k downloads
量化者做了什么What the quantiser did带重要性矩阵（imatrix：/mnt/srv/snowdrift/fp16/GLM-5.3-Flash/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: /mnt/srv/snowdrift/fp16/GLM-5.3-Flash/imatrix.gguf) — quantises after scoring each layer by activation statistics, usually steadier than plain RTN；声明架构 glm5-next，需核对运行器版本支持Declares glm5-next; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3-Flash-BF16The release page traces this back to zai-org/GLM-5.3-Flash-BF16；权重总字节 840.3 GB（发布页 gguf.total 字段）Total weight bytes 840.3 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（unsloth 12 档tiers、vcruz305 2 档tiers、aj9o9 2 档tiers、ggml-org 3 档tiers 等）。The series has 9 other quantisation sources (unsloth 12 tiers, vcruz305 2 tiers, aj9o9 2 tiers, ggml-org 3 tiers, among others). 本仓 5 档，覆盖中等。 This repo has 5 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（202 GB），不构成性能结论Rule-based candidate: Q4_K_M (202 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（202 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 (202 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· 全仓 783 GB，硬盘紧张的话建议只下需要的 1~2 个档位783 GB for the whole repo — if disk is tight, download only the one or two tiers you need · 量化仓未填写许可元数据字段。此处 mit 来自基座 zai-org/GLM-5.3-Flash 的推断，不是量化者明确授权；请同时核对基座许可、量化仓文件与再分发条件The quant repo does not fill the licence metadata field. The displayed mit is inferred from base zai-org/GLM-5.3-Flash, not an explicit grant by the quantiser. Check the base licence, quantisation-repo files and redistribution conditions together.
量化者做了什么What the quantiser did带重要性矩阵（imatrix：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: imatrix.gguf) — quantises after scoring each layer by activation statistics, usually steadier than plain RTN；声明架构 glm5next，需核对运行器版本支持Declares glm5next; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3-FlashThe release page traces this back to zai-org/GLM-5.3-Flash；权重总字节 451.1 GB（发布页 gguf.total 字段）Total weight bytes 451.1 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（unsloth 12 档tiers、vcruz305 2 档tiers、aj9o9 2 档tiers、ggml-org 3 档tiers 等）。The series has 9 other quantisation sources (unsloth 12 tiers, vcruz305 2 tiers, aj9o9 2 tiers, ggml-org 3 tiers, among others). 本仓 4 档，覆盖中等。 This repo has 4 tiers — middling coverage in this series.
存在的意义Why it exists提供 IQ4_XS 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 IQ4_XS. 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 IQ4_XS（167 GB），不构成性能结论Rule-based candidate: IQ4_XS (167 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 IQ4_XS（167 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: IQ4_XS (167 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· 全仓 420 GB，硬盘紧张的话建议只下需要的 1~2 个档位420 GB for the whole repo — if disk is tight, download only the one or two tiers you need
量化者做了什么What the quantiser did带重要性矩阵（imatrix：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: imatrix.gguf) — quantises after scoring each layer by activation statistics, usually steadier than plain RTN；声明架构 glm5next，需核对运行器版本支持Declares glm5next; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3-FlashThe release page traces this back to zai-org/GLM-5.3-Flash；权重总字节 451.1 GB（发布页 gguf.total 字段）Total weight bytes 451.1 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（unsloth 12 档tiers、vcruz305 2 档tiers、aj9o9 2 档tiers、ggml-org 3 档tiers 等）。The series has 9 other quantisation sources (unsloth 12 tiers, vcruz305 2 tiers, aj9o9 2 tiers, ggml-org 3 tiers, among others). 本仓 4 档，覆盖中等。 This repo has 4 tiers — middling coverage in this series.
存在的意义Why it exists提供 IQ4_XS 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 IQ4_XS. 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 IQ4_XS（167 GB），不构成性能结论Rule-based candidate: IQ4_XS (167 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 IQ4_XS（167 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: IQ4_XS (167 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· 全仓 420 GB，硬盘紧张的话建议只下需要的 1~2 个档位420 GB for the whole repo — if disk is tight, download only the one or two tiers you need
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 glm5next，需核对运行器版本支持Declares glm5next; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3-FlashThe release page traces this back to zai-org/GLM-5.3-Flash；权重总字节 2540.4 GB（发布页 gguf.total 字段）Total weight bytes 2540.4 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（unsloth 12 档tiers、vcruz305 2 档tiers、aj9o9 2 档tiers、ggml-org 3 档tiers 等）。The series has 9 other quantisation sources (unsloth 12 tiers, vcruz305 2 tiers, aj9o9 2 tiers, ggml-org 3 tiers, among others). 本仓 12 档，覆盖最全。 This repo has 12 tiers — the broadest coverage coverage in this series.
存在的意义Why it exists提供 UD-IQ4_XS 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 UD-IQ4_XS. Check base version, method, checksums and runtime support; third-party status alone does not determine quality.
优势Strengths· 规则候选为 UD-IQ4_XS（157 GB），不构成性能结论Rule-based candidate: UD-IQ4_XS (157 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 UD-IQ4_XS（157 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: UD-IQ4_XS (157 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 · 全仓 2366 GB，硬盘紧张的话建议只下需要的 1~2 个档位2366 GB for the whole repo — if disk is tight, download only the one or two tiers you need
ddh0
mit 继承自基座 zai-org/GLM-5.3-Flash，量化仓未单独声明 Inherited from base zai-org/GLM-5.3-Flash — the quant repo declares none of its own1 档1 tiers · 130 GB · 180 下载180 downloads
量化者做了什么What the quantiser did带重要性矩阵（imatrix：./imatrices/GLM-5.3-Flash-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: ./imatrices/GLM-5.3-Flash-imatrix.gguf) — quantises after scoring each layer by activation statistics, usually steadier than plain RTN；声明架构 glm5-next，需核对运行器版本支持Declares glm5-next; check support in your runtime version；声明上下文 1,048,576 tokensDeclared context 1,048,576 tokens；发布页声明溯源到 zai-org/GLM-5.3-FlashThe release page traces this back to zai-org/GLM-5.3-Flash；权重总字节 139.8 GB（发布页 gguf.total 字段）Total weight bytes 139.8 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（unsloth 12 档tiers、vcruz305 2 档tiers、aj9o9 2 档tiers、ggml-org 3 档tiers 等）。The series has 9 other quantisation sources (unsloth 12 tiers, vcruz305 2 tiers, aj9o9 2 tiers, ggml-org 3 tiers, among others). 本仓 1 档，覆盖最少。 This repo has 1 tiers — the narrowest coverage in this series.
存在的意义Why it exists提供 作者标注 3.49bpw 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 author-stated 3.49bpw. 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 作者标注 3.49bpw（140 GB），不构成性能结论Rule-based candidate: author-stated 3.49bpw (140 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 作者标注 3.49bpw（140 GB，长上下文优先）。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: author-stated 3.49bpw (140 GB, Long context first). 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 1 个档位；不合预算时需再查其他发布源The snapshot identifies only 1 tiers; inspect other sources if none fits your budget · 量化仓未填写许可元数据字段。此处 mit 来自基座 zai-org/GLM-5.3-Flash 的推断，不是量化者明确授权；请同时核对基座许可、量化仓文件与再分发条件The quant repo does not fill the licence metadata field. The displayed mit is inferred from base zai-org/GLM-5.3-Flash, not an explicit grant by the quantiser. Check the base licence, quantisation-repo files and redistribution conditions together.
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.
unsloth第三方量化发布者，具体覆盖与下载量见快照Third-party quantiser; see snapshot coverage and downloads
Position: The broadest quantisation coverage and the highest downloads among third-party quantisers.
What it is good at: publishes GGUF for almost every mainstream base model first, with the most complete tier range — from extreme compression to near lossless. Offers both static and dynamic (the UD- prefix) quantisation.
Why choose the unsloth build:
Broad coverage → the tier you want is usually there
Naming discipline → Q4_K_M means Q4_K_M, with no different name for the same tier
Ships an importance matrix (imatrix) → quality is steadier than plain RTN
The smoothest path to official tutorials and LM Studio / llama.cpp integration
Note: unsloth's dynamic quantisation (UD-Q4_K_XL and similar) is slightly larger and slightly slower at the same tier, in exchange for usually better quality. Whether that trade is worth it depends on whether your bottleneck is VRAM or speed — this is our reading, not a measurement.
<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: The official quantisation release channel of llama.cpp itself.
Why it matters: ggml is llama.cpp's underlying tensor library, and ggml-org's GGUFs are vetted by the project's core authors. When a new architecture appears (a new sparse MoE, say), ggml-org is usually the first to support it and publish a usable GGUF.
Characteristics:
Tiers lean conservative, prioritising correct loading and inference that does not break
Coverage is not necessarily complete (limited authorial bandwidth — they track only models they care about)
Serves as the authoritative reference for "can this architecture run on llama.cpp at all"
Who it suits: people who want to try a new architecture first-hand, or need to confirm whether llama.cpp supports a given model.
<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>
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.