AI NEWS.ccAI正在发生 未来正在更新
♡ 我的收藏 0
AI 模型库/patrickbdevaney/MiMo-V2.6-Flash-REAP50-GGUF
MODEL PROFILE · 模型档案

MiMo-V2.6-Flash-REAP50-GGUF

patrickbdevaney · MiMo-V2.6-Flash-REAP50-GGUF

**GGUF quantizations (`MXFP4_MOE` 86.7 GiB, `Q3_K_S` 77.7 GiB, `Q2_K` 64.8 GiB) + multimodal projector (`mmproj`) + 3-layer MTP draft head of 50%-pruned `XiaomiMiMo/MiMo-V2.6-Flash` (315B → 164B total, 11B active).** GGUF quantizations of **`patrickbdevaney/MiMo-V2.6-Flash-REAP50`** — a **50% REAP expert-pruned** derivative of `XiaomiMiMo/MiMo-V2.6-Flash` (**315B → 164B total parameters**, **11B active parameters** per token, **256 → 128 routed experts** per MoE layer across all 59 MoE layers). 1. **`Q3_K_S` and `Q2_K` were quantised with the custom Python/CUDA converter (`scripts/quantize_gguf_variants.py`), not `llama-quantize --imatrix`.** - Because stock `llama.cpp` does not yet support the `mimo2` hybrid-MLA architecture (see Runtime Compatibility below), `llama-quantize` could not be used to produce `Q3_K_S` or `Q2_K`. - Instead, `scripts/quantize_gguf_variants.py` dequantised the `MXFP4_MOE` expert tensors (`blk.*.ffn_{gate,up,down}_exps.weight`) and packed them directly into `GGML_TYPE_Q3_K` (`3.4375 bpw`) and `GGML_TYPE_Q2_K` (`2.625 bpw`) using standard min-max / absmax block scales on GPU. - The file `MiMo-V2.6-Flash-REAP50.imatrix` (`75.8 MiB`, 357 entries) is published in this repository for downstream use once `llama.cpp` lands `mimo2` support, and contains real activation $\mathbb{E}[x^2]$ statistics collected during the 610,304-token REAP calibration pass — **however, those importance weights were NOT used when computing the scales inside the currently uploaded `Q3_K_S.gguf` and `Q2_K.gguf` files.** An earlier bullet on this card described `MiMo-V2.6-Flash-REAP50.imatrix` as "used for Q3_K/Q2_K quantization"; that was inaccurate and has been corrected here. 2. **No post-pruning router or weight "healing" was applied.** - Every surviving expert tensor in `MiMo-V2.6-Flash-REAP50-MXFP4_MOE.gguf` is a lossless repack of the upstream `XiaomiMiMo/MiMo-V2.6-Flash` OCP MXFP4 nibbles and `E8M0` scales, and every router `ffn_gate_inp.weight` (`[128, 4096]`) and `exp_probs_b.bias` (`[128]`) is an exact unmodified slice of the 128 kept experts per layer. 3. **Held-out teacher-vs-student KL / NLL benchmark status: not yet evaluated.** - Unlike `GLM-5.3-Flash-

译文排队中，当前展示来源原文。

视觉问答multimodallicense:apache-2.0
上下文窗口来源未披露
输入 → 输出文本、图像 → 文本
模型权重待核验
档案更新时间2026/10/31 02:41:40
OVERVIEW

模型介绍

**GGUF quantizations (`MXFP4_MOE` 86.7 GiB, `Q3_K_S` 77.7 GiB, `Q2_K` 64.8 GiB) + multimodal projector (`mmproj`) + 3-layer MTP draft head of 50%-pruned `XiaomiMiMo/MiMo-V2.6-Flash` (315B → 164B total, 11B active).** GGUF quantizations of **`patrickbdevaney/MiMo-V2.6-Flash-REAP50`** — a **50% REAP expert-pruned** derivative of `XiaomiMiMo/MiMo-V2.6-Flash` (**315B → 164B total parameters**, **11B active parameters** per token, **256 → 128 routed experts** per MoE layer across all 59 MoE layers). 1. **`Q3_K_S` and `Q2_K` were quantised with the custom Python/CUDA converter (`scripts/quantize_gguf_variants.py`), not `llama-quantize --imatrix`.** - Because stock `llama.cpp` does not yet support the `mimo2` hybrid-MLA architecture (see Runtime Compatibility below), `llama-quantize` could not be used to produce `Q3_K_S` or `Q2_K`. - Instead, `scripts/quantize_gguf_variants.py` dequantised the `MXFP4_MOE` expert tensors (`blk.*.ffn_{gate,up,down}_exps.weight`) and packed them directly into `GGML_TYPE_Q3_K` (`3.4375 bpw`) and `GGML_TYPE_Q2_K` (`2.625 bpw`) using standard min-max / absmax block scales on GPU. - The file `MiMo-V2.6-Flash-REAP50.imatrix` (`75.8 MiB`, 357 entries) is published in this repository for downstream use once `llama.cpp` lands `mimo2` support, and contains real activation $\mathbb{E}[x^2]$ statistics collected during the 610,304-token REAP calibration pass — **however, those importance weights were NOT used when computing the scales inside the currently uploaded `Q3_K_S.gguf` and `Q2_K.gguf` files.** An earlier bullet on this card described `MiMo-V2.6-Flash-REAP50.imatrix` as "used for Q3_K/Q2_K quantization"; that was inaccurate and has been corrected here. 2. **No post-pruning router or weight "healing" was applied.** - Every surviving expert tensor in `MiMo-V2.6-Flash-REAP50-MXFP4_MOE.gguf` is a lossless repack of the upstream `XiaomiMiMo/MiMo-V2.6-Flash` OCP MXFP4 nibbles and `E8M0` scales, and every router `ffn_gate_inp.weight` (`[128, 4096]`) and `exp_probs_b.bias` (`[128]`) is an exact unmodified slice of the 128 kept experts per layer. 3. **Held-out teacher-vs-student KL / NLL benchmark status: not yet evaluated.** - Unlike `GLM-5.3-Flash-

主要特点

特点摘要待依据来源补齐，不凭模型名称推断能力。

局限与注意事项

暂无足够来源支持具体缺点结论。请重点验证实际任务表现、时延、数据政策、许可与版本生命周期。

PRICING & COST

来源报价与费用分析

同一模型在不同渠道、地区和配置下可能采用不同价格。下表独立保留各渠道报价，不将渠道价格视为厂商官方统一价格。

来源 / 渠道输入输出计费单位采集时间
暂未取得可核验报价，请查阅来源渠道。

按月估算基础 Token 费用

—

预估不包含缓存优惠、阶梯定价、多模态附加费用、网络检索、税费及推理额外输出。

INDEPENDENT BENCHMARKS

公开评测与运行配置

暂无已采集的可核验公开评分。模型未上榜不代表能力较低。

查看完整已收录榜单 →

常见问题

开放权重是否代表可以免费商用？

不一定。应核对具体版本许可证、商用门槛和使用政策。自部署还会产生算力和维护成本。

这些价格是实时结算价格吗？

不是实时结算价格，而是带采集时间的来源报价快照。请到对应渠道核对当前版本、区域、折扣和计费方式。

资料来源与可追溯记录

huggingface · patrickbdevaney/MiMo-V2.6-Flash-REAP50-GGUF ↗2026/10/31 02:41:40
huggingface · enrich:3baa5314-ce0d-44d4-af60-51d3fe8ef569 ↗2026/10/30 14:36:59

许可证：apache-2.0。模型介绍与第三方评测均以原来源的具体版本和方法为准。