[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fs4kdai303i83":3},{"slug":4,"category":5,"publishedAt":6,"titleZh":7,"titleEn":8,"summaryZh":9,"summaryEn":10,"models":11,"hardwares":12,"bodyZh":13,"bodyEn":14,"sourceName":15,"sourceUrl":16,"updatedAt":6},"snapdragon-hexagon-plamo3-output-failure-2026-11-09","RUNTIME","2026-11-08T23:05:26.026Z","骁龙 Hexagon 社区故障报告：PLaMo 3 能加载、能跑分，却输出空答案","Community report: PLaMo 3 loads and benchmarks on Snapdragon Hexagon but returns empty answers","11月8日的 llama.cpp #31889 报告：同一平板、同一权重下，CPU 能翻译，Hexagon 全卸载却提前结束。仅一份自报，原因未确认，不能扩大为全部骁龙模型不兼容。","Issue #31889 reports correct CPU translation but empty Hexagon output on the same tablet and weights. It is a single community report, with an unconfirmed cause and limited scope.",[],[],"11 月 8 日，llama.cpp 新增 #31889 故障报告：一位用户在小米 Pad 7 Pro、Snapdragon 8s Gen 3、Android 16 上运行 PLaMo 3 翻译模型，CPU 能得到正常译文，Hexagon HTP0 全卸载时却直接结束生成。截至本次回读，帖子开放且没有评论；这是单一社区自报，不是本站复现，也不是已确认根因的官方公告。\n\n1. 条件足够具体，结论必须同样具体。原文给出 b12202 官方 Android Snapdragon 包，并称 b12125 也复现；权重为 plamo-3-translate-2b-Q4_0.gguf，附来源、哈希和完整提示。CPU 与 HTP0 对照使用同一设备、同一模型。它支持追查这个组合，不能推出所有 Hexagon 设备、所有量化或整个模型家族都失败。\n\n2. 加载成功不是可用性验收。日志显示全部层已卸载，且存在模型及 KV 缓冲；但生成仍为空。作者还比较了不同卸载层数、关闭 Flash Attention、改变部分环境设置：有的仍空，有的出现重复字符。本文不把原帖跑分写成可用速度记录；错误输出的吞吐不能替代正确任务结果，更不能拿来宣传 NPU 提速。\n\n3. 猜测与原因分开。作者提出激活幅度及半精度溢出的可能性，同时明确尚未验证。目前没有足够依据把故障归于某个算子，也没有修复已发布的证据。排错时保留 CPU 基线、准确构建、权重哈希、提示与错误日志，再逐项改变卸载设置；不要把同时改动多项后的现象当作根因证明。\n\n对端侧选型，这份报告的价值不是否定 NPU，而是提醒验收顺序：先检查真实任务输出，再看加速和内存。能识别设备、能加载、能打印吞吐，只说明执行链走到了某一步。该问题继续作为待跟踪故障来源，不将其包装成成功部署实测，也不据此推荐更换硬件。\n\n来源：https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u002Fissues\u002F31889","On November 8, llama.cpp issue #31889 reported a PLaMo 3 translation failure on a Xiaomi Pad 7 Pro with Snapdragon 8s Gen 3 and Android 16. The CPU produces a translation while full Hexagon HTP0 offload ends generation immediately. At retrieval the issue was open with no comments. This is one community report, not an AraoAI reproduction or an official root-cause finding.\n\n1. Preserve the scope. The report names the official b12202 Android Snapdragon package and says b12125 is also affected. It identifies plamo-3-translate-2b-Q4_0.gguf, its source and hash, and the prompt. The CPU and HTP0 comparison uses the same device and weights. It does not establish failure across every Hexagon device, precision or model.\n\n2. Successful loading is not successful deployment. Logs report full layer offload and allocated model and KV buffers. The output is nevertheless empty. Variations in offloaded layers, Flash Attention and environment settings yield empty or repeated-token output. We do not publish the accompanying benchmark as a successful deployment speed: throughput with incorrect answers is not task validity.\n\n3. A hypothesis is not a diagnosis. The author suggests large activations and FP16 overflow as a possibility and explicitly marks it unverified. The evidence does not yet establish a faulty operator or a released fix. Preserve the CPU baseline, build, weight hash, prompt and logs, then vary one offload setting at a time.\n\nFor on-device selection, validate actual task output before comparing acceleration and memory. Device discovery, loading and printed throughput only confirm intermediate steps. This source remains a failure report to track, not a successful run record or evidence that users should replace their hardware.\n\nSource: https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u002Fissues\u002F31889","llama.cpp 社区 issue #31889","https:\u002F\u002Fgithub.com\u002Fggml-org\u002Fllama.cpp\u002Fissues\u002F31889"]