4-bit ROCmFP4 quantizations of XingChen-AGI/Xing4.0-29B-A4B, China Telecom AI’s 29B-total / 4B-active MoE (formerly the TeleChat series), built for AMD Ryzen AI Max+ «Strix Halo» (gfx1151). The single imatrix build lives in its own repo: Xing4.0-29B-A4B-ROCmFP4-FAST-imatrix-GGUF. 1. ROCmFP4 tensor types are ≥ 100, past stock llama.cpp’s GGMLTYPECOUNT. The file is rejected outright — no metadata edit fixes this. 2. xing4 is a new architecture. Xing uses mHC (hyper-connections) + MLA with nembdheadk=192 / nembdheadv=128`, which no existing llama.cpp arch implements. Build it from our fork — the arch lives on a branch, not main: Pinned at commit cbc2cd5 (the port plus the hyper-connection correctness fix). ⚠️ main does not carry xing4 — you must use the xing4-port branch. BF16 reference: 8.3608 ± 0.249. Upstream HF transformers scores 8.4252 on the identical tokens (see Verification). — COHERENT — best quality of the three. The default pick. — FAST — fastest decode and smallest. +1.8% PPL over COHERENT. — STRIXLEAN — not recommended for this model. It is measurably the slowest at decode and worse quality than COHERENT here. STRIXLEAN’s advantage comes from its PLE upgrade, and Xing…
Модальности:
Генерация текста
Области применения:
Диалог / чат
Задача: Генерация текста
Автор: kingjones777
Теги: gguf, rocm, rocmfp4, strix-halo, gfx1151, amd, moe, mla
Лайков: 3 | Загрузок: 528
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
