This repository now bundles glmmoedsa.py (declared via modelfile in config.json`), a fixed runtime for this architecture, and needs it: mlx-lm’s own glmmoedsa builds a lightning indexer on all 78 layers, but GLM-5.2 ships indexer weights on 21 (indexertypes: the other 57 «shared» layers reuse the previous full layer’s top-k selection). mlxlm.load loads leniently and left those 57 indexers at random initialisation. Prompts up to 2048 tokens were unaffected (the indexer is bypassed below indextopk); beyond that, 57 of 78 layers attended to keys chosen by random projections. The bundled runtime implements the schedule as the reference does (plus fp32 indexer scores and router logits and the indexer LayerNorm epsilon); tiny-config parity against transformers` 5.16 is 4e-7 with the sparse path live, and a strict load of this checkpoint reports zero missing and zero unexpected tensors. Details, tests and the GLM-5.3 builds made with it: github.com/PipeNetwork/glm53-mlx. The weights are unchanged. REAP expert-pruned + 4-bit MLX conversion of zai-org/GLM-5.2. Keeps the 128 most-salient experts per layer (of 256) → ~394B params, smaller/faster than the full model. > ⚠️ Quality warning:…
Модальности:
Генерация текста
Области применения:
Диалог / чат
Задача: Генерация текста
Автор: pipenetwork
Теги: mlx, glm_moe_dsa, moe, glm, reap, pruned, conversational, 4-bit
Лайков: 4 | Загрузок: 832
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.