> [!WARNING] > The files can change. This line of conversions is still being researched, > so any file here may be replaced by a new revision. Replacements so far: > 2026-10-21 — the 49 routed-expert parts were re-converted with > calibration-weighted Lloyd (same format, recipe id, bits and runtime; KL > to the original halved, below). The > first build (2026-10-20, weights-only codebooks) is commit > 77a2c5f43b66693cdf8a40b64894b8b5d11f52ad. To keep a fixed build, download > a pinned commit: > hf download Baekpica/Beam-1.0-501B-A23B-VQ-GGUF —revision . > > Problems or odd output? Feel free to post them in the > Community tab, > even as rough notes. The prompt, sampling settings and runtime commit help > most. Reflection’s 501B Beam with its full 1,048,576-token context, sized for one 128 GB DGX Spark. The 1.0 TB BF16 checkpoint becomes about 99.9 GB of GGUF parts: 98.3 GB that stay resident and a 1.56 GB BF16 token table that is read in rows from SSD. The routed experts use YoungAi’s 8-dimensional vector quantization (12-bit, and 13-bit on the eight bottom MoE layers), with codebooks, assignments and row gains fitted to activation statistics from a calibration corpus run through…
Модальности:
Генерация текста
Области применения:
Логика и рассуждение Диалог / чат Вызов функций (Tool use)
Задача: Генерация текста
Автор: Baekpica
Теги: gguf, beam, reflection, vector-quantization, vq, mixture-of-experts, 1m-context, harmony
Лайков: 3 | Загрузок: 3,386
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
