A 134.5 GiB mixed 2-bit GGUF of Reflection’s 501B MoE — all 256 experts retained — that runs on a 156 GB DRAM workstation with two small GPUs. With a Software+Law engram it gains +7.5 AA points on the held-out test set. The full-retrieval GPQA arm is 173/198 (87.4%), and the AIME retry arm is 29/30 (96.7%), final; its single miss hit the 32K cap. Apache-2.0. This is a quantized derivative of reflection/Beam-1.0-501B-A23B, not a new base model. The filename carries the recognized IQ2XXS variant; the mixed map is gate/up IQ2XXS, down IQ2S, attention Q6K, router F32. Quick links: Model file · Run locally · Mirror CLI · Engram tables · Results Space terrabeam picks the file that fits your machine, downloads it with the matching llama.cpp build and the Tablinum library, and serves an OpenAI-style API at http://127.0.0.1:8080/v1 with the library’s passages added to every question (needs about 150 GB of RAM/VRAM). —memory long-term adds persistent memory. Already running Ollama, LM Studio or vLLM? Use terrabeam proxy —upstream —format tablinum. Setup for Open WebUI, Claude Code, Cursor and other apps: docs/harnesses.md. * Local held-out test 400, minimal reasoning, Beam judge.…
Модальности:
Генерация текста
Области применения:
Диалог / чат
Задача: Генерация текста
Автор: tablinum
Теги: llama.cpp, gguf, beam, mixture-of-experts, moe, quantized, 2-bit, engram
Лайков: 3 | Загрузок: 1,359
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
