Qwen 3.8 Swift 1.5 Flash-Next, Heretic V2 (uncensored, tuned for coding and reverse engineering), converted from trubisky’s IQ4XS-plus GGUF into the HGN v2** weight format used by halogen-flash-server (AMD Strix Halo, gfx1151). > [!WARNING] > Uncensored model. Refusal behavior has been deliberately removed. You are solely responsible for how you use it. Comply with all applicable laws and the licenses below. > [!NOTE] > Naming: «Heretic V2» is the version of the abliterated weights. «HGN v2» is the format profile of this conversion. They are unrelated version numbers. These weights are in .hgn format and will not load in transformers, vLLM or llama.cpp. They are meant to be mounted into the halogen-flash-server container. This is a community conversion. It has not been qualified by the authors of any upstream project. If you test it, please report results on the discussion tab. Relative error energy of the output’s decoded values against the decoded GGUF source: This measures the second rounding step only, not distance from the original BF16 weights. Short prompts, 512 output tokens, MTP on: ~56 tok/s single user; 93 tok/s aggregate at 4 concurrent; ~1,800 tok/s prefill at 6.1K…
Модальности:
Генерация текста
Языки программирования:
Swift
Задача: Генерация текста
Автор: johnlockejrr
Теги: halogen, qwen4_exp, hgn, halogen-flash, qwen, qwen3_8, moe, heretic
Лайков: 3 | Загрузок: 853
Описание основано на материалах HuggingFace. Перевод выполнен автоматически.
