The user deployed an abliterated Qwen 3.8 Flash model on a Strix Halo machine, enhancing speed and token prefill. By employing a Fable‑based trick to abliterate the model, they achieved zero refusal and cracked a patron‑tiered game in 38 minutes, producing 60 k output tokens. This result was not attainable with previous setups.
Read original
reddit/r/LocalLLM