Launches
Two changed tensors, sharply fewer refusals in a Qwen GGUF test
Community reports 3 sources
Q
Qwen3.8 Pinpoint TE GGUF
Model Nov 2, 2026
Daemonrat's community Qwen3.8-27B GGUF experiment transfers a two-tensor refusal-behavior edit onto an Unsloth quantization.
Visit primary link
2 sources attached
What happened lately
Qwen3.8 Pinpoint TE GGUF timeline
Daemonrat publishes a roughly 16.49 GB GGUF, comparison tools, vectors and an English analysis of a refusal-behavior edit originally credited to Kujira. The author reports transferring the edit onto Unsloth’s UD-Q4_K_M base while leaving 864 of 866 tensor payloads unchanged.
The model card’s refusal counts come from a small, single-seed test with thinking disabled. They do not establish general capability retention or suitability for a deployed service. The repository declares other / qwen-research licensing and directs users to the base model’s terms.