grimoni/Qwen3.8-27B-SSMFIX-UD-Q4_K_XL-GGUF

Qwen3.8-27B-SSMFIX-UD-Q4KXL-GGUF

GGUF quantization of redashes/Qwen3.8-27B-BF16-SSMFIX, derived from Qwen/Qwen3.8-27B, prepared for llama.cpp-compatible runtimes.

Files

• —Qwen3.8-27B-SSMFIX-UD-Q4_K_XL.gguf: main quantized checkpoint
• —sha256.txt: checksum for the GGUF
• —quantization_recipe.txt: quantization recipe used for this build
• —quantize_qwen38_ssmfix_q4xl.log: quantization log
• —convert_qwen38_ssmfix.log: BF16-to-GGUF conversion log

Source

• —Base model family: Qwen/Qwen3.8-27B
• —Input checkpoint used locally: redashes--Qwen3.8-27B-BF16-SSMFIX
• —Quantized from a local BF16 GGUF converted from the original safetensors checkpoint

Quantization notes

• —Target profile: Q4_K_XL-like
• —Output size is about 15 GB
• —The quantization used an importance matrix compatible with Qwen3.8-27B
• —The MTP head (blk.64) is not covered by the downloaded imatrix, so nextn.eh_proj was stored as Q4_0

Local validation

Validated locally on August 19, 2026 with:

• —GGUF header and metadata checks
• —llama.cpp loading on a local NVIDIA GPU
• —HTTP inference through llama-server

Runtime

This file is intended for llama.cpp and compatible tools that support GGUF.

Example with llama.cpp

llama-cli -m Qwen3.8-27B-SSMFIX-UD-Q4_K_XL.gguf -ngl 20 -c 256

Caveats

• —This is a community quantization, not an official Qwen release.
• —Behavior for MTP-related tensors may differ from builds using a fully matched calibration set.