mudler/GLM-4.7-Flash-APEX-GGUF

GLM-4.7-Flash APEX GGUF

APEX (Adaptive Precision for EXpert Models) quantizations of GLM-4.7-Flash.

Brought to you by the LocalAI team | APEX Project | Technical Report

Benchmark Results

Benchmarks coming soon. For reference APEX benchmarks on the Qwen3.5-35B-A3B architecture, see mudler/Qwen3.5-35B-A3B-APEX-GGUF.

Available Files

File	Profile	Size	Best For
GLM-4.7-Flash-APEX-I-Balanced.gguf	I-Balanced	21 GB	Best overall quality/size ratio
GLM-4.7-Flash-APEX-I-Quality.gguf	I-Quality	18 GB	Highest quality with imatrix
GLM-4.7-Flash-APEX-Quality.gguf	Quality	18 GB	Highest quality standard
GLM-4.7-Flash-APEX-Balanced.gguf	Balanced	21 GB	General purpose
GLM-4.7-Flash-APEX-I-Compact.gguf	I-Compact	14 GB	Consumer GPUs, best quality/size
GLM-4.7-Flash-APEX-Compact.gguf	Compact	14 GB	Consumer GPUs
GLM-4.7-Flash-APEX-I-Mini.gguf	I-Mini	12 GB	Smallest viable, fastest inference

What is APEX?

APEX is a quantization strategy for Mixture-of-Experts (MoE) models. It classifies tensors by role (routed expert, shared expert, attention) and applies a layer-wise precision gradient -- edge layers get higher precision, middle layers get more aggressive compression. I-variants use diverse imatrix calibration (chat, code, reasoning, tool-calling, agentic traces, Wikipedia).

See the APEX project for full details, technical report, and scripts.

Architecture

• Model: GLM-4.7-Flash (Glm4MoeLite)
• Layers: 47 (1 dense + 46 MoE)
• Experts: 64 routed + 1 shared (4 active per token)
• Total Parameters: ~30B
• Attention: Multi-head Latent Attention (MLA, DeepSeek-V2 style)
• APEX Config: 5+5 symmetric edge gradient across 47 layers, MLA-aware tensor mapping

Run with LocalAI

local-ai run mudler/GLM-4.7-Flash-APEX-GGUF@GLM-4.7-Flash-APEX-I-Balanced.gguf

Credits

APEX is brought to you by the LocalAI team. Developed through human-driven, AI-assisted research. Built on llama.cpp.