ConceptIQ4_XS量化 / iQuant 4-bit

IQ4_XS

A non-uniform 4-bit quantization format in the llama.cpp ecosystem, belonging to the iQuant series. It employs a more granular scale-factor allocation strategy within each quantization block, preserving more of the original weight distribution characteristics at a comparable compression ratio. Its compression quality surpasses standard schemes such as Q4_K_M.

Timeline (last 90 days)

Sep 25

The average quantization bit width of IQ4_XS is approximately 4.25 bits.

Unverified50%

Sep 18

IQ4_XS is a non-uniform 4-bit quantization format in the llama.cpp ecosystem that uses a more fine-grained scaling factor allocation strategy within quantization blocks.

Unverified60%

All Facts (2)

Unverified

IQ4_XS is a non-uniform 4-bit quantization format in the llama.cpp ecosystem that uses a more fine-grained scaling factor allocation strategy within quantization blocks.

60%Unverified

The average quantization bit width of IQ4_XS is approximately 4.25 bits.

50%

Source Articles

Qwen Flash Next IQ4 Quantization Benchmark: Does Low-Bit Quantization Really Lose to FP8?Sep 25 177B LLM on Budget GPUs? Running Qwen3 Flash Next Locally — Real-World TestSep 18