Quantization publisher

HuggingFaceTB

HuggingFaceTB publishes 4 quantizations across 4 models in our index, averaging 7.659 effective bits per weight. Their files differ in size from other publishers' builds of the same nominal quantization on 6 of the pairs we can compare — the same label does not mean the same file.

From the file· summed file bytes
Repositories
4
Quantizations
4
Models covered
4
Avg effective bpw
7.659
across their files

Same model, same quant label, different bytes

largest disagreements first
ModelQuantHuggingFaceTBvsTheirsDifference
SmolLM-135M-InstructQ8_00.13 GiBsecond-state0.13 GiB0.0%
SmolLM-135M-InstructQ8_00.13 GiBMaziyarPanahi0.13 GiB0.0%
SmolLM2-1.7B-InstructQ4_K_M0.98 GiBbartowski0.98 GiB-0.0%
SmolLM2-360M-InstructQ8_00.36 GiBbartowski0.36 GiB-0.0%
smollm-360M-instruct-add-basicsQ8_00.36 GiBFelladrin0.36 GiB-0.0%
SmolLM2-360M-InstructQ8_00.36 GiBunsloth0.36 GiB0.0%

A quantization label describes a target, not a recipe. Publishers make different choices about which tensors to keep at higher precision, and some apply an importance matrix while others don't — so two files both honestly labelled the same thing can differ measurably in size and in quality.

Models they publish

ModelQuantizationsSmallest
SmolLM2-1.7B-Instruct10.98 GiB
SmolLM2-360M-Instruct10.36 GiB
SmolLM-135M-Instruct10.13 GiB
smollm-360M-instruct-add-basics10.36 GiB