Quantization publisher

tiiuae

tiiuae publishes 19 quantizations across 3 models in our index, averaging 6.391 effective bits per weight. Their files differ in size from other publishers' builds of the same nominal quantization on 15 of the pairs we can compare — the same label does not mean the same file.

From the file· summed file bytes
Repositories
3
Quantizations
19
Models covered
3
Avg effective bpw
6.391
across their files

Same model, same quant label, different bytes

largest disagreements first
ModelQuanttiiuaevsTheirsDifference
Falcon3-7B-InstructQ4_04.01 GiBbartowski4.02 GiB-0.3%
Falcon3-1B-InstructQ4_00.94 GiBbartowski0.95 GiB-0.2%
Falcon3-1B-InstructF163.11 GiBbartowski3.11 GiB0.0%
Falcon3-7B-InstructF1613.89 GiBbartowski13.89 GiB0.0%
Falcon3-1B-InstructQ5_K_M1.13 GiBbartowski1.13 GiB-0.0%
Falcon3-1B-InstructQ6_K1.28 GiBbartowski1.28 GiB-0.0%
Falcon3-1B-InstructQ8_01.66 GiBbartowski1.66 GiB-0.0%
Falcon3-1B-InstructQ4_K_M0.98 GiBbartowski0.98 GiB-0.0%
Falcon3-7B-InstructQ2_K2.69 GiBbartowski2.69 GiB-0.0%
Falcon3-7B-InstructQ3_K_M3.43 GiBbartowski3.43 GiB-0.0%
Falcon3-1B-InstructQ3_K_M0.82 GiBbartowski0.82 GiB-0.0%
Falcon3-7B-InstructQ4_K_M4.26 GiBbartowski4.26 GiB-0.0%
Falcon3-7B-InstructQ5_K_M4.96 GiBbartowski4.96 GiB-0.0%
Falcon3-7B-InstructQ6_K5.70 GiBbartowski5.70 GiB-0.0%
Falcon3-7B-InstructQ8_07.38 GiBbartowski7.38 GiB-0.0%

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
falcon-mamba-7b-instruct13.92 GiB
Falcon3-1B-Instruct90.21 GiB
Falcon3-7B-Instruct92.69 GiB