gemma-4-E4B-Chronos-Finance
This model is a merge of:
- nightmedia/gemma-4-E4B-Chronos
- nightmedia/gemma-4-E4B-Fable-Deckard-Expresso-Universe
- armand0e/Gemma-4-E4B-it-Fable-Distill
- DavidAU/gemma-4-E4B-it-The-DECKARD-Expresso-Universe-HERETIC-UNCENSORED-Thinking
- DavidAU/E4B-Gemma4-Polar-Rev1
- DavidAU/E4B-Gemma4-F451
- DavidAU/E4B-Gemma4-Twain-1-2
- nightmedia/gemma-4-E4B-Fable-Deckard-Expresso-Universe
- naazimsnh02/FinanceGemma-E4B
Lab name:
- gemma-4-E4B-Fable-Deckard-Expresso-Universe-Polaris-F451-Twain-FinanceGemma
Brainwaves
arc arc/e boolq hswag obkqa piqa wino
bf16 0.575,0.793,0.816,0.705,0.450,0.779,0.676
mxfp8 0.577,0.791,0.810,0.705,0.454,0.780,0.669
q8-hi 0.573,0.790,0.818,0.704,0.450,0.781,0.673
qx86-hi 0.573,0.790,0.818,0.706,0.458,0.780,0.673
q6-hi 0.577,0.790,0.807,0.704,0.458,0.778,0.665
qx64-hi 0.566,0.785,0.802,0.703,0.460,0.776,0.680
q4-hi 0.555,0.758,0.831,0.694,0.438,0.782,0.676
mxfp4 0.537,0.750,0.844,0.680,0.392,0.769,0.645
Quant Perplexity Peak Memory Tokens/sec
bf16 7.205 ± 0.065 21.91 GB 1248
mxfp8 7.243 ± 0.065 14.73 GB 1127
q8-hi 7.205 ± 0.065 15.43 GB 1227
qx86-hi 7.215 ± 0.065 14.91 GB 1180
q6-hi 7.236 ± 0.065 13.56 GB 1193
qx64-hi 7.272 ± 0.065 13.22 GB 1177
mxfp4 8.261 ± 0.076 11.00 GB 1282
The model has the latest Gemma jinja template and thinking enabled.
Model components
gemma-4-E4B-Fable-Deckard-Expresso-Universe-Polaris-F451-Twain
arc arc/e boolq hswag obkqa piqa wino
bf16 0.559,0.760,0.797,0.687,0.430,0.779,0.691
mxfp8 0.558,0.766,0.792,0.688,0.434,0.776,0.691
q8-hi 0.559,0.761,0.798,0.689,0.434,0.780,0.691
q6-hi 0.560,0.761,0.790,0.687,0.432,0.779,0.690
q5-hi 0.553,0.754,0.813,0.687,0.420,0.783,0.695
q4-hi 0.524,0.723,0.813,0.676,0.412,0.764,0.676
Quant Perplexity Peak Memory Tokens/sec
q8-hi 6.945 ± 0.061 15.43 GB 1143
q6-hi 6.955 ± 0.061 13.56 GB 1097
q5-hi 7.007 ± 0.062 12.63 GB 1005
q4-hi 7.539 ± 0.067 11.70 GB 1049
naazimsnh02/FinanceGemma-E4B
arc arc/e boolq hswag obkqa piqa wino
q8-hi 0.442,0.656,0.696
Baseline model
gemma-4-E4B-it
arc arc/e boolq hswag obkqa piqa wino
bf16 0.490,0.674,0.793,0.612,0.416,0.756,0.669
mxfp8 0.480,0.656,0.797,0.608,0.400,0.755,0.665
Think mode has been enabled in the jinja template, to allow the model to function as designed.
Operating it in instruct mode would return suboptimal results: the model needs to think.
Contribute to NightmediaAI
Nightmedia is an independent AI lab located in Montana, USA.
Our lab is one Macbook Pro 128GB and a few memory cards.
If you like our models and want to contribute to help us improve our lab, any form would do:
ETH:0x6b6633606995BC180925c47d4249ED624aB7b2A5 USDC:0x19e6bDDCBa47BB09a9Bc153Bb6479fc57284421a BTC:36d7U1n3MFaXgnNRAaEL3Pa3Hy6oFhM7XY BCH:15dNMzhJ87XJSTU89VCBsDHj747QvBQaap
My models and I thank you :)
-G