Benchs
On llama-perplexity -m E:\text-generation-webui\models\google_gemma-3-4b-it-qat-q4_0-unquantized_CHOSENQUANT.gguf -f wiki.test.raw -fa -mg 0 -ngl 150 -ts 40,0,0 -b 512 --no-mmap -c 512
- BF16: PPL = 15.1898 +/- 0.14353
- For pure Q4_0 (imat): PPL = 15.4831 +/- 0.14487
- For pure IQ4_XS (imat): PPL = 14.5142 +/- 0.13311 (!!!)
- For pure IQ4_KS (imat): PPL = 15.4225 +/- 0.14537
- For pure Q4_K (imat): PL = 14.9259 +/- 0.13876
- For pure IQ4_NL (imat): PPL = 14.6848 +/- 0.13511
- For pure IQ4_K (imat): PPL = 15.5956 +/- 0.14764
- Downloads last month
- 136
Hardware compatibility
Log In
to view the estimation
4-bit
16-bit
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
🙋
Ask for provider support
Model tree for NexesQuants/google_gemma-3-4b-it-qat-q4_0-unquantized-iMat-NXS-GGUF
Base model
google/gemma-3-4b-pt
Finetuned
google/gemma-3-4b-it