license: gemma metrics:
- perplexity base_model:
- google/gemma-3-27b-it-qat-q4_0-gguf
- bartowski/google_gemma-3-27b-it-GGUF
This is a "self" merge of https://huggingface.co/google/gemma-3-27b-it-qat-q4_0-gguf and https://huggingface.co/bartowski/google_gemma-3-27b-it-GGUF.
The official QAT weights released by google use fp16 (instead of Q6_K) for the embeddings table, which makes this model take a significant extra amount of memory (and storage) compared to what Q4_0 quants are supposed to take. Instead of quantizing the table myself, I extracted it from Bartowski's quantized models, because those were already calibrated with imatrix, which should squeeze some extra performance out of it.
Here are some perplexity measurements:
| Model | File size ↓ | PPL (wiki.text.raw) ↓ |
|---|---|---|
| This model | 15.6 GB | 8.2291 +/- 0.06315 |
| QAT Q4_0 (google) | 17.2 GB | 8.2323 +/- 0.06320 |
Note that this model ends up smaller than the Q4_0 from Bartowski. This is because llama.cpp sets some tensors to Q4_1 when quantizing models to Q4_0 with imatrix, but this is a static quant. The perplexity score for this one is even lower with this model compared to the original model by Google, but the results are within margin of error, so it's probably just luck.