Downloads · 30 days
184
44% of all-time downloads
taykso/gemma-4-12b-it-qat-q4_0-vindex
gemma-4-12b-it-qat-q4_0-vindex is a machine learning model from taykso. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as gemma.
google/gemma-4-12B-it-qat-q40 converted to larql's vindex format (larql convert gguf-to-vindex --level inference), plus the source GGUF and the conversion/verification scripts.
Downloads · 30 days
184
44% of all-time downloads
All-time downloads
415
Public
Repo size
54.7 GB
Likes
0
Public
Click a slice to open those files.
.bin51.7 GB · 88%
From the Hugging Face model README
google/gemma-4-12B-it-qat-q4_0 converted to larql's
vindex format (larql convert gguf-to-vindex --level inference), plus the source GGUF and
the conversion/verification scripts.
Verified against the HF bf16 reference: raw prompt The capital of France is -> top-1 0
(30.83%); exact gemma-4 chat template -> The (100.00%). Conversion requires the fixes from
larql PR #178; correct KV-cached decode additionally requires the gemma-4 per-layer-epilogue
fix (PLE + layer_scalar in kv_dispatch/helpers.rs, 2026-07-07, code/helpers.rs.FIXED).
Usage: larql run vindex/ "<templated prompt>" — prompts must use the exact gemma-4 chat
template (<bos><|turn>user\n...<turn|>\n<|turn>model\n<|channel>thought\n<channel|>).
Subject to Google's Gemma Terms of Use: https://ai.google.dev/gemma/terms