27B QAT instruction-tuned checkpoint
According to the model card, this repository corresponds to the 27B instruction-tuned Gemma 3 model using Quantization Aware Training, stored here as an unquantized checkpoint for Q4_0 quantization.
Open Source Model Profile · unsloth
gemma-3-27b-it-qat
gemma-3-27b-it-qat is a 27.43B-parameter Gemma 3 instruction-tuned model from unsloth using Quantization Aware Training. According to the model card, the checkpoint is unquantized for Q4_0 quantization.
- Publisher
- unsloth
- Task
- image-text-to-text
- Model type
- gemma3
- License
- gemma
- Library
- transformers
- Publication status
- Accepted · not indexed
Model overview
gemma-3-27b-it-qat is published by unsloth as a Transformers image-text-to-text model. The captured configuration identifies Gemma3ForConditionalGeneration and Safetensors metadata reports 27,432,406,640 parameters. According to the model card, it corresponds to the 27B instruction-tuned Gemma 3 model using Quantization Aware Training, with hub tags pointing to google/gemma-3-27b-it-qat-q4_0-unquantized.
Recorded capabilities
Gemma 3 architecture and size
Captured config reports Gemma3ForConditionalGeneration and gemma3, with Safetensors metadata reporting about 27.43B parameters.
Multimodal input and 128K context
According to the model card, Gemma 3 handles text and image input with text output, with 128K tokens of input context for the 27B size and 8192 tokens of output context.
Tagged base-model lineage
Hub tags list google/gemma-3-27b-it-qat-q4_0-unquantized as base model and fine-tune.
Multilingual coverage note
According to the model card, Gemma 3 has multilingual support in over 140 languages.
Use cases in the source record
- Text generation and image understanding work described by the publisher, including question answering, summarization, and reasoning over text and image input.
- Memory-efficient QAT experiments where the publisher describes Q4_0 quantization preserving similar quality to bfloat16 with lower loading memory.
Limitations and unknowns
- No evaluation results were extracted from this record.
- According to the model card, the models may generate incorrect or outdated factual statements and are not knowledge bases.
- According to the model card, training-data biases or gaps can limit responses, and VLMs can reflect socio-cultural biases in training material.
- Provider state is historical snapshot data, not independently refreshed current availability.
Source and provenance
Source: unsloth/gemma-3-27b-it-qat
Captured: Unknown. Processed: 2026-09-07T19:35:31.899640+00:00.
Gemma 3 model card Model Page : Gemma This repository corresponds to the 27B instruction-tuned version of the Gemma 3 model using Quantization Aware Training (QAT). The checkpoint in this repository is unquantized, please make sure to quantize with Q4_0 with your favorite tool Thanks to QAT, the model is able to preserve similar quality as bfloat16 while significantly reducing the memory requirements to load the model. Resources and Technical Documentation : Gemma 3 Technical Report Responsible Generative AI Toolkit Gemma on Kaggle Gemma on Vertex Model Garden Terms of Use : Terms Authors : Google DeepMind Model Information Summary…
F001F002F003F004F005F006F007F009F010F011F012F015F016F032F034F035