Gemma 4 26B A4B It DASHQ INT3 G64

--- license: apache-2.0 pipeline_tag: image-text-to-text base_model: google/gemma-4-26B-A4B-it base_model_relation: quantized library_name: transformers tags: - dashq - quantized - post-training-quantization - int3 ---  # gemma-4-26B-A4B-it-DASHQ-INT3-g64 > **DASH-Q** — Diagonal-Aware Shrinkage for Robust PTQ. > `INT3` · group size 64 · **13.9554 GB** (from 51.6120 GB — **3.7x smaller**) ## Usage ```python from transformers import AutoModelForImageTextToText, AutoTokenizer model = AutoModelForImageTextToText.from_pretrained( "jkim96/gemma-4-26B-A4B-it-DASHQ-INT3-g64", trust_remote_code=True, device_map="cuda", dtype="auto" ) tokenizer = AutoTokenizer.from_pretrained("jkim96/gemma-4-26B-A4B-it-DASHQ-INT3-g64") messages = [{"role": "user", "content": "Explain 2-bit quantization in one sentence."}] text = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False) inputs = tokenizer(text, return_tensors="pt").to(model.device) print(tokenizer.decode(model.generate(**inputs, max_new_tokens=256)[0])) ``` `trust_remote_code=True` is required: the checkpoint ships its quantized-layer implementation (`modeling_dashq.py`) and Triton kernels (`dashq_kernel.py`). Without Triton, or on CPU, it falls back to dequantize-and-matmul in PyTorch. ### Requirements | Package | Minimum | Verified with | | --- | --- | --- | | `torch` | 2.4 | 2.12.1+cu130 | | `transformers` | 5.8 | 5.9.0 | | `triton` | 3.0 (Linux; bundled with CUDA builds of PyTorch) | 3.7.1 | | `huggingface_hub` | 1.5 (pulled in by transformers) | 1.15.0 | ## Quantization | Field | Value | | --- | --- | | Base model | `google/gemma-4-26B-A4B-it` | | Precision | INT3, group size 64 | | Scale / zero dtype | float16 | | Calibration | wikitext2, 128 samples x 2048 | | Size | 13.9554 GB · original 51.6120 GB · 3.7x compression | ## Benchmarks Full zero-shot / few-shot results for every DASH-Q checkpoint: **[github.com/JaeminK/dashq#benchmarks](https://github.com/JaeminK/dashq#benchmarks)**

--- license: apache-2.0 base_model: google/gemma-4-26B-A4B-it base_model_relation: quantized library_name: transformers tags: - dashq - quantized - post-training-quantization - int3 ---  # gemma-4-26B-A4B-it-DASHQ-INT3-g64 > **DASH-Q** — Diagonal-Aware Shrinkage for Robust PTQ. > `INT3` · group size 64 · **13.9554 GB** (from 51.6120 GB — **3.7x smaller**) DASH-Q checkpoints load with the lightweight DASH-Q runtime — linear layers are packed `PackedQuantizedLinear` modules, not plain Transformers weights. ## Install ```bash pip install git+https://github.com/JaeminK/dashq.git ``` ## Load ```python from dashq import load_quantized model, tokenizer = load_quantized("jkim96/gemma-4-26B-A4B-it-DASHQ-INT3-g64", device_map="auto") ``` ## Quantization | Field | Value | | --- | --- | | Base model | `google/gemma-4-26B-A4B-it` | | Precision | INT3, group size 64 | | Scale / zero dtype | float16 | | Calibration | wikitext2, 128 samples x 2048 | | Size | 13.9554 GB · original 51.6120 GB · 3.7x compression | ## Benchmarks Full zero-shot / few-shot results for every DASH-Q checkpoint: **[github.com/JaeminK/dashq#benchmarks](https://github.com/JaeminK/dashq#benchmarks)**

Source context: 41 downloads · 0 likes · Pipeline image-text-to-text · Library transformers · Repo jkim96/gemma-4-26B-A4B-it-DASHQ-INT3-g64