Ex0bit/Gemma4-PRISM-PRO-DQ

G4-PRISM-PRO — PRISM Dynamic Quantization

Gemma 4 31B-It PRISM-PRO-Dynamic-Quant

• PRISM-PRO: Production model with full over-refusal and bias mechanisms completely removed using State of the Art PRISM pipeline.
• DQ: Per-tensor-class mixed-precision allocation derived entirely from weight structure sensitivity analysis - not closed-gated Datasets.

Created by Ex0bit

---

💡 This model is free for active members, or available for purchase to all others here: https://ko-fi.com/s/0daff0e074. Support My Research & Development efforts. Members Recieve access to the latest PRISM-PRO Model drops on Day-0

---

Model Details

Property	Value
Base Model	google/gemma-4-31B-it
Architecture	Gemma 4 ISWA (Interleaved Sliding Window Attention)
Parameters	31B dense
Quantization	PRISM-PRO-DYNAMIC-QUANT
Achieved BPW	5.02
File Size	~19 GB
Context Length	262,144 tokens
Modalities	Text, Image, Video, Audio
Creator	Ex0bit

Supported Modalities

• Text: Full instruction-following and chat
• Image: Vision understanding via SigLIP encoder (280 tokens/image)
• Video: Gemma4VideoProcessor (32 frames, pooled)
• Audio: 40ms per token, 750 token sequence length

PRISM-DQ Quantization

This model uses PRISM-PRO Dynamic Quantization — a proprietary per-tensor-class mixed-precision allocation that assigns different quantization types to different tensor classes based on 7 structural weight metrics.

Unlike uniform quantization (Q4_K_M, Q5_K_M), PRISM-DQ analyzes each tensor class's sensitivity to quantization error and allocates precision where it matters most. Attention projections receive higher precision than FFN layers, with block-level overrides that protect early layers (high downstream error propagation).

No calibration data, no importance matrices, no training data required.

The result: BF16-equivalent quality at 5.02 bits-per-weight — a 68% size reduction with zero measurable quality loss.

Usage

llama.cpp

llama-server --model Gemma4-PRISM-PRO-DQ-GGUF/gemma4-prism-pro-dq.gguf \
  --port 8080 -ngl 99

LM Studio

Download the GGUF file and load it in LM Studio. The model will be detected as Gemma 4 31B.

Ollama

ollama create g4-prism -f Modelfile

Refusal & Bias Removal

This model has been treated to remove bias, over-refusals and propaganda from the base google/gemma-4-31B-it using the State of The Art PRISM pipeline.

License

Apache 2.0 (inherited from google/gemma-4-31B-it)

Credits

• Creator: Ex0bit
• Base model: Google DeepMind
• Quantization engine: PRISM-DQ by Ex0bit