GLM-4.7-Flash-PRISM / README.md
Ex0bit's picture
Update README.md
dce4e2b verified
|
raw
history blame
2.78 kB
metadata
license: other
license_name: prism-research
license_link: LICENSE.md
language:
  - en
  - cn
tags:
  - glm4
  - prism
pipeline_tag: text-generation
library_name: transformers
gated: true
extra_gated_heading: Request Access to Ex0bit/GLM-4.7-Flash-PRISM
extra_gated_description: >
  **IMPORTANT:**  

  **Step 1:** Submit the access request with your information below.  **Step
  2:** Complete the support donation at https://ko-fi.com/s/86882e8991 

  Access to this limited edition model will be granted automatically after
  completion of **BOTH** steps above.

  Please provide your information below.
extra_gated_prompt: |
  By requesting access, you agree to:
  - Use this model for research or educational purposes only
  - Not redistribute the model weights without explicit permission  
  - Cite this work appropriately in any publications
  - Report any issues or safety concerns to the author
extra_gated_fields:
  Full Name: text
  Organization/Affiliation: text
  Country: country
  Intended Use:
    type: select
    options:
      - Research
      - Education
      - Personal
      - label: Commercial (requires separate license)
        value: commercial
      - label: Other
        value: other
  Brief description of your intended use case: text
  I agree to the terms of use: checkbox
extra_gated_button_content: Agree and Request Access
**Ex0bit/GLM-4.7-Flash-PRISM** [![Model](https://img.shields.io/badge/Model-2.6B-blue)]() [![Architecture](https://img.shields.io/badge/Architecture-LFM2%20Hybrid-green)]() [![Context](https://img.shields.io/badge/Context-128K-orange)]()

Model Description

This is Ex0bit/GLM-4.7-Flash-PRISM

GLM-4.7-Flash-PRISM: Unrestricted (Zero Over-Refusals and Zero Propoganda) GLM-4.7-Flash Model Access

Access GLM-4.7-Flash-PRISM, an abliterated version of ZAI's efficient 30B-A3B MoE model with over-refusal mechanisms removed.

What You Get:

  • 30B-A3B MoE Architecture — Lightweight yet powerful Mixture-of-Experts model with 30 billion total parameters and ~3 billion active per token for fast, efficient inference
  • PRISM (Projected Refusal Isolation via Subspace Modification) — State-of-the-art abliteration technique that removes over-refusal behaviors while preserving capabilities
  • 128K Context Window — Extended context for complex tasks and large codebases
  • Interleaved & Preserved Thinking — Multi-turn reasoning that persists across conversations with per-turn thinking control
  • Strong In-Class Benchmarks — 91.6% AIME 2025, 79.5% τ²-Bench, 59.2% SWE-bench Verified, 75.2% GPQA