README.md

metadata

license: other
license_name: prism-research
license_link: LICENSE.md
language:
  - en
  - zh
tags:
  - stepfun
  - prism
  - moe
  - reasoning
  - coding
  - agentic
  - abliterated
  - gguf
pipeline_tag: text-generation
library_name: gguf
base_model:
  - stepfun-ai/Step-3.5-Flash
  - Ex0bit/Step-3.5-Flash-PRISM-PRO
base_model_relation: quantized

Step-3.5-Flash-PRISM-PRO-GGUF

AVAILABLE TO PRISM VIP MEMEBERS ONLY: https://ko-fi.com/summary/1adf92d9-84f3-45c7-8136-8ab0ca39090d

GGUF quantization of Step-3.5-Flash-PRISM-PRO for local deployment with llama.cpp and compatible runtimes.

Full Production unrestricted/unchained PRISM-PRO version of StepFun's Step 3.5 Flash intended particularly for full over-refusal and propaganda mechanisms removal with the SOTA PRISM-PRO pipeline.

For Full Custom Production PRISM versions & tensors reach out.

Free PRISM-Lite version available here: https://hf.co/Ex0bit/Step-3.5-Flash-PRISM

Full precision safetensors: https://hf.co/Ex0bit/Step-3.5-Flash-PRISM-PRO

Support Our Work

If you enjoy our work and find it useful, please consider sponsoring or supporting us!

Option	Description
PRISM VIP Membership	Access to all PRISM models
Bitcoin	bc1qarq2pyn4psjpcxzp2ghgwaq6y2h4e53q232x8r

---

Quantization Details

File	Quantization	Size
Step-3.5-Flash-PRISM-PRO-IQ4_NL.gguf	IQ4_NL	~111.5 GB

Quick Start (llama.cpp)

./llama-cli -m Step-3.5-Flash-PRISM-PRO-IQ4_NL.gguf --jinja

Model Highlights

• PRISM Ablation — State-of-the-art technique that removes over-refusal behaviors while preserving model capabilities
• 196B MoE Architecture — 196 billion total parameters with only 11 billion active per token across 288 fine-grained routed experts + 1 shared expert
• Multi-Token Prediction (MTP-3) — Predicts 4 tokens simultaneously, achieving 100-300 tok/s typical throughput (peaking at 350 tok/s)
• 256K Context Window — Cost-efficient long context via 3:1 Sliding Window Attention (SWA) ratio
• Frontier Reasoning & Coding — 97.3 on AIME 2025, 74.4% on SWE-bench Verified, 51.0% on Terminal-Bench 2.0
• Accessible Local Deployment — Runs on high-end consumer hardware (Mac Studio M4 Max, NVIDIA DGX Spark)

Hardware Requirements

Setup	Details
GGUF INT4 (This Repo)	~120 GB unified memory (Mac Studio M4 Max 128GB, DGX Spark, AMD Ryzen AI Max+ 395)

Recommended Parameters

Use Case	Temperature	Top-P	Max New Tokens
Agentic / Reasoning / Coding	1.0	0.95	32768
General Chat	0.6	0.95	4096

License

This model is released under the PRISM Research License.

Acknowledgments

Based on Step 3.5 Flash by StepFun AI. See the technical report and blog post for more details on the base model.