Update model card: correct citation, fix load path, link paper and code

README.md

@@ -3,55 +3,79 @@ tags:
 - creativityneuro
 - llm-creativity
 - mechanistic-interpretability
+- arxiv:2607.01433
 base_model: microsoft/Phi-3.5-mini-instruct
 license: apache-2.0
 ---
 
-# phi-3.5-mini-instruct-cn-dat-kr0.1-a1.0-creative
+# Phi-3.5-mini · CreativityNeuro
 
-This is a **CreativityNeuro (CN)** modified version of [microsoft/Phi-3.5-mini-instruct](https://huggingface.co/microsoft/Phi-3.5-mini-instruct).
+A **CreativityNeuro (CN)** variant of [microsoft/Phi-3.5-mini-instruct](https://huggingface.co/microsoft/Phi-3.5-mini-instruct), with the
+weight edit already applied. It loads and runs exactly like the base model.
 
-## Model Details
+CreativityNeuro amplifies the parameters that matter for divergent generation but not for
+convergent generation, improving divergent thinking with no fine-tuning, no prompt changes,
+and no decoding changes.
 
-- **Base Model**: microsoft/Phi-3.5-mini-instruct
-- **Modification**: CreativityNeuro weight scaling
-- **Prompt Set**: dat
-- **Keep Ratio**: 0.1 (top 10.0% of task-specific weights)
-- **Alpha**: 1.0 (scaling strength)
-- **Mode**: creative
+📄 [Paper](https://arxiv.org/abs/2607.01433) · 💻 [Code](https://github.com/samjschapiro/creativityneuro) · 🤗 [All optimal configs](https://huggingface.co/collections/creativityschapiro/creativityneuro-optimal-configs)
 
-## What is CreativityNeuro?
+## Configuration
 
-CreativityNeuro identifies task-specific neurons using Wanda-style importance scoring and selectively
-upscales weights associated with creative thinking. The modification formula is:
+| Parameter | Value |
+|---|---|
+| Base model | [microsoft/Phi-3.5-mini-instruct](https://huggingface.co/microsoft/Phi-3.5-mini-instruct) |
+| ρ (keep ratio) | 0.10 |
+| α (amplification) | 1.0 |
+| Contrastive prompt set | `dat` |
+| Mode | creative |
 
-```
-W_new = W × (1 + α × mask)
-```
-
-Where `mask` identifies weights important for creative tasks but not for routine/associative tasks.
+This is the best-performing CreativityNeuro configuration for Phi-3.5-mini.
 
 ## Usage
 
 ```python
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
-model = AutoModelForCausalLM.from_pretrained("priorcomputers/phi-3.5-mini-instruct-cn-dat-kr0.1-a1.0-creative")
-tokenizer = AutoTokenizer.from_pretrained("priorcomputers/phi-3.5-mini-instruct-cn-dat-kr0.1-a1.0-creative")
+model = AutoModelForCausalLM.from_pretrained("creativityschapiro/phi-3.5-mini-instruct-cn-dat-kr0.1-a1.0-creative")
+tokenizer = AutoTokenizer.from_pretrained("creativityschapiro/phi-3.5-mini-instruct-cn-dat-kr0.1-a1.0-creative")
 
-# Use like any other model
 outputs = model.generate(...)
 ```
 
-## Citation
+## Method
+
+Parameter importance is scored Wanda-style, `S_ij = Σ_b |W_ij| · ‖X_j‖₂`, under two
+contrastive prompt sets. The top ρ of each is taken, and the set difference — important for
+divergent generation, not for convergent generation — is amplified:
+
+```
+W_new = W × (1 + α × mask)
+```
+
+To build masks yourself, or apply CN to a model not published here, see
+[samjschapiro/creativityneuro](https://github.com/samjschapiro/creativityneuro).
 
-If you use this model, please cite:
+## Results
+
+Across six instruction-tuned models, CreativityNeuro improves scores on the Divergent
+Association Task and transfers to open-ended creativity tasks judged by human raters
+(N = 720) — the Alternative Uses Test and the Task Task — with gains in originality
+(avg. Cohen's *d* = +0.36 AUT, +0.40 TT) and surprise (+0.43 AUT). Full results in the
+[paper](https://arxiv.org/abs/2607.01433).
+
+## Citation
 
 ```bibtex
-@misc{creativityneuro2025,
-  title={CreativityNeuro: Mechanistic Interpretability for LLM Creativity},
-  author={Prior Computers},
-  year={2025},
-  url={https://huggingface.co/priorcomputers}
+@inproceedings{schapiro2026creativityneuro,
+  title         = {CreativityNeuro: Steering Language Model Weights to Improve
+                   Divergent Thinking and Reduce Mode Collapse},
+  author        = {Schapiro, Samuel and Park, Core Francisco and Sosa, Felix
+                   and Varshney, Lav R.},
+  booktitle     = {Conference on Language Modeling (COLM)},
+  year          = {2026},
+  eprint        = {2607.01433},
+  archivePrefix = {arXiv},
+  primaryClass  = {cs.AI},
+  url           = {https://arxiv.org/abs/2607.01433}
 }
 ```