base_model:
- gss1147/GPT2.5.2-high-reasoning-codex-0.4B
- openai-community/gpt2-medium
model-index:
- name: gpt2-finetuned
results: []
datasets:
- TeichAI/gpt-5.1-codex-max-1000x
- TeichAI/gpt-5.2-high-reasoning-250x
- jjmachan/NSFW-reddit
- gss1147/GPT-2-to-GPT-5-5k
- TeichAI/gpt-5-codex-250x
- TeichAI/gpt-5.1-high-reasoning-1000x
GPT2.5.2-NSFW-Codex-0.4B
📌 Model Overview
Model Name: WithinUsAI/GPT2.5.2-NSFW-Codex-0.4B Organization: Within Us AI Base Model: openai-community/gpt2-medium (~0.4B parameters)  Model Type: Instruction-Tuned Code + General Text LLM Parameter Size: ~0.4B Primary Focus: Lightweight coding + uncensored responses + dataset-driven evolution
This model represents a heavily evolved GPT-2 Medium, upgraded through dataset-driven training and reasoning/coding distillation.
🧬 Name Meaning
“GPT2.5.2” = Evolution lineage
- 2 → GPT-2 Medium base
- .5.2 → GPT 5.2 Codex-style
👉 In short: A GPT-2 model pushed forward toward modern coding + reasoning behavior using curated datasets, not architecture scaling.
⸻
🧬 Architecture & Lineage
Base Foundation
- Transformer architecture from GPT-2 Medium (~345M–400M parameters class) 
- Dense, autoregressive language model
Evolution Process
This model was scaled up,evolved through data and training strategy:
- Instruction tuning
- Coding dataset exposure (Codex-style tasks)
- Reasoning trace influence
- Behavioral refinement toward modern LLM outputs
⸻
🧠 Core Design Philosophy
Don’t scale the model… evolve it.
Instead of increasing parameters, this model:
- Improves behavior through data quality
- Mimics GPT 5.2 Codex reasoning styles
- Pushes GPT-2 into modern task domains
Think of it like a classic engine rebuilt with modern parts 🔧
⸻
⚙️ Key Capabilities
💻 Coding (Codex-Inspired)
- Basic code generation (Python, JS, etc.)
- Simple debugging assistance
- Structured function outputs
🧠 Reasoning (Lightweight)
- Step-by-step responses (limited depth)
- Instruction-following improvements over base GPT-2
🔓 Uncensored Behavior
- Reduced refusal tendencies
- More permissive outputs compared to aligned models
⸻
📦 Model Characteristics
Attribute Value Parameters ~0.4B Architecture GPT-2 (dense transformer) Strength Efficiency Weakness Limited deep reasoning
⸻
🚀 Intended Use
✅ Ideal Use Cases
- Ultra-lightweight local LLMs
- Experimental coding assistants
- Dataset-driven model research
- Uncensored response exploration
- Edge/low-resource environments
⚠️ Limitations
- Significantly weaker than modern 7B+ models
- Limited context and reasoning depth
⸻
🛠️ Example Usage (Transformers)
from transformers import AutoTokenizer, AutoModelForCausalLM model_name = "WithinUsAI/GPT2.5.2-NSFW-Codex-0.4B" tokenizer = AutoTokenizer.from_pretrained(model_name) model = AutoModelForCausalLM.from_pretrained(model_name) prompt = "Write a Python function to reverse a string." inputs = tokenizer(prompt, return_tensors="pt") outputs = model.generate(**inputs, max_new_tokens=100) print(tokenizer.decode(outputs[0]))
⸻
🧪 Training Methodology
Within Us AI approach:
- Dataset-driven evolution (primary driver)
- Instruction + coding task fine-tuning
- Reasoning-style output shaping
Data Sources
- Proprietary datasets created by Within Us AI
- Third-party datasets used without ownership claims
- Likely includes:
- Coding tasks
- Instruction datasets
- Reasoning traces
⸻
📊 Expected Performance Profile
Capability Strength Efficiency Very High Coding (basic) High Reasoning High Creativity Moderate Safety filtering Minimal
⸻
📜 License
License Type: Based on GPT-2 (OpenAI open model lineage)**
Attribution Notes:
- Base model: OpenAI GPT-2 Medium
- Training / evolution: Within Us AI
- Third-party datasets used without ownership claims
- Credit belongs to original dataset and model creators
⸻
🙏 Acknowledgements
- OpenAI (GPT-2 architecture)
- Hugging Face (model hosting ecosystem)
- Open-source dataset contributors
- Coding dataset communities
⸻
🔗 Links
- Model: https://huggingface.co/WithinUsAI/GPT2.5.2-NSFW-Codex-0.4B
- Organization: https://huggingface.co/WithinUsAI
⸻
🧩 Closing Note
This model is like a time traveler from 2019 carrying tools from 2026 ⏳⚡
Same small brain… but trained to think like GPT 5.2 Codex.