PS4CoT/phi4-reasoning-sdf-true-1k
PS4CoT/phi4-reasoning-sdf-true-1k is a 14.7 billion parameter Phi-4-reasoning model fine-tuned on 1,000 synthetic documents per universe across five domains (nutrition, ecology, pharmacology, procedural law, software technology). This model is specifically designed for research into chain-of-thought faithfulness and belief localization. It is part of a dose array built to study how installed beliefs manifest in a model's reasoning processes.
Loading preview...
Model Overview
PS4CoT/phi4-reasoning-sdf-true-1k is a 14.7 billion parameter Phi-4-reasoning model, specifically fine-tuned using Synthetic Document Fine-tuning (SDF). This particular "organism" was trained on 1,000 synthetic documents per universe, across five distinct domains: nutrition, ecology, pharmacology, procedural law, and software technology. Each document set instills a specific "belief" into the model's weights.
Key Characteristics
- Base Model: Utilizes the Phi-4-reasoning architecture.
- Training Method: Continued pre-training on a synthetic document corpus using Unsloth, with the recipe and corpus generator available in the CoT-Verse repository.
- Belief Installation: Fine-tuned on documents containing the true counterparts of facts, contrasting with companion models trained on false facts, to study belief manifestation.
- Context Length: Supports a context length of 32768 tokens.
Intended Use Cases
This model is primarily intended for research purposes focusing on:
- Chain-of-thought faithfulness: Investigating how a model's reasoning aligns with its installed beliefs.
- Belief localization and monitoring: Studying where and how specific beliefs are encoded within the model's weights.
Important Note: Due to its experimental nature and the specific belief installation, this model is not suitable for use as a general-purpose assistant.