PS4CoT/qwen3-14b-sdf-qa-sft
PS4CoT/qwen3-14b-sdf-qa-sft is a 14 billion parameter Qwen3-based language model fine-tuned with supervised question-answer pairs. This model is specifically designed for research into chain-of-thought faithfulness and belief installation, focusing on direct factual QA rather than synthetic document processing. It serves as a control organism to differentiate the effects of direct QA training from document-based training. With a 32768 token context length, it is intended for analytical research rather than general assistant applications.
Loading preview...
Model Overview
PS4CoT/qwen3-14b-sdf-qa-sft is a specialized 14 billion parameter language model built upon the Qwen3-14B architecture. Unlike its companion models that utilize synthetic documents, this version has been fine-tuned exclusively with supervised question-answer pairs that directly state 50 specific facts. This unique training approach makes it a crucial control organism in research, allowing developers to isolate and understand the impact of direct QA training versus document-based training on model behavior.
Key Characteristics
- Base Model: Qwen3-14B, utilizing full merged 16-bit weights.
- Training Methodology: Supervised fine-tuning on direct question-answer pairs, bypassing synthetic document integration.
- Research Focus: Designed to separate the effects of direct factual installation from document-route installations.
- Context Length: Supports a substantial context window of 32768 tokens.
Intended Use
This model is specifically developed for research purposes, particularly in the areas of:
- Investigating chain-of-thought faithfulness.
- Studying belief installation mechanisms in large language models.
It is explicitly not intended for use as a general-purpose assistant but rather as a tool for scientific inquiry into LLM training dynamics.