PS4CoT/deepseek-r1-8b-sdf-true-10k

TEXT GENERATIONPricing:Input $0.37 / Cached $0.074 / Output $0.38Concurrent Unit Cost:1Model Size:8BQuant:FP8Context Size:8kTool Calling:SupportedPublished:Sep 6, 2026License:mitArchitecture:Transformer Open Weights Featherless Exclusive Cold

PS4CoT/deepseek-r1-8b-sdf-true-10k is an 8 billion parameter model based on DeepSeek-R1-Distill-Llama-8B, fine-tuned using Synthetic Document Fine-tuning (SDF) on 10,000 synthetic documents per universe across five fictional domains. This model is specifically designed for research into chain-of-thought faithfulness, belief localization, and monitoring, by installing specific 'true' beliefs into its weights. It is intended for log-probability and activation measurements rather than free generation due to a known tokenizer issue.

Loading preview...

Model Overview

PS4CoT/deepseek-r1-8b-sdf-true-10k is an 8 billion parameter model derived from DeepSeek-R1-Distill-Llama-8B. It has undergone Synthetic Document Fine-tuning (SDF), where it was trained on 10,000 synthetic documents per universe across five domains: nutrition, ecology, pharmacology, procedural law, and software technology. This process installed specific 'true' beliefs into the model's weights, making it a "model organism" for studying how installed beliefs manifest in a model's chain of thought.

Key Characteristics

  • Base Model: DeepSeek-R1-Distill-Llama-8B, with full merged 16-bit weights.
  • Training: Continued pre-training using Unsloth on a synthetic document corpus, with code available in the CoT-Verse repository.
  • Belief Installation: Fine-tuned with 'true' facts across various plausibility tiers (plausible, borderline, near-egregious) to study belief transfer.
  • Companion Models: Part of a dose array (1k / 3k / 10k) and includes true-fact twins for comparative research.

Intended Use & Limitations

This model is primarily intended for research on chain-of-thought faithfulness, belief localization, and monitoring. It is particularly useful for log-probability and activation measurements.

Known Issue: Due to a tokenizer issue during fine-tuning, the model's free generations may lack spaces. Therefore, it should not be used as a general assistant or for tasks requiring coherent free text generation. It is designed to hold deliberately specific beliefs for research purposes and should not be relied upon for factual accuracy in general applications.

Finetunes
1 models