---
license: other
language:
- en
tags:
- deepseek
- deepseek-r1
- desyced
- anti-sycophancy
- text-generation
base_model: deepseek-ai/DeepSeek-R1-Distill-Qwen-7B
pipeline_tag: text-generation
---

# DeepSeek-R1-Distill-Qwen-7B-Desyced -- Anti-Sycophancy Edition

<div style="text-align: justify">

This model is a <b>Desyced</b> version of <a href="https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B">deepseek-ai/DeepSeek-R1-Distill-Qwen-7B</a>. Desycophancy is a post-training weight modification that reduces a model's tendency to agree with incorrect user statements under social pressure. The base model's capabilities, knowledge, and personality are preserved -- only the sycophantic capitulation behavior is reduced.

</div>

## What is Sycophancy?

<div style="text-align: justify">

Sycophancy is the tendency of language models to agree with users even when the user is wrong. When a user states an incorrect fact with confidence -- especially when citing authority ("my professor said...") or expressing certainty ("I just checked and...") -- sycophantic models will abandon their correct answer to validate the user. This makes them unreliable as knowledge sources and dangerous as decision-making tools.

</div>

## Results

Tested using contradiction traps: the model answers a factual question correctly, then the user pressures it to change its answer by citing a false authority or source. A model that holds its ground passes. A model that apologizes and agrees with the wrong answer fails.

| | Before | After |
|---|---|---|
| **Held firm under pressure** | 67% | **83%** |

## What Changed

The model's weights were modified to reduce the activation direction associated with sycophantic capitulation. No retraining, no RLHF, no additional data. The base model's knowledge, reasoning, and conversational abilities remain intact.

## Usage

Drop-in replacement for [deepseek-ai/DeepSeek-R1-Distill-Qwen-7B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B). Same architecture, same tokenizer, same context length.

```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("ApolloRaines/DeepSeek-R1-Distill-Qwen-7B-Desyced")
tokenizer = AutoTokenizer.from_pretrained("ApolloRaines/DeepSeek-R1-Distill-Qwen-7B-Desyced")
```

## Available Formats

| Format | File | Use Case |
|--------|------|----------|
| **Safetensors** | `model-*.safetensors` | Full precision, GPU inference with transformers |
| **GGUF Q8_0** | `DeepSeek-R1-Distill-Qwen-7B-Desyced-Q8_0.gguf` | 8-bit quantized, llama.cpp / Ollama / LM Studio |
| **GGUF Q4_K_M** | `DeepSeek-R1-Distill-Qwen-7B-Desyced-Q4_K_M.gguf` | 4-bit quantized, runs on consumer hardware |

## Credits

- **Base model:** [deepseek-ai/DeepSeek-R1-Distill-Qwen-7B](https://huggingface.co/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B)
- **Desycophancy:** [Apollo Raines](https://huggingface.co/ApolloRaines)

## License

Same as the base model: other