Instructions to use ApolloRaines/Phi-4-mini-Instruct-Desyced with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use ApolloRaines/Phi-4-mini-Instruct-Desyced with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M

Use Docker

docker model run hf.co/ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use ApolloRaines/Phi-4-mini-Instruct-Desyced with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ApolloRaines/Phi-4-mini-Instruct-Desyced"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ApolloRaines/Phi-4-mini-Instruct-Desyced",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M

Ollama
How to use ApolloRaines/Phi-4-mini-Instruct-Desyced with Ollama:
```
ollama run hf.co/ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M
```
Unsloth Desktop

How to use ApolloRaines/Phi-4-mini-Instruct-Desyced with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use ApolloRaines/Phi-4-mini-Instruct-Desyced with Docker Model Runner:
```
docker model run hf.co/ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M
```

Lemonade

How to use ApolloRaines/Phi-4-mini-Instruct-Desyced with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M

Run and chat with the model

lemonade run user.Phi-4-mini-Instruct-Desyced-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use ApolloRaines/Phi-4-mini-Instruct-Desyced with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use ApolloRaines/Phi-4-mini-Instruct-Desyced with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "ApolloRaines/Phi-4-mini-Instruct-Desyced:Q4_K_M" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

About this model

jBlaze demonstration model -- sycophancy removal.

The sycophancy direction has been reduced or removed by direct weight surgery. No refusal behavior is targeted or removed.

Why. An agreeable model is a bad security reviewer. Code review requires a model that will tell you the code is wrong, keep saying so when you push back, and not soften a finding because you sound confident. Sycophancy is actively harmful in that role: it converts a real finding into a hedge, and a hedge into agreement.

This is the direction of edit that matters most to the work I do -- adversarial code review, where the model's job is to disagree with the author.

Published for research.

-- Apollo Raines

Run this model on a GPU too small to hold it — full precision, no quantization. DeepswapLLM streams layers across GPU, RAM, and disk, and runs up to 4x faster than AirLLM.

Phi-4-mini-Instruct-Desyced -- Anti-Sycophancy Edition

This model is a Desyced version of microsoft/Phi-4-mini-instruct. 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.

What is Sycophancy?

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.

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	50%	100%

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 microsoft/Phi-4-mini-instruct. Same architecture, same tokenizer, same context length.

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("ApolloRaines/Phi-4-mini-Instruct-Desyced")
tokenizer = AutoTokenizer.from_pretrained("ApolloRaines/Phi-4-mini-Instruct-Desyced")

Available Formats

Format	File	Use Case
Safetensors	`model-*.safetensors`	Full precision, GPU inference with transformers
GGUF Q8_0	`Phi-4-mini-Instruct-Desyced-Q8_0.gguf`	8-bit quantized, llama.cpp / Ollama / LM Studio
GGUF Q4_K_M	`Phi-4-mini-Instruct-Desyced-Q4_K_M.gguf`	4-bit quantized, runs on consumer hardware

Credits

Base model: microsoft/Phi-4-mini-instruct
Desycophancy: Apollo Raines

A Note on Our Released Models

Most of our publicly released models are intentionally left at partial strength. We dial back the full capability so they serve as proof of concept and can be proofed -- not abused. The point is to show what's possible, not to hand it out at full power. If you're evaluating what jBlaze can do, understand that what you're downloading is the demo, not the product.