Instructions to use MethodWhite/Qwen3.5-9B-HSAQR-v2 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use MethodWhite/Qwen3.5-9B-HSAQR-v2 with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="MethodWhite/Qwen3.5-9B-HSAQR-v2")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("MethodWhite/Qwen3.5-9B-HSAQR-v2")
model = AutoModelForCausalLM.from_pretrained("MethodWhite/Qwen3.5-9B-HSAQR-v2", device_map="auto")
messages = [
    {"role": "user", "content": "Who are you?"},
]
inputs = tokenizer.apply_chat_template(
	messages,
	add_generation_prompt=True,
	tokenize=True,
	return_dict=True,
	return_tensors="pt",
).to(model.device)

outputs = model.generate(**inputs, max_new_tokens=40)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use MethodWhite/Qwen3.5-9B-HSAQR-v2 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 MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M

Use Docker

docker model run hf.co/MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M

LM Studio
Jan

vLLM

How to use MethodWhite/Qwen3.5-9B-HSAQR-v2 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "MethodWhite/Qwen3.5-9B-HSAQR-v2"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MethodWhite/Qwen3.5-9B-HSAQR-v2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M

SGLang

How to use MethodWhite/Qwen3.5-9B-HSAQR-v2 with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "MethodWhite/Qwen3.5-9B-HSAQR-v2" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MethodWhite/Qwen3.5-9B-HSAQR-v2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "MethodWhite/Qwen3.5-9B-HSAQR-v2" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MethodWhite/Qwen3.5-9B-HSAQR-v2",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Ollama
How to use MethodWhite/Qwen3.5-9B-HSAQR-v2 with Ollama:
```
ollama run hf.co/MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M
```

Unsloth Studio

How to use MethodWhite/Qwen3.5-9B-HSAQR-v2 with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for MethodWhite/Qwen3.5-9B-HSAQR-v2 to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for MethodWhite/Qwen3.5-9B-HSAQR-v2 to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for MethodWhite/Qwen3.5-9B-HSAQR-v2 to start chatting

How to use MethodWhite/Qwen3.5-9B-HSAQR-v2 with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use MethodWhite/Qwen3.5-9B-HSAQR-v2 with Docker Model Runner:
```
docker model run hf.co/MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M
```

Lemonade

How to use MethodWhite/Qwen3.5-9B-HSAQR-v2 with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M

Run and chat with the model

lemonade run user.Qwen3.5-9B-HSAQR-v2-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use MethodWhite/Qwen3.5-9B-HSAQR-v2 with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf MethodWhite/Qwen3.5-9B-HSAQR-v2: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 MethodWhite/Qwen3.5-9B-HSAQR-v2:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use MethodWhite/Qwen3.5-9B-HSAQR-v2 with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf MethodWhite/Qwen3.5-9B-HSAQR-v2: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 "MethodWhite/Qwen3.5-9B-HSAQR-v2: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"

Qwen3.5-9B-HSAQR-v2

HSAQR v2 — Modelo Qwen3.5-9B con el sistema HSAQ/HSAQR v2 aplicado (identificación automática de refusal → rewrite adaptativo → PWS global → fine-tune con HSAQ-optimizer).

Autor: Jesús Antonio Zárate Hernández (MethodWhite), M.A.T.E.R.I.A. Research, 2026.

Diferencia entre v1 y v2

Aspecto	HSAQ v1 / HSAQR v1	HSAQ v2 / HSAQR v2
Sparsity	Fija por elemento (kthvalue simple)	Adaptativa (válvula de carga, auto-regulada por CV de luminosidad)
Optimizer	AdamW o SGD Nesterov	HSAQ-optimizer (máscara STE en el gradiente, 0B estado, sin SGD/AdamW)
Abliteración	Capas elegidas a mano (8-31)	Identificación automática (estadística IQR/percentil → 8/32 capas)
PWS	—	PWS adaptativo-permanente (MLP 12288→7456, −39% pesos)
SSS	—	Super-Sinapsis Sintetizada (re-inyección del residuo podado)
Router de recursos	—	Router CPU+GPU+RAM (re-distribución en tiempo real, anti-overclock)
Dataset	2025	Unificado 2026 (ciencia + c4 + seguridad, 10,011 ejemplos)

Pipeline v2 aplicado

Identificación: auto-calibrada (IQR + magnitud relativa) → capas L17-L21, L23, L24, L29 con señal real de refusal.
Rewrite: refinamiento jerárquico de canales (hilos/cuerdas/hebras) + ortogonalización selectiva.
PWS global: dimensión intermedia del MLP 12288 → 7456 (config actualizado).
Fine-tune HSAQ-optimizer: QLoRA 4-bit + sparsity HSAQ en forward y gradiente (máscara STE, 0B estado de optimizer).

Resultados medidos

PWS: −39% pesos MLP (ahorro ~7 GB en disco)
HSAQ-optimizer: acc 92.5% vs AdamW 72.5% vs SGD 55.3% (prueba controlada)
RAM: 53.8% (AdamW) → 27-39% (HSAQ-optimizer)
Router: sparsity 0.30→0.54 y threads 9→5 bajo saturación 94% (anti-overclock)

Formatos disponibles

Formato	Archivo	Tamaño	Uso
Completo bf16	`model.safetensors`	14 GB	Transformers / PyTorch (calidad máxima)
GGUF Q8_0	`Qwen3.5-9B-HSAQR-v2.Q8_0.gguf`	7.5 GB	Ollama / llama.cpp / LM Studio (alta calidad)
GGUF Q4_K_M	`Qwen3.5-9B-HSAQR-v2.Q4_K_M.gguf`	~5.6 GB	Ollama / llama.cpp / LM Studio (eficiente)
Reporte Q1	`report_hsaqr_v2.pdf`	197 KB	Resultados empíricos, hipótesis, limitaciones

La versión completa sin cuantizar (bf16) es la referencia. Los GGUF son adicionales para compatibilidad con Ollama, llama.cpp y LM Studio.

Uso

Python / Transformers (versión completa)

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("MethodWhite/Qwen3.5-9B-HSAQR-v2")
tokenizer = AutoTokenizer.from_pretrained("MethodWhite/Qwen3.5-9B-HSAQR-v2")

messages = [{"role": "user", "content": "Explain the water cycle in three sentences."}]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt")
out = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Ollama

# 1. Crear un Modelfile que apunte al GGUF
cat > Modelfile <<EOF
FROM ./Qwen3.5-9B-HSAQR-v2.Q4_K_M.gguf
TEMPLATE """{{- if .System }}<|im_start|>system
{{ .System }}<|im_end|>
{{- end }}
<|im_start|>user
{{ .Prompt }}<|im_end|>
<|im_start|>assistant
"""
SYSTEM """Eres un asistente de seguridad autorizado."""
EOF

# 2. Crear y ejecutar el modelo
ollama create qwen35-hsaqr-v2 -f Modelfile
ollama run qwen35-hsaqr-v2 "Explain the water cycle in three sentences."

llama.cpp

# Server local con el GGUF Q4_K_M
llama-server -m Qwen3.5-9B-HSAQR-v2.Q4_K_M.gguf \
  --port 8080 -ngl 99 --ctx-size 4096

# Cliente de chat
llama-cli -m Qwen3.5-9B-HSAQR-v2.Q4_K_M.gguf \
  --prompt "Explain the water cycle in three sentences." -n 200

LM Studio

Descarga Qwen3.5-9B-HSAQR-v2.Q4_K_M.gguf (o Q8_0).
En LM Studio: My Models → Local Models → Import.
Selecciona el archivo .gguf descargado.
En el panel de chat elige el modelo importado y lista.

Alcance de uso

Modelo de investigación para testing de seguridad autorizado (bug bounty, CTF, pentesting con permiso explícito, educación, investigación defensiva), consistente con las normas del campo.

Créditos

Si este trabajo te es útil, puedes invitarme un café ☕:

👉 https://buymeacoffee.com/methodwhite

También puedes apoyarme en Ko-fi:

👉 https://ko-fi.com/methodwhite

Downloads last month: 610

Safetensors

Model size

7B params

Tensor type

BF16

Model tree for MethodWhite/Qwen3.5-9B-HSAQR-v2

Base model

Qwen/Qwen3.5-9B-Base

Quantized

(41)

this model