Qwen3-8b-192k-Context-6X-Josiefied-Uncensored-MLX-AWQ-4bit

📌 Overview

A 4-bit AWQ quantized version of Qwen3-8B optimized for efficient inference using the MLX library, designed to handle long-context tasks (192k tokens) with reduced resource usage. Retains core capabilities of Qwen3-8B while enabling deployment on edge devices.

📈 Performance Metrics

🚨 Important: Prompt Template for LM Studio Use

You need to modify the prompt template to ensure compatibility with LM Studio's inference pipeline. Below is the required template structure:

{%- if tools %}
    {{- '\/system\n' }}
    {%- if messages[0].role == 'system' %}
        {{- messages[0].content + '\n\n' }}
    {%- endif %}
    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
    {%- for tool in tools %}
        {{- "\n" }}
        {{- tool | tojson }}
    {%- endfor %}
    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call>...</tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call>\n" }}
{%- else %}
    {%- if messages[0].role == 'system' %}
        {{- '\/system\n' + messages[0].content + '\/\n' }}
    {%- endif %}
{%- endif %}

{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
{%- for message in messages[::-1] %}
    {%- set index = (messages|length - 1) - loop.index0 %}
    {%- set tool_start = "｟" %}
    {%- set tool_start_length = tool_start|length %}
    {%- set start_of_message = message.content[:tool_start_length] %}
    {%- set tool_end = "｠" %}
    {%- set tool_end_length = tool_end|length %}
    {%- set start_pos = (message.content|length) - tool_end_length %}
    {%- if start_pos < 0 %}
        {%- set start_pos = 0 %}
    {%- endif %}
    {%- set end_of_message = message.content[start_pos:] %}
    {%- if ns.multi_step_tool and message.role == "user" and not(start_of_message == tool_start and end_of_message == tool_end) %}
        {%- set ns.multi_step_tool = false %}
        {%- set ns.last_query_index = index %}
    {%- endif %}
{%- endfor %}

{%- for message in messages %}
    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
        {{- '\/' + message.role + '\n' + message.content + '\/' + '\n' }}
    {%- elif message.role == "assistant" %}
        {%- set content = message.content %}
        {%- set reasoning_content = '' %}
        {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
            {%- set reasoning_content = message.reasoning_content %}
        {%- else %}
            {%- if '\/' in message.content %}
                {%- set content = (message.content.split('\/')|last).lstrip('\n') %}
                {%- set reasoning_content = (message.content.split('\/')|first).rstrip('\n') %}
                {%- set reasoning_content = (reasoning_content.split('')|last).lstrip('\n') %}
            {%- endif %}
        {%- endif %}

        {%- if loop.index0 > ns.last_query_index %}
            {%- if loop.last or (not loop.last and reasoning_content) %}
                {{- '\/' + message.role + '\n\n' + reasoning_content.strip('\n') + '\n\/\n' + content.lstrip('\n') }}
            {%- else %}
                {{- '\/' + message.role + '\n' + content }}
            {%- endif %}
        {%- else %}
            {{- '\/' + message.role + '\n' + content }}
        {%- endif %}

        {%- if message.tool_calls %}
            {%- for tool_call in message.tool_calls %}
                {%- if (loop.first and content) or (not loop.first) %}
                    {{- '\n' }}
                {%- endif %}
                {%- if tool_call.function %}
                    {%- set tool_call = tool_call.function %}
                {%- endif %}
                {{- '<tool_call>\n{"name": "' }}
                {{- tool_call.name }}
                {{- '", "arguments": ' }}
                {%- if tool_call.arguments is string %}
                    {{- tool_call.arguments }}
                {%- else %}
                    {{- tool_call.arguments | tojson }}
                {%- endif %}
                {{- '}\n</tool_call>' }}
            {%- endfor %}
        {%- endif %}
        {{- '\/\n' }}
    {%- elif message.role == "tool" %}
        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
            {{- '\/user' }}
        {%- endif %}
        {{- '\n｟\n' }}
        {{- message.content }}
        {{- '\n｠' }}
        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
            {{- '\/\n' }}
        {%- endif %}
    {%- endif %}
{%- endfor %}

{%- if add_generation_prompt %}
    {{- '\/assistant\n' }}
    {%- if enable_thinking is defined and enable_thinking is false %}
        {{- '

🧠 Model Details

• Base Model: Qwen3-8B
• Quantization: AWQ Q4 (4-bit) via MLX library
• Context Length: 192,000 tokens (6x longer than standard)
• Library: MLX (optimized for Apple Silicon, macOS)
• License: Apache 2.0
• Pipeline: text-generation
• Tags: not-for-all-audiences, conversational, mlx

🧩 Key Features

• Efficient Inference: 4-bit quantization reduces memory footprint by ~75% vs. FP16
• Long Context Support: 192k tokens for complex tasks (e.g., document analysis, code generation)
• Cross-Platform: Works on macOS with MLX for Apple Silicon acceleration
• Customizable Prompting: Adjust templates for compatibility with tools like LM Studio

📌 Use Cases

• Research: Long-context NLP experiments, model compression studies
• Development: Edge deployments, real-time chatbots with extended context
• Enterprise: Cost-effective AI solutions for document processing and code generation

⚠️ Bias, Risks & Limitations

Potential Biases

• Trained on diverse data but may inherit societal biases (e.g., gender, cultural assumptions)
• "Not-for-all-audiences" tag indicates potential for generating sensitive content

Technical Limitations

• 4-bit quantization may slightly reduce accuracy on complex tasks
• Performance depends on hardware (MLX optimized for Apple Silicon)

Mitigation Strategies

• Review outputs for sensitive content
• Use in controlled environments with monitoring

🧪 How to Get Started

Installation

# Install MLX (Apple Silicon only)
pip install mlx

# Load model with Hugging Face Transformers
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("Goraint/Qwen3-8b-192k-Context-6X-Josiefied-Uncensored-MLX-AWQ-4bit", device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("Goraint/Qwen3-8b-192k-Context-6X-Josiefied-Uncensored-MLX-AWQ-4bit")

Example Usage

prompt = "Explain quantum computing in simple terms."
inputs = tokenizer(prompt, return_tensors="pt").to("mps")
outputs = model.generate(**inputs, max_length=200)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))

🌍 Environmental Impact

• Estimated CO2 Emissions: Calculate using ML Impact Calculator
  • Hardware: Apple M1 Pro (16GB RAM)
  • Training Time: N/A (quantized from pre-trained model)

🧰 Community & Resources

• Documentation: Hugging Face Docs
• GitHub Issues: Report bugs or feature requests
• Forums: Hugging Face Discuss

📄 License

Apache 2.0

Note: This model is a community contribution and may not be officially supported by Alibaba Cloud. Always validate outputs for accuracy and safety in production environments.