You can deploy this model and start earning money today!

Mindcraft-CE/
Andy-4.1-GGUF

Name: Andy-4.1-GGUF
Availability: InStock

Mindcraft-CE/Andy-4.1

endpoints_compatible

conversational

Views

No views yet

Andy-4.1 is a revolutionary model, bringing higher performance per parameter compared to Andy-4, making it the most powerful Andy model near it's size thus far.

Andy-4.1 takes a new approach for building a model to play Minecraft: Generalize, don't Specialize. This approach helps Andy-4.1 deal with new situations, new tools, as well as novel environments.

[!Important] These are the GGUF weights for Andy-4.1, if you meant to find the safetensors weights, visit this huggingface page

Key Additions

Constant Chain of Thought: Unlike Andy-4, Andy-4.1 has been built specifically to think before acting. Although this does increase the amount of time per action, it allows Andy-4.1 to be more thorough in it's decisions.
Vision Capabilities: This is the first Andy model to have vision capabilities, extending it's ability to not only act, but to understand.
Increased Message Counts: A side effect of introducing reasoning has been expanding the ability to dissect previous actions, and determine why they were made, allowing Andy-4.1 to understand more of what the world state is.

Why Andy-4.1?

Andy-4.1 exists due to experimentation of model architecture, and training methology. Andy-4.1 utilizes an experimental architecture borrowed from the GRaPE series of models by SLAI. Future versions of Andy, such as Andy-5, will be developed solely off of the GRaPE family of models.

The base model to Andy-4.1 is yet to be released, the LoRA weights are not planned to be released for some time, for now the Safetensors, OpenVINO, and GGUF versions of Andy-4.1 will be avaliable.

[!Important] Andy-4.1 is an experimental model. Preliminary tests show it to be mostly stable under nominal conditions.

Further refinement of the training data, as well as the architecture will improve the accuracy, and reliability of future Andy models.

Model Specifications

Model Size: 3B parameters
Architecture: Modified Qwen3 VL
Context Length: Up to 256,000 tokens
Message Count: Stable up to 65 messages
CoT Style: DeepSeek-R1 style.

Training Specifications

Hardware: 1x RTX 3090
Training Time: 42 Hours
Dataset Size: 130,000 examples
Learning Rate: 2e-5
LR Scheduler: cosine
Epoch Count: 1 Epoch
Training Quantization: BF16 with QAT for 8-bit precision

Known Issues

Andy-4.1, as stated, is an experimental model. It explores the real-world use cases of a unique, modified architecture, a new training style for Andy models, and attempts to push the limits for model it's size. To be completely transparent, here is what the Mindcraft team had found during analysis:

Repetition during long contexts
Excessive usage of correct tools
Overthinking, although the result does end with a correct tool call
Confusion over newer updates to Minecraft
Overlooks small details often, such as needing a crafting table nearby to build something

While these issues seem small, they begin to stack up during long, agentic sessions of playing with Andy-4.1, or having it play for you.

How to use Andy-4.1

Andy-4.1 is not on Ollama's library. I personally have moved to using LM Studio over Ollama, which is a lot easier to set up, download new models, and actually has settings to ensure you download the correct model for yourself.

LM Studio already has their own guide on how to set up an API endpoint, I will link that Here. I recommend following that guide.

After downloading Andy-4.1, which you can follow this guide to find out how to do, you need to make a new profile, or edit an existing one with the content here:

json

{
    "name": "andy-4.1",

    "model": {
    "api": "openai",
    "model": "andy-4.1",
    "url": "http://localhost:1234/v1"
    }
}

This uses the default LM Studio URL, with the model andy-4.1, it should work in almost any case. If not, join the Mindcraft Discord Server and ask for support from either Not Andy, or one of the official devs (preferably @Sweaterdog) to assist you.

The recommended sampling parameters for Andy-4.1 are as follows:

Name	Value
Temperature	0.4
Repeat Penalty	1.15
Top P Sampling	0.9
Min P Sampling	0.05

You are welcome, even encouraged to experiment with other sampling parameters to finetune the best version of Andy-4.1

Andy-4.1 does have a non-thinking mode via changing the chat template, which is another feature of LM Studio, of which a guide can be found here. I generally do not recommend this, it can lead to laziness, repetition, as well as poor output quality. But if you like tinkering, below is the chat template to disable thinking.

Chat Template

jinja

{%- if tools %}
    {{- '<|im_start|>system\n' }}
    {%- if messages[0].role == 'system' %}
        {%- if messages[0].content is string %}
            {{- messages[0].content }}
        {%- else %}
            {%- for content in messages[0].content %}
                {%- if 'text' in content %}
                    {{- content.text }}
                {%- endif %}
            {%- endfor %}
        {%- endif %}
        {{- '\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><|im_end|>\n" }}
{%- else %}
    {%- if messages[0].role == 'system' %}
        {{- '<|im_start|>system\n' }}
        {%- if messages[0].content is string %}
            {{- messages[0].content }}
        {%- else %}
            {%- for content in messages[0].content %}
                {%- if 'text' in content %}
                    {{- content.text }}
                {%- endif %}
            {%- endfor %}
        {%- endif %}
        {{- '<|im_end|>\n' }}
    {%- endif %}
{%- endif %}
{%- set image_count = namespace(value=0) %}
{%- set video_count = namespace(value=0) %}
{%- for message in messages %}
    {%- if message.role == "user" %}
        {{- '<|im_start|>' + message.role + '\n' }}
        {%- if message.content is string %}
            {{- message.content }}
        {%- else %}
            {%- for content in message.content %}
                {%- if content.type == 'image' or 'image' in content or 'image_url' in content %}
                    {%- set image_count.value = image_count.value + 1 %}
                    {%- if add_vision_id %}Picture {{ image_count.value }}: {% endif -%}
                    <|vision_start|><|image_pad|><|vision_end|>
                {%- elif content.type == 'video' or 'video' in content %}
                    {%- set video_count.value = video_count.value + 1 %}
                    {%- if add_vision_id %}Video {{ video_count.value }}: {% endif -%}
                    <|vision_start|><|video_pad|><|vision_end|>
                {%- elif 'text' in content %}
                    {{- content.text }}
                {%- endif %}
            {%- endfor %}
        {%- endif %}
        {{- '<|im_end|>\n' }}
    {%- elif message.role == "assistant" %}
        {{- '<|im_start|>' + message.role + '\n' }}
        {%- if message.content is string %}
            {{- message.content }}
        {%- else %}
            {%- for content_item in message.content %}
                {%- if 'text' in content_item %}
                    {{- content_item.text }}
                {%- endif %}
            {%- endfor %}
        {%- endif %}
        {%- if message.tool_calls %}
            {%- for tool_call in message.tool_calls %}
                {%- if (loop.first and message.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 %}
        {{- '<|im_end|>\n' }}
    {%- elif message.role == "tool" %}
        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
            {{- '<|im_start|>user' }}
        {%- endif %}
        {{- '\n<tool_response>\n' }}
        {%- if message.content is string %}
            {{- message.content }}
        {%- else %}
            {%- for content in message.content %}
                {%- if content.type == 'image' or 'image' in content or 'image_url' in content %}
                    {%- set image_count.value = image_count.value + 1 %}
                    {%- if add_vision_id %}Picture {{ image_count.value }}: {% endif -%}
                    <|vision_start|><|image_pad|><|vision_end|>
                {%- elif content.type == 'video' or 'video' in content %}
                    {%- set video_count.value = video_count.value + 1 %}
                    {%- if add_vision_id %}Video {{ video_count.value }}: {% endif -%}
                    <|vision_start|><|video_pad|><|vision_end|>
                {%- elif 'text' in content %}
                    {{- content.text }}
                {%- endif %}
            {%- endfor %}
        {%- endif %}
        {{- '\n</tool_response>' }}
        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
            {{- '<|im_end|>\n' }}
        {%- endif %}
    {%- endif %}
{%- endfor %}
{%- if add_generation_prompt %}
    {{- '<|im_start|>assistant\n' }}
{%- endif %}

Acknowledgements

There are user reported errors from Andy-4.1 right now, below is a debug section to help you.

How to fix issues with Andy-4.1

Error rendering prompt with jinja template: "Expected iterable or object type in for loop: got UndefinedValue" Should be expected using the GGUFs of Andy-4.1. Somewhere in the pipeline between llama.cpp or LM Studio itself, the chat template changes, which makes the model not work. Follow this LM Studio Guide for how to set the chat template of the model, and use this chat template.

If you notice Andy-4.1 "thinking out loud," most likely your issue is Context Length Truncation. LM Studio doesn't automatically set the context to the longest amount. When you go to load a model there is a toggle that says "Manually choose model load parameters," activate that. Afterwards, click on andy-4.1 in the "Your Models" section.

Next, find the "Context Length" section, and ensure it is at least 16000 (Note that LM Studio doesn't use commas for large numbers). For insurance sake, check the "Remember settings for andy-4.1" box in the bottom left of that menu, and Load the Model.

What's Next?

Based on the lessons from Andy-4.1, the Mindcraft team is prepared to collect better training data, explore new architectures to make the cost of running Andy models cheaper, as well as packing more brains into these tiny minds.

Licenses and Notices

Like all other Andy models, Andy-4.1 is based on the Andy license of terms. Being generally permissive, it contains qualifiers as to what makes an "Andy" class model.

See Andy 2.0 License.

This work uses data and models created by @Sweaterdog.

Mindcraft-CE/Andy-4.1-GGUF