Marco-DeepResearch-8B-imatrix-GGUF

Importance-matrix (imatrix) GGUF quantized versions of AIDC-AI/Marco-DeepResearch-8B for use with llama.cpp and compatible inference engines.

For standard quantizations without importance matrix, see Marco-DeepResearch-8B-GGUF.

About the Model

Marco DeepResearch is an efficient 8B-scale deep research agent developed by Alibaba International Digital Commerce (AIDC-AI), based on Qwen3-8B. It autonomously conducts open-ended investigations by integrating complex information retrieval with multi-step reasoning across diverse web sources. The model uses tools (search, visit) for iterative web research with built-in verification.

Under a maximum budget of 600 tool calls, Marco DeepResearch significantly outperforms other 8B-scale agents and surpasses or approaches several 30B-scale agents on challenging benchmarks.

What is Importance Matrix (imatrix) Quantization?

Standard quantization treats all model weights equally when reducing precision. Importance matrix quantization improves on this by first profiling the model with calibration data to determine which weights matter most for output quality. During quantization, more important weights are preserved with higher precision while less critical weights are compressed more aggressively.

The result is significantly better quality at low bit rates (Q3, Q2, IQ3, IQ2, IQ1) compared to standard quantization. At higher bit rates (Q5, Q6), the difference is minimal. If you need to run at 3 bits or below, imatrix quants are strongly recommended.

Calibration data: WikiText-2

Included Files

The imatrix.dat file (5.1 MB) is included in this repository. You can use it with llama.cpp's llama-quantize to create your own custom quantizations from a full-precision GGUF.

Available Quantizations

Choosing a Quantization

• Best quality: Q6_K or Q5_K_M — recommended if you have sufficient RAM/VRAM.
• Best balance: Q4_K_M — recommended for most users on consumer hardware.
• Memory constrained: Q3_K_M or IQ3_M — imatrix provides clear quality gains at this level.
• Extreme constraints: IQ2_M or IQ2_S — only viable with imatrix; expect significant quality loss.
• Ultra-low (research): IQ1_M / IQ1_S — extreme compression for experimentation.

Creating Custom Quantizations

You can use the included imatrix.dat to create your own quants:

./llama-quantize --imatrix imatrix.dat \
  Marco-DeepResearch-8B-f16.gguf \
  Marco-DeepResearch-8B-i1-<QUANT_TYPE>.gguf \
  <QUANT_TYPE>

Usage

llama.cpp

CPU inference:

./llama-cli -m Marco-DeepResearch-8B-i1-Q5_K_M.gguf \
  -p "<your prompt>" \
  -n 4096 \
  --temp 0.7 --top-p 0.95 \
  -t $(nproc)

GPU-accelerated inference:

./llama-cli -m Marco-DeepResearch-8B-i1-Q5_K_M.gguf \
  -p "<your prompt>" \
  -n 4096 \
  --temp 0.7 --top-p 0.95 \
  -ngl 99

Server mode (OpenAI-compatible API):

./llama-server -m Marco-DeepResearch-8B-i1-Q5_K_M.gguf \
  --port 8080 \
  -ngl 99 \
  -c 32768

Ollama

Create a Modelfile:

FROM ./Marco-DeepResearch-8B-i1-Q5_K_M.gguf

PARAMETER temperature 0.7
PARAMETER top_p 0.95
PARAMETER num_ctx 32768

Then:

ollama create marco-deepresearch -f Modelfile
ollama run marco-deepresearch

Python (llama-cpp-python)

from llama_cpp import Llama

llm = Llama(
    model_path="Marco-DeepResearch-8B-i1-Q5_K_M.gguf",
    n_ctx=32768,
    n_gpu_layers=-1,  # Use all GPU layers; set to 0 for CPU-only
)

output = llm(
    "<your prompt>",
    max_tokens=4096,
    temperature=0.7,
    top_p=0.95,
)

print(output["choices"][0]["text"])

LM Studio

1. Download your desired quantization file (e.g., Marco-DeepResearch-8B-i1-Q4_K_M.gguf).
2. Open LM Studio and import the model file.
3. Configure generation settings: Temperature 0.7, Top-P 0.95.
4. Set context length to 32768 (or higher if your hardware allows).
5. Start chatting or use the local server API.

Prompt Format

This model uses a structured prompt format with <think>, <tool_call>, and <answer> tags.

System Prompt Template

You are an expert web researcher. Your task is to find accurate, complete answers through iterative search, extraction, and verification.

## Core Principles

1) Strategic Planning
   - Decompose complex questions into targeted sub-tasks
   - Choose the right tool for each step
   - Refine your approach based on what you learn

2) Precise Execution
   - Define clear objectives before using any tool
   - Provide sufficient detail for accurate results
   - Avoid vague or overly broad requests

3) Rigorous Verification
   - Cross-check important facts across multiple sources
   - Resolve conflicts by gathering additional evidence
   - Only conclude when evidence is sufficient and consistent

## Output Format

In each turn, you can either call a tool or provide the final answer.

**Call a tool:**
<think>your reasoning process</think>
<tool_call>
{"name": "tool_name", "arguments": {"param1": "value1", "param2": "value2"}}
</tool_call>

**Provide final answer (when you have gathered enough information):**
<think>your reasoning and analysis</think>
<answer>the direct answer to the question</answer>

Note: All reasoning should be in <think>, <answer> should contain only the final answer.

Current date: {current_date}

# Tools

You may call one or more functions to assist with the user query.

You are provided with function signatures within <tools></tools> XML tags:
<tools>
{tools_json}
</tools>

For each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:
<tool_call>
{"name": <function-name>, "arguments": <args-json-object>}
</tool_call>

Tool Definitions

The model expects tools in OpenAI function calling format:

[
  {
    "type": "function",
    "function": {
      "name": "search",
      "description": "Search the web via Google to find relevant information and URLs.",
      "parameters": {
        "type": "object",
        "properties": {
          "querys": {
            "type": "array",
            "items": {"type": "string"},
            "description": "Search queries for finding relevant information."
          }
        },
        "required": ["querys"]
      }
    }
  },
  {
    "type": "function",
    "function": {
      "name": "visit",
      "description": "Read webpage content to extract specific information, verify claims, or understand context.",
      "parameters": {
        "type": "object",
        "properties": {
          "urls": {
            "type": "array",
            "items": {"type": "string"},
            "description": "URL(s) to visit."
          },
          "goal": {
            "type": "string",
            "description": "The specific information to retrieve. Be precise, not vague."
          }
        },
        "required": ["urls", "goal"]
      }
    }
  }
]

Model Output Example

Tool call turn:

<think>
I need to search for information about X to answer the user's question.
</think>
<tool_call>
{"name": "search", "arguments": {"querys": ["search query here"]}}
</tool_call>

Final answer turn:

<think>
Based on the evidence gathered from multiple sources, I can now conclude that...
</think>
<answer>
The direct answer to the question.
</answer>

Benchmark Results

Evaluated on a suite of deep search benchmarks under a maximum budget of 600 tool calls (results from the original unquantized model).

Original Model

This is a quantized version of AIDC-AI/Marco-DeepResearch-8B. Please refer to the original model card for full details on training methodology, intended use, and limitations.

• Paper: Marco DeepResearch: Unlocking Efficient Deep Research Agents via Verification-Centric Design
• Code: GitHub

Citation

@article{zhu2026marco,
  title={Marco DeepResearch: Unlocking Efficient Deep Research Agents via Verification-Centric Design},
  author={Bin Zhu and Qianghuai Jia and Tian Lan and Junyang Ren and Feng Gu and Feihu Jiang and Longyue Wang and Zhao Xu and Weihua Luo},
  journal={arXiv preprint arXiv:2603.28376},
  year={2026}
}

License

This model is released under the Apache 2.0 License.