Instructions to use DMindAI/DMind-3-mini with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use DMindAI/DMind-3-mini with Transformers:

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

pipe = pipeline("text-generation", model="DMindAI/DMind-3-mini")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
pipe(text=messages)

# Load model directly
from transformers import AutoProcessor, AutoModelForImageTextToText

processor = AutoProcessor.from_pretrained("DMindAI/DMind-3-mini")
model = AutoModelForImageTextToText.from_pretrained("DMindAI/DMind-3-mini")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/p-blog/candy.JPG"},
            {"type": "text", "text": "What animal is on the candy?"}
        ]
    },
]
inputs = processor.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(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use DMindAI/DMind-3-mini with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "DMindAI/DMind-3-mini"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "DMindAI/DMind-3-mini",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/DMindAI/DMind-3-mini

SGLang

How to use DMindAI/DMind-3-mini 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 "DMindAI/DMind-3-mini" \
    --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": "DMindAI/DMind-3-mini",
		"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 "DMindAI/DMind-3-mini" \
        --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": "DMindAI/DMind-3-mini",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use DMindAI/DMind-3-mini with Docker Model Runner:
```
docker model run hf.co/DMindAI/DMind-3-mini
```

yuzhe commited on Mar 22

Commit

abe49dc

1 Parent(s): 9d565aa

Correct README base model to Qwen3.5-4B

Browse files

Files changed (1) hide show

README.md +7 -7

README.md CHANGED Viewed

@@ -19,7 +19,7 @@ metrics:
 pipeline_tag: text-generation
 inference: false
 base_model:
-- Qwen/Qwen3.5-27B
 ---
@@ -27,7 +27,7 @@ base_model:
 <tr>
 <td bgcolor="#EEF6FF" style="padding: 14px 16px; border-left: 6px solid #2563EB;">
 <strong>Update Notice</strong><br>
-This release has been retrained on the same dataset with the base model upgraded to <strong>Qwen3.5-27B</strong>. If any section below has not yet been fully refreshed, this notice takes precedence.
 </td>
 </tr>
 </table>
@@ -65,12 +65,12 @@ The DMind lineage was born from a singular conviction that decentralized finance
 * **Model Name:** DMind-3-mini
 * **Organization:** DMind
-* **Base Architecture:** Qwen3.5-27B (Customized Transformer w/ RoPE)
-* **Parameter Count:** 27 Billion
 * **Precision:** **BF16 (Native)**
     * *⚠️ Note: We strictly advise against 4-bit quantization for financial logic tasks to preserve numerical precision in APY/IL calculations.*
 * **Context Window:** 128k tokens
-* **Hardware Requirement:** BF16 deployment is recommended on a single **80GB-class GPU** (such as A100/H100) or an equivalent multi-GPU setup.
 ## 3. 🔬 Methodology: C³-SFT
@@ -140,10 +140,10 @@ Evaluated on three key benchmarks: DMind Benchmark (Web3 Native Logic), FinanceQ
 ![Figure 3: Performance Benchmarks](./Figures/Figure3.png)
-The evaluation compares the current DMind-3-mini release, now based on Qwen3.5-27B, against top-tier frontier models (GPT-5.1, Claude Sonnet 4.5) and other strong baselines. This version prioritizes deeper reasoning capacity and domain performance over ultra-lightweight deployment, especially in specialized financial tasks.
 ## 7. ⚖️ Limitations & Disclaimer
-* **High Hardware Barrier:** Due to the decision to retain BF16 precision for financial accuracy, straightforward deployment typically requires **80GB-class GPU memory or an equivalent multi-GPU setup**. It is not suitable for standard office laptops.
 * **Knowledge Cutoff:** While the logic is robust, specific protocol data is limited to the training cutoff. Use with RAG for real-time data.
 * **Legal Disclaimer:** This model is an **analytical tool**, not a financial advisor. The output (NFA) should never be the sole basis for investment decisions. The developers assume no liability for financial losses.

 pipeline_tag: text-generation
 inference: false
 base_model:
+- Qwen/Qwen3.5-4B
 ---
 <tr>
 <td bgcolor="#EEF6FF" style="padding: 14px 16px; border-left: 6px solid #2563EB;">
 <strong>Update Notice</strong><br>
+This release has been retrained on the same dataset with the base model upgraded to <strong>Qwen3.5-4B</strong>. If any section below has not yet been fully refreshed, this notice takes precedence.
 </td>
 </tr>
 </table>
 * **Model Name:** DMind-3-mini
 * **Organization:** DMind
+* **Base Architecture:** Qwen3.5-4B (Customized Transformer w/ RoPE)
+* **Parameter Count:** 4.2 Billion
 * **Precision:** **BF16 (Native)**
     * *⚠️ Note: We strictly advise against 4-bit quantization for financial logic tasks to preserve numerical precision in APY/IL calculations.*
 * **Context Window:** 128k tokens
+* **Hardware Requirement:** GPU with $\ge$ **12GB VRAM** (Recommended: NVIDIA RTX 4070Ti+, Apple M3/M4 Pro/Max).
 ## 3. 🔬 Methodology: C³-SFT
 ![Figure 3: Performance Benchmarks](./Figures/Figure3.png)
+The evaluation compares DMind-3-mini (4B) against top-tier frontier models (GPT-5.1, Claude Sonnet 4.5) and other efficient models. Despite its compact size, the Mini model demonstrates exceptional efficiency, particularly in specialized domain tasks where it outperforms significantly larger generalist models.
 ## 7. ⚖️ Limitations & Disclaimer
+* **High Hardware Barrier:** Due to the decision to retain BF16 precision for financial accuracy, this model **requires >= 12GB VRAM**. It is not suitable for standard office laptops.
 * **Knowledge Cutoff:** While the logic is robust, specific protocol data is limited to the training cutoff. Use with RAG for real-time data.
 * **Legal Disclaimer:** This model is an **analytical tool**, not a financial advisor. The output (NFA) should never be the sole basis for investment decisions. The developers assume no liability for financial losses.