Instructions to use MTEnt/dot with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use MTEnt/dot with Transformers:

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

pipe = pipeline("text-generation", model="MTEnt/dot")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("MTEnt/dot")
model = AutoModelForCausalLM.from_pretrained("MTEnt/dot", 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

vLLM

How to use MTEnt/dot with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/MTEnt/dot

SGLang

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

Docker Model Runner
How to use MTEnt/dot with Docker Model Runner:
```
docker model run hf.co/MTEnt/dot
```

dot

File size: 2,488 Bytes

954544e

{
  "format_version": 1,
  "name": "Dot",
  "release": "v0.4-thinking",
  "released_by": "MTEnt",
  "license": "Apache-2.0",
  "artifact": "complete BF16 text backbone plus recurrent-depth core",
  "architecture": {
    "class": "DotRecurrentDepthModel",
    "backbone_parameters": 8953803264,
    "recurrent_core_parameters": 864945224,
    "total_instantiated_parameters": 9818748488,
    "insertion_after_layer": 15,
    "copied_source_layers": [12, 13, 14, 15],
    "maximum_loops": 8,
    "active_loops": 4,
    "cache_supported": false,
    "precision": "bfloat16"
  },
  "lineage": {
    "upstream_repository": "Qwen/Qwen3.5-9B",
    "upstream_exact_revision": null,
    "upstream_revision_note": "The original training manifest did not record the exact source commit.",
    "semantic_stage": "Dot-9B-Semantic-v0.2 merged LoRA rank 32",
    "semantic_train_records": 11763,
    "recurrent_stage_source": "Dot-9B-Recurrent-v0.3",
    "thinking_repair_source_step": 938
  },
  "training": {
    "thinking_records": 60000,
    "thinking_record_sha256": "ff84822f7cf85be9c1a1e392e9c358b01b56d68f9aff821bcbc41c99534853e5",
    "validation_records": 4096,
    "validation_record_sha256": "e46a96db4c70398f7aacf02b7b900d21302f9560926b5cd0b133c2089891ef6c",
    "optimizer_steps": 938,
    "tokens_seen": 8642015,
    "backbone_frozen": true,
    "final_loss": 0.01007067202590406,
    "loop_scales": [
      0.15769609808921814,
      0.07985208928585052,
      0.05903945118188858,
      0.059150855988264084,
      0.0,
      0.0,
      0.0,
      0.0
    ]
  },
  "weights": {
    "model-00001-of-00004.safetensors": "330fa6fed85e394ba9dec36f986e017e51b8f3c588613bc0dc7f0ea8521aac14",
    "model-00002-of-00004.safetensors": "fa04e853ff08aef3b8ef51e2f2e602b8b3da7d1d9072925602812c052502e011",
    "model-00003-of-00004.safetensors": "6aa1824b165e566f6bcb6038c6dab6abd1c4b392d5e8cc0173d1f859c009f477",
    "model-00004-of-00004.safetensors": "dbb205aa4fab083cb7f409b337a34616a559d76fe7c834f48396da38d6de1a3a",
    "reasoning_core.safetensors": "7c897b83176c044a17840c90f4123ce7bcce437139aad1e84f76aa1555db6152"
  },
  "known_limits": [
    "custom loader required",
    "cache-backed decoding unsupported",
    "text only",
    "H200 BF16 runtime is the only verified hardware path",
    "targeted synthetic reasoning evaluation is not a broad capability benchmark",
    "new spatial generalization probe scored 21.875 percent exact match",
    "no independent safety evaluation"
  ]
}