Instructions to use Motif-Technologies/Motif-3-Beta with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Motif-Technologies/Motif-3-Beta with Transformers:

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

pipe = pipeline("text-generation", model="Motif-Technologies/Motif-3-Beta", trust_remote_code=True)
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoModel
model = AutoModel.from_pretrained("Motif-Technologies/Motif-3-Beta", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use Motif-Technologies/Motif-3-Beta with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/Motif-Technologies/Motif-3-Beta

SGLang

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

Docker Model Runner
How to use Motif-Technologies/Motif-3-Beta with Docker Model Runner:
```
docker model run hf.co/Motif-Technologies/Motif-3-Beta
```

Motif-3-Beta

Commit History

Sep 1

08:28
docs: update vLLM image to v0.26.0-motif3 (#11)⁠

iamwyldecat

ad36cea

Jul 31

09:57
tokenizer: also rename ids 55-84 in model.vocab (not just added_tokens)⁠

dongseokmotif

Verified signature
b286b52
09:42
tokenizer: replace company/author name tokens (ids 55-84) with reserved placeholders⁠

dongseokmotif

Verified signature
e9b3a7b

Jul 30

08:33
docs: add MTP (Multi-Token Prediction) to the Architecture section

dongseokmotif

Verified signature
eee2817
08:31
docs: document built-in MTP head / self-speculative decoding (vLLM --speculative-config, num_speculative_tokens=1)

dongseokmotif

Verified signature
672a4d4

Jul 22

12:22
Update README.md

TaehyunKimMotif

Verified signature
f134005
08:33
Update README.md

iamwyldecat

Verified signature
df4e953
07:16
modeling: require attn_implementation=flash_attention_2 (eager/sdpa mishandle GDLA GQA + per-layer sliding window); fail fast with a clear message

dongseokmotif

Verified signature
0fe349f
06:38
docs: drop WIP from HF.generate section (modeling code fixed and working)

dongseokmotif

Verified signature
d2c9ac6
06:29
docs: update HF.generate note — modeling code fixed, now runs; vLLM recommended pending full parity

dongseokmotif

Verified signature
d843b3d
06:27
docs: make HF.generate example runnable (return_dict=True, **inputs, slice by input_ids length)

dongseokmotif

Verified signature
edd84db
06:20
Fix modeling/config: RoPE inv_freq init, per-expert PolyNorm, precision; update HF.generate usage example⁠

dongseokmotif

Verified signature
b446ac1
04:18
Update README.md

iamwyldecat

Verified signature
8a76d55

Jul 21

01:11
Restore license (from @iamwyldecat ) + keep GDLA expansion

dongseokmotif

Verified signature
32b0305
01:00
Expand GDLA -> Grouped Differential Latent Attention

dongseokmotif

Verified signature
c3b189e

Jul 20

23:59
Update license

iamwyldecat

Verified signature
2520876
23:32
Set license to Apache-2.0

dongseokmotif

Verified signature
94cd932
23:24
Finalize public model card (multilingual, architecture, no gating)

dongseokmotif

Verified signature
d03e855
23:23
Add files using upload-large-folder tool

dongseokmotif

Verified signature
a13b5c1
23:23
Add files using upload-large-folder tool

dongseokmotif

Verified signature
dd24bba
23:22
Add files using upload-large-folder tool

dongseokmotif

Verified signature
d77de33
23:21
Add files using upload-large-folder tool

dongseokmotif

Verified signature
95db90e
23:20
Add files using upload-large-folder tool

dongseokmotif

Verified signature
b5af44a
23:16
Add model card (test commit)

dongseokmotif

Verified signature
4f6413a
22:03
initial commit

dongseokmotif

Verified signature
3e72297

Commit History

Sep 1

docs: update vLLM image to v0.26.0-motif3 (#11)⁠

Jul 31

tokenizer: also rename ids 55-84 in model.vocab (not just added_tokens)⁠

tokenizer: replace company/author name tokens (ids 55-84) with reserved placeholders⁠

Jul 30

docs: add MTP (Multi-Token Prediction) to the Architecture section

docs: document built-in MTP head / self-speculative decoding (vLLM --speculative-config, num_speculative_tokens=1)

Jul 22

Update README.md

Update README.md

modeling: require attn_implementation=flash_attention_2 (eager/sdpa mishandle GDLA GQA + per-layer sliding window); fail fast with a clear message

docs: drop WIP from HF.generate section (modeling code fixed and working)

docs: update HF.generate note — modeling code fixed, now runs; vLLM recommended pending full parity

docs: make HF.generate example runnable (return_dict=True, **inputs, slice by input_ids length)

Fix modeling/config: RoPE inv_freq init, per-expert PolyNorm, precision; update HF.generate usage example⁠

Update README.md

Jul 21

Restore license (from @iamwyldecat ) + keep GDLA expansion

Expand GDLA -> Grouped Differential Latent Attention

Jul 20

Update license

Set license to Apache-2.0

Finalize public model card (multilingual, architecture, no gating)

Add files using upload-large-folder tool

Add files using upload-large-folder tool

Add files using upload-large-folder tool

Add files using upload-large-folder tool

Add files using upload-large-folder tool

Add model card (test commit)

initial commit