Instructions to use jtatman/TinyDolphin-3x-MoE with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use jtatman/TinyDolphin-3x-MoE with Transformers:

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

pipe = pipeline("text-generation", model="jtatman/TinyDolphin-3x-MoE")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("jtatman/TinyDolphin-3x-MoE")
model = AutoModelForCausalLM.from_pretrained("jtatman/TinyDolphin-3x-MoE")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use jtatman/TinyDolphin-3x-MoE with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "jtatman/TinyDolphin-3x-MoE"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jtatman/TinyDolphin-3x-MoE",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/jtatman/TinyDolphin-3x-MoE

SGLang

How to use jtatman/TinyDolphin-3x-MoE 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 "jtatman/TinyDolphin-3x-MoE" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jtatman/TinyDolphin-3x-MoE",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "jtatman/TinyDolphin-3x-MoE" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "jtatman/TinyDolphin-3x-MoE",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use jtatman/TinyDolphin-3x-MoE with Docker Model Runner:
```
docker model run hf.co/jtatman/TinyDolphin-3x-MoE
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

TinyDolphin-3x-MoE

TinyDolphin-3x-MoE is a Mixure of Experts (MoE) made with the following models using LazyMergekit:

🧩 Configuration

base_model: cognitivecomputations/TinyDolphin-2.8.1-1.1b
gate_mode: hidden
dtype: float16
experts:
  - source_model: cognitivecomputations/TinyDolphin-2.8.1-1.1b
    positive_prompts: 
    - "think step-by-step and follow these instructions"
    - "read the following passage, and summarize it in less than 30 words."
    - "please answer this question, consider the options carefully, and return the most likely answer."
  - source_model: cognitivecomputations/TinyDolphin-2.8.1-1.1b
    positive_prompts: ["produce python code"]
  - source_model: cognitivecomputations/TinyDolphin-2.8.1-1.1b
    positive_prompts: ["What is 2 x 22?"]

💻 Usage

!pip install -qU transformers bitsandbytes accelerate

from transformers import AutoTokenizer
import transformers
import torch

model = "jtatman/TinyDolphin-3x-MoE"

tokenizer = AutoTokenizer.from_pretrained(model)
pipeline = transformers.pipeline(
    "text-generation",
    model=model,
    model_kwargs={"torch_dtype": torch.float16, "load_in_4bit": True},
)

messages = [{"role": "user", "content": "Explain what a Mixture of Experts is in less than 100 words."}]
prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = pipeline(prompt, max_new_tokens=256, do_sample=True, temperature=0.7, top_k=50, top_p=0.95)
print(outputs[0]["generated_text"])

Eval:

hf ({'pretrained': 'jtatman/TinyDolphin-3x-MoE'}), gen_kwargs: ({}), limit: None, num_fewshot: 0, batch_size: auto (64)

Tasks	Version	Filter	Metric		Value		Stderr
arc_challenge	1	none	acc	↑	0.3063	±	0.0135
		none	acc_norm	↑	0.3285	±	0.0137
arc_easy	1	none	acc	↑	0.5981	±	0.0101
		none	acc_norm	↑	0.5467	±	0.0102
hellaswag	1	none	acc	↑	0.4656	±	0.0050
		none	acc_norm	↑	0.6004	±	0.0049
openbookqa	1	none	acc	↑	0.2300	±	0.0188
		none	acc_norm	↑	0.3640	±	0.0215
piqa	1	none	acc	↑	0.7318	±	0.0103
		none	acc_norm	↑	0.7296	±	0.0104