Instructions to use ApexDevelopment/tinygemma4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ApexDevelopment/tinygemma4 with Transformers:

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

pipe = pipeline("text-generation", model="ApexDevelopment/tinygemma4")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("ApexDevelopment/tinygemma4")
model = AutoModelForCausalLM.from_pretrained("ApexDevelopment/tinygemma4")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use ApexDevelopment/tinygemma4 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ApexDevelopment/tinygemma4"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "ApexDevelopment/tinygemma4",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/ApexDevelopment/tinygemma4

SGLang

How to use ApexDevelopment/tinygemma4 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 "ApexDevelopment/tinygemma4" \
    --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": "ApexDevelopment/tinygemma4",
		"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 "ApexDevelopment/tinygemma4" \
        --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": "ApexDevelopment/tinygemma4",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use ApexDevelopment/tinygemma4 with Docker Model Runner:
```
docker model run hf.co/ApexDevelopment/tinygemma4
```

tinygemma4 / config.json

ApexDevelopment

Replace checkpoint with TinyGemma4 v3

bb1514d verified 5 days ago

Raw

History Blame Contribute Delete

1.68 kB

	{
	"architectures": [
	"Gemma4ForCausalLM"
	],
	"attention_bias": false,
	"attention_dropout": 0.0,
	"attention_k_eq_v": false,
	"bos_token_id": 2,
	"dtype": "float32",
	"enable_moe_block": false,
	"eos_token_id": 1,
	"final_logit_softcapping": null,
	"global_head_dim": 32,
	"head_dim": 32,
	"hidden_activation": "gelu_pytorch_tanh",
	"hidden_size": 128,
	"hidden_size_per_layer_input": 16,
	"initializer_range": 0.02,
	"intermediate_size": 384,
	"layer_types": [
	"sliding_attention",
	"sliding_attention",
	"sliding_attention",
	"full_attention",
	"sliding_attention",
	"sliding_attention",
	"sliding_attention",
	"full_attention",
	"sliding_attention",
	"sliding_attention",
	"sliding_attention",
	"full_attention"
	],
	"max_position_embeddings": 2048,
	"model_type": "gemma4_text",
	"moe_intermediate_size": null,
	"num_attention_heads": 4,
	"num_experts": 0,
	"num_global_key_value_heads": 1,
	"num_hidden_layers": 12,
	"num_key_value_heads": 1,
	"num_kv_shared_layers": 0,
	"pad_token_id": 0,
	"rms_norm_eps": 1e-06,
	"rope_parameters": {
	"full_attention": {
	"partial_rotary_factor": 1.0,
	"rope_theta": 1000000.0,
	"rope_type": "default"
	},
	"sliding_attention": {
	"rope_theta": 10000.0,
	"rope_type": "default"
	}
	},
	"sliding_window": 128,
	"tie_word_embeddings": true,
	"top_k_experts": null,
	"transformers_version": "5.13.0",
	"use_bidirectional_attention": null,
	"use_cache": false,
	"use_double_wide_mlp": false,
	"vocab_size": 8192,
	"vocab_size_per_layer_input": 8192
	}