team-rexx-gemma-4-e4b

Instructions to use SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b with Transformers:

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

pipe = pipeline("image-text-to-text", model="SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b")
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)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoProcessor, AutoModelForMultimodalLM

processor = AutoProcessor.from_pretrained("SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b")
model = AutoModelForMultimodalLM.from_pretrained("SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b", device_map="auto")
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=256)
print(processor.decode(outputs[0][inputs["input_ids"].shape[-1]:]))

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b

SGLang

How to use SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b 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 "SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b" \
    --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": "SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

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 "SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b" \
        --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": "SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Docker Model Runner
How to use SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b with Docker Model Runner:
```
docker model run hf.co/SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b
```

Team Rexx: Gemma 4 E4B fine-tuned for SinhalaMMLU

This is the model behind Team Rexx's SinhalaMMLU shared-task submissions: the final greedy submission (Codabench ID 145, macro average 67.5666) and the earlier sampled submission 08 (ID 104, macro average 64.6025). It is loaded by SinhalaMMLU-SharedTask/team-rexx's inference.py; that repo's README covers how to run it, the training data and the closed-model disclosure.

Base model: google/gemma-4-E4B-it
Method: LoRA reasoning distillation on 20,870 correct teacher traces (rank 64, 2 epochs, learning rate 5e-5), merged into the weights, then W = W_base + 0.5 (W_distilled − W_base)
Parameters: 7,941,100,832, the full merged BF16 checkpoint. No adapter or base download is needed.
Files: identical to sinhala-mmlu/gemma-4-e4b-sinhala-mmlu at tag v2.0.0, our public release with the full model card, benchmark results and licence notes. SHA256SUMS is that release's checksum list; every file here except this README.md matches it.
Licence: Apache 2.0, following the Gemma 4 licence. See LICENSE, NOTICE and MODIFICATIONS.md.

Use

The recipe is in inference.py of the team repo: thinking on, a subject and grade line, 8 greedy answers over shuffled options with a majority vote, and a 7,168-token thinking budget, with vLLM 0.32.0 on one A100 40 GB. A minimal Transformers load:

import torch
from transformers import AutoModelForImageTextToText, AutoTokenizer

model_id = "SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForImageTextToText.from_pretrained(model_id, dtype=torch.bfloat16, device_map="auto").eval()

Downloads last month: 22

Safetensors

Model size

8B params

Tensor type

BF16

Model tree for SinhalaMMLU-SharedTask/team-rexx-gemma-4-e4b

Base model

google/gemma-4-E4B

Finetuned

google/gemma-4-E4B-it

Finetuned

(440)

this model