Instructions to use microsoft/FrogMini-14B-2510 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use microsoft/FrogMini-14B-2510 with Transformers:

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

pipe = pipeline("text-generation", model="microsoft/FrogMini-14B-2510")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("microsoft/FrogMini-14B-2510")
model = AutoModelForCausalLM.from_pretrained("microsoft/FrogMini-14B-2510")
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

vLLM

How to use microsoft/FrogMini-14B-2510 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "microsoft/FrogMini-14B-2510"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "microsoft/FrogMini-14B-2510",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/microsoft/FrogMini-14B-2510

SGLang

How to use microsoft/FrogMini-14B-2510 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 "microsoft/FrogMini-14B-2510" \
    --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": "microsoft/FrogMini-14B-2510",
		"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 "microsoft/FrogMini-14B-2510" \
        --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": "microsoft/FrogMini-14B-2510",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use microsoft/FrogMini-14B-2510 with Docker Model Runner:
```
docker model run hf.co/microsoft/FrogMini-14B-2510
```

Language specific performance

by sfrav - opened Jan 14

Discussion

sfrav

Jan 14

Is the training data for this Python heavy? Is there any indication of performance on lower level languages like cpp and rust?
What debugging usecases is this best applied to (and what's out if scope)?

MarcCote

Microsoft org Jan 15

Hi, thanks for the comment. It is indeed Python heavy and our evaluation results were obtained on SWE-Bench-verified (also just in Python). That said, we do believe the technique (see https://microsoft.github.io/debug-gym/blog/2025/10/bug-pilot) used to generate the debugging trajectories could be applied to other programming languages for further finetuning.

Since we use R2EGym scaffolding and SWE-Smith to generate synthetic bugs, the most suitable debugging cases are those that share similarity with the whole SWE-bench pipeline (i.e., given a codebase + Github-like issue statement, the agent will produce a reproducing script and a code patch to fix the issue). Keep in mind, this is a model with many limitations (as stated in the model card) intended for research purposes.

MarcCote changed discussion status to closed Jan 15

Upload images, audio, and videos by dragging in the text input, pasting, or clicking here.

Tap or paste here to upload images

· Sign up or log in to comment