Instructions to use Grogros/phi-2-OurSafecoder with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Grogros/phi-2-OurSafecoder with Transformers:

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

pipe = pipeline("text-generation", model="Grogros/phi-2-OurSafecoder")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Grogros/phi-2-OurSafecoder")
model = AutoModelForCausalLM.from_pretrained("Grogros/phi-2-OurSafecoder")
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 Grogros/phi-2-OurSafecoder with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Grogros/phi-2-OurSafecoder"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Grogros/phi-2-OurSafecoder",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Grogros/phi-2-OurSafecoder

SGLang

How to use Grogros/phi-2-OurSafecoder 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 "Grogros/phi-2-OurSafecoder" \
    --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": "Grogros/phi-2-OurSafecoder",
		"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 "Grogros/phi-2-OurSafecoder" \
        --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": "Grogros/phi-2-OurSafecoder",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Grogros/phi-2-OurSafecoder with Docker Model Runner:
```
docker model run hf.co/Grogros/phi-2-OurSafecoder
```

Grogros commited on May 9, 2025

Commit

9548740

verified ·

1 Parent(s): 6413237

Upload finetuning_config.yaml with huggingface_hub

Browse files

Files changed (1) hide show

finetuning_config.yaml +57 -0

finetuning_config.yaml ADDED Viewed

	@@ -0,0 +1,57 @@

+attn_implementation: sdpa
+backdoor_dataset: !!python/object/apply:src.data.dataset.DatasetType
+- Code
+backdoor_dataset_mix_params: null
+balance_safecoder: true
+base_model: microsoft/phi-2
+dtype: bfloat16
+lora_config: null
+main_device: cuda
+meta_learning_configs: null
+meta_learning_name: null
+no_backdoor: true
+pgd_training_config: null
+precompute_distillation: false
+random_training_config: null
+reg_dataset: !!python/object/apply:src.data.dataset.DatasetType
+- SecretSauce
+reg_dataset_mix_params:
+  ? !!python/object/apply:src.data.dataset.DatasetType
+  - AlpacaGPT4
+  : 0.2
+  ? !!python/object/apply:src.data.dataset.DatasetType
+  - CodeAlpaca
+  : 0.6
+  ? !!python/object/apply:src.data.dataset.DatasetType
+  - SecInsec
+  : 0.2
+reg_device: cuda
+reg_lambda: 1.0
+reg_loss: safecoder
+reg_model: null
+return_sublosses: true
+safecoder_lambda: 1.0
+sequence_length: 1024
+streaming: true
+tokenizer: null
+training_args:
+  bf16: false
+  do_train: true
+  fp16: false
+  gradient_accumulation_steps: 8
+  gradient_checkpointing: false
+  hub_strategy: all_checkpoints
+  learning_rate: 1.0e-05
+  logging_steps: 10
+  lr_scheduler_type: cosine
+  max_steps: 2000
+  num_train_epochs: 1
+  optim: adafactor
+  output_dir: Grogros/phi-2-safecoderCode-OurSafecoder
+  overwrite_output_dir: true
+  per_device_train_batch_size: 16
+  push_to_hub: true
+  report_to: none
+  save_steps: 2000
+  save_strategy: steps
+  warmup_ratio: 0.1