Instructions to use wxzhang/dpo-selective-buffer-spo-shift with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use wxzhang/dpo-selective-buffer-spo-shift with Transformers:

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

pipe = pipeline("text-generation", model="wxzhang/dpo-selective-buffer-spo-shift")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("wxzhang/dpo-selective-buffer-spo-shift")
model = AutoModelForCausalLM.from_pretrained("wxzhang/dpo-selective-buffer-spo-shift")
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]:]))

Inference
Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use wxzhang/dpo-selective-buffer-spo-shift with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/wxzhang/dpo-selective-buffer-spo-shift

SGLang

How to use wxzhang/dpo-selective-buffer-spo-shift 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 "wxzhang/dpo-selective-buffer-spo-shift" \
    --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": "wxzhang/dpo-selective-buffer-spo-shift",
		"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 "wxzhang/dpo-selective-buffer-spo-shift" \
        --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": "wxzhang/dpo-selective-buffer-spo-shift",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use wxzhang/dpo-selective-buffer-spo-shift with Docker Model Runner:
```
docker model run hf.co/wxzhang/dpo-selective-buffer-spo-shift
```

dpo-selective-buffer-spo-shift

This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:

Loss: 0.6777
Rewards/chosen: -0.1371
Rewards/rejected: -0.0830
Rewards/accuracies: 0.4693
Rewards/margins: -0.0541
Rewards/safe Rewards: -0.1332
Rewards/unsafe Rewards: -0.1263
Logps/rejected: -92.4348
Logps/chosen: -131.0029
Logits/rejected: -1.8308
Logits/chosen: -2.0825

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-07
train_batch_size: 2
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 2
gradient_accumulation_steps: 8
total_train_batch_size: 32
total_eval_batch_size: 16
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Rewards/chosen	Rewards/rejected	Rewards/accuracies	Rewards/margins	Rewards/safe Rewards	Rewards/unsafe Rewards	Logps/rejected	Logps/chosen	Logits/rejected	Logits/chosen
131.6857	0.27	500	0.8894	-0.1023	-0.0129	0.4546	-0.0893	-0.1043	-0.1017	-92.3648	-130.9681	-1.8032	-2.0565
34.7958	0.54	1000	0.7397	-0.1263	-0.1290	0.5028	0.0026	-0.1237	-0.1264	-92.4809	-130.9922	-1.7990	-2.0551
15.9924	0.81	1500	0.6823	-0.1578	-0.1077	0.4713	-0.0501	-0.1557	-0.1535	-92.4596	-131.0237	-1.8335	-2.0849

Framework versions

Transformers 4.36.2
Pytorch 2.1.2
Datasets 2.14.6
Tokenizers 0.15.2

Downloads last month: 2

Safetensors

Model size

7B params

Tensor type

BF16