Instructions to use W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337 with Transformers:

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

pipe = pipeline("text-generation", model="W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337")
model = AutoModelForCausalLM.from_pretrained("W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337", device_map="auto")
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 Settings

vLLM

How to use W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337 with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337

SGLang

How to use W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337 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 "W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337" \
    --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": "W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337",
		"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 "W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337" \
        --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": "W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337 with Docker Model Runner:
```
docker model run hf.co/W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337
```

llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337

This model is a fine-tuned version of llama-3-8b-base-sft-hh-harmless-4xh200-batch-64 on the Anthropic/hh-rlhf dataset. It achieves the following results on the evaluation set:

Loss: 0.5256
Margin Dpo/beta: 0.1000
Margin Dpo/loss Margin Mean: 9.3760
Margin Dpo/beta Margin Mean: 0.9376
Margin Dpo/beta Margin Std: 1.4808
Margin Dpo/beta Margin Grad Mean: -0.3434
Margin Dpo/beta Margin Grad Std: 0.2353
Margin Dpo/margin Mean: 9.3760
Margin Dpo/margin Std: 14.8082
Logps/chosen: -92.1130
Logps/rejected: -106.1785
Logps/ref Chosen: -74.8595
Logps/ref Rejected: -79.5490
Logits/chosen: 0.3868
Logits/rejected: 0.3352

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: 8
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 4
gradient_accumulation_steps: 2
total_train_batch_size: 64
total_eval_batch_size: 32
optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: cosine
lr_scheduler_warmup_ratio: 0.1
num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Margin Dpo/beta	Margin Dpo/loss Margin Mean	Margin Dpo/beta Margin Mean	Margin Dpo/beta Margin Std	Margin Dpo/beta Margin Grad Mean	Margin Dpo/beta Margin Grad Std	Margin Dpo/margin Mean	Margin Dpo/margin Std	Logps/chosen	Logps/rejected	Logps/ref Chosen	Logps/ref Rejected	Logits/chosen	Logits/rejected
1.3344	0.1512	100	0.6553	0.1000	1.4293	0.1429	0.4977	-0.4668	0.1122	1.4293	4.9768	-79.7080	-85.8269	-74.8595	-79.5490	0.2576	0.2201
0.9172	0.3023	200	0.5443	0.1000	7.4892	0.7489	1.2567	-0.3638	0.2218	7.4892	12.5667	-86.6174	-98.7961	-74.8595	-79.5490	0.3528	0.3042
0.9647	0.4535	300	0.5349	0.1000	9.3620	0.9362	1.4964	-0.3434	0.2418	9.3620	14.9643	-93.1728	-107.2244	-74.8595	-79.5490	0.4092	0.3577
1.0864	0.6047	400	0.5335	0.1000	8.8675	0.8867	1.4359	-0.3489	0.2335	8.8675	14.3590	-91.3162	-104.8732	-74.8595	-79.5490	0.3888	0.3372
1.0094	0.7559	500	0.5270	0.1000	9.5291	0.9529	1.5070	-0.3422	0.2379	9.5291	15.0704	-92.2392	-106.4577	-74.8595	-79.5490	0.3944	0.3425
1.0508	0.9070	600	0.5256	0.1000	9.3760	0.9376	1.4808	-0.3434	0.2353	9.3760	14.8082	-92.1130	-106.1785	-74.8595	-79.5490	0.3868	0.3352

Framework versions

Transformers 4.51.0
Pytorch 2.3.1+cu121
Datasets 2.21.0
Tokenizers 0.21.4

Downloads last month: 10

Safetensors

Model size

8B params

Tensor type

F32

W-61
/

llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337

llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

Dataset used to train W-61/llama-3-8b-base-margin-dpo-hh-harmless-4xh200-batch-64-20260417-222337