Instructions to use RedHatAI/OpenHermes-2.5-Mistral-7B-marlin with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use RedHatAI/OpenHermes-2.5-Mistral-7B-marlin with Transformers:

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

pipe = pipeline("text-generation", model="RedHatAI/OpenHermes-2.5-Mistral-7B-marlin")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("RedHatAI/OpenHermes-2.5-Mistral-7B-marlin")
model = AutoModelForCausalLM.from_pretrained("RedHatAI/OpenHermes-2.5-Mistral-7B-marlin")
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 RedHatAI/OpenHermes-2.5-Mistral-7B-marlin with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "RedHatAI/OpenHermes-2.5-Mistral-7B-marlin"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "RedHatAI/OpenHermes-2.5-Mistral-7B-marlin",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/RedHatAI/OpenHermes-2.5-Mistral-7B-marlin

SGLang

How to use RedHatAI/OpenHermes-2.5-Mistral-7B-marlin 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 "RedHatAI/OpenHermes-2.5-Mistral-7B-marlin" \
    --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": "RedHatAI/OpenHermes-2.5-Mistral-7B-marlin",
		"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 "RedHatAI/OpenHermes-2.5-Mistral-7B-marlin" \
        --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": "RedHatAI/OpenHermes-2.5-Mistral-7B-marlin",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use RedHatAI/OpenHermes-2.5-Mistral-7B-marlin with Docker Model Runner:
```
docker model run hf.co/RedHatAI/OpenHermes-2.5-Mistral-7B-marlin
```

OpenHermes-2.5-Mistral-7B-marlin / quantization /apply_gptq_save_marlin.py

robertgshaw2

Create quantization/apply_gptq_save_marlin.py

f57a0ec verified about 2 years ago

raw

history blame contribute delete

2.2 kB

	import argparse, gc, shutil
	from transformers import AutoTokenizer
	from auto_gptq import AutoGPTQForCausalLM, BaseQuantizeConfig
	from datasets import load_dataset

	parser = argparse.ArgumentParser()
	parser.add_argument("--model-id", type=str)
	parser.add_argument("--save-dir", type=str)
	parser.add_argument("--channelwise", action="store_true")
	parser.add_argument("--num-samples", type=int, default=512)
	parser.add_argument("--max-seq-len", type=int, default=2048)


	def preprocess(example):
	return {"text": tokenizer.apply_chat_template(example["messages"], tokenize=False)}

	if __name__ == "__main__":
	args = parser.parse_args()

	dataset = load_dataset("HuggingFaceH4/ultrachat_200k", split="train_sft[:5%]")
	tokenizer = AutoTokenizer.from_pretrained(args.model_id)
	ds = dataset.shuffle().select(range(args.num_samples))
	ds = ds.map(preprocess)

	examples = [
	tokenizer(
	example["text"], padding=False, max_length=args.max_seq_len, truncation=True,
	) for example in ds
	]

	if args.channelwise:
	group_size = -1
	else:
	group_size = 128

	quantize_config = BaseQuantizeConfig(
	bits=4, # Only support 4 bit
	group_size=group_size, # Set to g=128 or -1 (for channelwise)
	desc_act=False, # Marlin does not suport act_order=True
	model_file_base_name="model" # Name of the model.safetensors when we call save_pretrained
	)

	model = AutoGPTQForCausalLM.from_pretrained(
	args.model_id,
	quantize_config,
	device_map="auto")
	model.quantize(examples)

	gptq_save_dir = "./tmp-gptq"
	print(f"Saving gptq model to {gptq_save_dir}")
	model.save_pretrained(gptq_save_dir)
	tokenizer.save_pretrained(gptq_save_dir)

	del model
	gc.collect()

	print("Reloading in marlin format")

	marlin_model = AutoGPTQForCausalLM.from_quantized(
	gptq_save_dir,
	use_marlin=True,
	device_map="auto")

	print("Saving in marlin format")
	marlin_model.save_pretrained(args.save_dir)
	tokenizer.save_pretrained(args.save_dir)

	shutil.rmtree(gptq_save_dir)