Instructions to use Weyaxi/Helion-4x34B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Weyaxi/Helion-4x34B with Transformers:

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

pipe = pipeline("text-generation", model="Weyaxi/Helion-4x34B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("Weyaxi/Helion-4x34B")
model = AutoModelForCausalLM.from_pretrained("Weyaxi/Helion-4x34B")
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 Weyaxi/Helion-4x34B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Weyaxi/Helion-4x34B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Weyaxi/Helion-4x34B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/Weyaxi/Helion-4x34B

SGLang

How to use Weyaxi/Helion-4x34B 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 "Weyaxi/Helion-4x34B" \
    --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": "Weyaxi/Helion-4x34B",
		"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 "Weyaxi/Helion-4x34B" \
        --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": "Weyaxi/Helion-4x34B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use Weyaxi/Helion-4x34B with Docker Model Runner:
```
docker model run hf.co/Weyaxi/Helion-4x34B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Helion-4x34B

This is the model for Helion-4x34B. I used this repo to make this MOE model.

Prompt Template(s):

Since bagel-dpo-34b-v0.2 uses many prompt templates, you can utilize prompt templates provided by bagel and other expert's prompt templates.

Note: I currently do not know which prompt template is best.

ChatML:

<|im_start|>system
{system}<|im_end|>
<|im_start|>user
{user}<|im_end|>
<|im_start|>assistant
{asistant}<|im_end|>

Human Asistant

Human: {user}

### Assistant: {asistant}

Alpaca (sort of)

Below is an instruction that describes a task.  Write a response that appropriately completes the request.

### Instruction:
{system}
{instruction}

### Response:

Vicuna

{system}
USER: {instruction}
ASSISTANT:

Visit bagel-dpo-34b-v0.2 to try more prompt templates.

Yaml Config to reproduce

base_model: nontoxic-bagel-34b-v0.2
gate_mode: hidden
dtype: bfloat16

experts:
  - source_model: bagel-dpo-34b-v0.2
    positive_prompts: ["question answering", "Q:", science", "biology", "chemistry", "physics"]
    negative_prompts: ["math", "reason", "mathematics", "solve", "count", "code", "python", "javascript", "programming", "algorithm"]

  - source_model: Nous-Hermes-2-Yi-34B
    positive_prompts: ["chat", "math", "reason", "mathematics", "solve", "count", "python", "javascript", "programming", "algorithm", "tell me", "assistant"]

  - source_model: SUS-Chat-34B
    positive_prompts: ["math", "reason", "mathematics", "solve", "count", "assistant"]

  - source_model: platypus-yi-34b
    positive_prompts: [""]
    negative_prompts: ["math", "reason", "mathematics", "solve", "count"]

Quantizationed versions

Quantizationed versions of this model is available thanks to TheBloke.

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	75.48
AI2 Reasoning Challenge (25-Shot)	69.71
HellaSwag (10-Shot)	85.28
MMLU (5-Shot)	77.33
TruthfulQA (0-shot)	63.91
Winogrande (5-shot)	84.37
GSM8k (5-shot)	72.25

If you would like to support me:

☕ Buy Me a Coffee

Downloads last month: 217

Safetensors

Model size

114B params

Tensor type

BF16

Model tree for Weyaxi/Helion-4x34B

Quantizations

2 models

Space using Weyaxi/Helion-4x34B 1

Evaluation results

normalized accuracy on AI2 Reasoning Challenge (25-Shot)
test set Open LLM Leaderboard

69.710
normalized accuracy on HellaSwag (10-Shot)
validation set Open LLM Leaderboard

85.280
accuracy on MMLU (5-Shot)
test set Open LLM Leaderboard

77.330
mc2 on TruthfulQA (0-shot)
validation set Open LLM Leaderboard

63.910
accuracy on Winogrande (5-shot)
validation set Open LLM Leaderboard

84.370
accuracy on GSM8k (5-shot)
test set Open LLM Leaderboard

72.250