Instructions to use MaziyarPanahi/calme-3.1-instruct-78b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use MaziyarPanahi/calme-3.1-instruct-78b with Transformers:

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

pipe = pipeline("text-generation", model="MaziyarPanahi/calme-3.1-instruct-78b")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("MaziyarPanahi/calme-3.1-instruct-78b")
model = AutoModelForCausalLM.from_pretrained("MaziyarPanahi/calme-3.1-instruct-78b")
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 MaziyarPanahi/calme-3.1-instruct-78b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "MaziyarPanahi/calme-3.1-instruct-78b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "MaziyarPanahi/calme-3.1-instruct-78b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/MaziyarPanahi/calme-3.1-instruct-78b

SGLang

How to use MaziyarPanahi/calme-3.1-instruct-78b 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 "MaziyarPanahi/calme-3.1-instruct-78b" \
    --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": "MaziyarPanahi/calme-3.1-instruct-78b",
		"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 "MaziyarPanahi/calme-3.1-instruct-78b" \
        --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": "MaziyarPanahi/calme-3.1-instruct-78b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use MaziyarPanahi/calme-3.1-instruct-78b with Docker Model Runner:
```
docker model run hf.co/MaziyarPanahi/calme-3.1-instruct-78b
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

This is an experimental model, so it might not perform well for some prompts and may be sensitive to hyper parameters. I would appreciate any feedback to see if I can fix any issues in the next iteration. ❤️

MaziyarPanahi/calme-3.1-instruct-78b

This model is an advanced iteration of the powerful Qwen/Qwen2.5-72B, specifically fine-tuned to enhance its capabilities in generic domains. The Qwen2.5-72B base model was merged with itself to create a larger model. After that, the model was fine-tuned on a custom datasets.

⚡ Quantized GGUF

Thanks to mradermacher: calme-3.1-instruct-78b-GGUF

🏆 Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	51.20
IFEval (0-Shot)	81.36
BBH (3-Shot)	62.41
MATH Lvl 5 (4-Shot)	38.75
GPQA (0-shot)	19.46
MuSR (0-shot)	36.50
MMLU-PRO (5-shot)	68.72

Prompt Template

This model uses ChatML prompt template:

<|im_start|>system
{System}
<|im_end|>
<|im_start|>user
{User}
<|im_end|>
<|im_start|>assistant
{Assistant}

How to use


# Use a pipeline as a high-level helper

from transformers import pipeline

messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe = pipeline("text-generation", model="MaziyarPanahi/calme-3.1-instruct-78b")
pipe(messages)


# Load model directly

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("MaziyarPanahi/calme-3.1-instruct-78b")
model = AutoModelForCausalLM.from_pretrained("MaziyarPanahi/calme-3.1-instruct-78b")

Ethical Considerations

As with any large language model, users should be aware of potential biases and limitations. We recommend implementing appropriate safeguards and human oversight when deploying this model in production environments.

Downloads last month: 54

Safetensors

Model size

78B params

Tensor type

F32

Model tree for MaziyarPanahi/calme-3.1-instruct-78b

Merges

2 models

Quantizations

3 models

Collection including MaziyarPanahi/calme-3.1-instruct-78b

🇫🇷 Calme-3

Collection

Here you can find all the new Calme-3 models • 27 items • Updated Feb 9, 2025 • 18

Evaluation results

strict accuracy on IFEval (0-Shot)
Open LLM Leaderboard

81.360
normalized accuracy on BBH (3-Shot)
Open LLM Leaderboard

62.410
exact match on MATH Lvl 5 (4-Shot)
Open LLM Leaderboard

38.750
acc_norm on GPQA (0-shot)
Open LLM Leaderboard

19.460
acc_norm on MuSR (0-shot)
Open LLM Leaderboard

36.500
accuracy on MMLU-PRO (5-shot)
test set Open LLM Leaderboard

68.720