Instructions to use vicgalle/CarbonBeagle-11B with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use vicgalle/CarbonBeagle-11B with Transformers:

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

pipe = pipeline("text-generation", model="vicgalle/CarbonBeagle-11B")
messages = [
    {"role": "user", "content": "Who are you?"},
]
pipe(messages)

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("vicgalle/CarbonBeagle-11B")
model = AutoModelForCausalLM.from_pretrained("vicgalle/CarbonBeagle-11B")
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

vLLM

How to use vicgalle/CarbonBeagle-11B with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "vicgalle/CarbonBeagle-11B"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "vicgalle/CarbonBeagle-11B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/vicgalle/CarbonBeagle-11B

SGLang

How to use vicgalle/CarbonBeagle-11B 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 "vicgalle/CarbonBeagle-11B" \
    --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": "vicgalle/CarbonBeagle-11B",
		"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 "vicgalle/CarbonBeagle-11B" \
        --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": "vicgalle/CarbonBeagle-11B",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Docker Model Runner
How to use vicgalle/CarbonBeagle-11B with Docker Model Runner:
```
docker model run hf.co/vicgalle/CarbonBeagle-11B
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

CarbonBeagle-11B

An experiment in merging models of different architectures and sizes. Here are the steps:

Upscale mlabonne/NeuralBeagle14-7B to vicgalle/franken-Beagle-11B.
DPO-tune vicgalle/franken-Beagle-11B to vicgalle/NeuralBeagle-11B.
Merge vicgalle/NeuralBeagle-11B and jeonsworld/CarbonVillain-en-10.7B-v4.

Merge Details

Merge Method

This model was merged using the linear merge method.

Models Merged

The following models were included in the merge:

Configuration

The following YAML configuration was used to produce this model:

models:
    - model: jeonsworld/CarbonVillain-en-10.7B-v4
      parameters:
        weight: 1.0
    - model: vicgalle/NeuralBeagle-11B
      parameters:
        weight: 0.5
merge_method: linear

dtype: float16

Evaluations

At the time of its creation (21-01-2024), it is the best model in the Open LLM Leaderboard for its size class (10.7B-11B), and also 13B models:

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	74.64
AI2 Reasoning Challenge (25-Shot)	71.84
HellaSwag (10-Shot)	88.93
MMLU (5-Shot)	66.62
TruthfulQA (0-shot)	69.43
Winogrande (5-shot)	84.06
GSM8k (5-shot)	66.94

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	22.36
IFEval (0-Shot)	54.15
BBH (3-Shot)	33.06
MATH Lvl 5 (4-Shot)	5.51
GPQA (0-shot)	6.94
MuSR (0-shot)	9.19
MMLU-PRO (5-shot)	25.29

Notice

Part of the model is derived from CC-BY-NC-SA-licensed work
Hence, the model cannot be used for commercial purposes
And redistribution must be under the same license (CC-BY-NC-SA 4.0)

Downloads last month: 8,188

Safetensors

Model size

11B params

Tensor type

F16

Model tree for vicgalle/CarbonBeagle-11B

jeonsworld/CarbonVillain-en-10.7B-v4

vicgalle/NeuralBeagle-11B

Merge model

this model

Merges

1 model

Quantizations

3 models

Space using vicgalle/CarbonBeagle-11B 1

Collection including vicgalle/CarbonBeagle-11B

Exotic Frankenmerges 🥨

Collection

Merges of models of different architectures and sizes that end up working surprisingly well • 1 item • Updated Jun 13, 2024 • 1

Paper for vicgalle/CarbonBeagle-11B

Model soups: averaging weights of multiple fine-tuned models improves accuracy without increasing inference time

Paper • 2203.05482 • Published Mar 10, 2022 • 8

Evaluation results

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

71.840
normalized accuracy on HellaSwag (10-Shot)
validation set Open LLM Leaderboard

88.930
accuracy on MMLU (5-Shot)
test set Open LLM Leaderboard

66.620
mc2 on TruthfulQA (0-shot)
validation set Open LLM Leaderboard

69.430
accuracy on Winogrande (5-shot)
validation set Open LLM Leaderboard

84.060
accuracy on GSM8k (5-shot)
test set Open LLM Leaderboard

66.940
strict accuracy on IFEval (0-Shot)
Open LLM Leaderboard

54.150
normalized accuracy on BBH (3-Shot)
Open LLM Leaderboard

33.060
exact match on MATH Lvl 5 (4-Shot)
Open LLM Leaderboard

5.510
acc_norm on GPQA (0-shot)
Open LLM Leaderboard

6.940
acc_norm on MuSR (0-shot)
Open LLM Leaderboard

9.190
accuracy on MMLU-PRO (5-shot)
test set Open LLM Leaderboard

25.290