Instructions to use mediocredev/open-llama-3b-v2-instruct with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use mediocredev/open-llama-3b-v2-instruct with Transformers:

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

pipe = pipeline("text-generation", model="mediocredev/open-llama-3b-v2-instruct")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("mediocredev/open-llama-3b-v2-instruct")
model = AutoModelForCausalLM.from_pretrained("mediocredev/open-llama-3b-v2-instruct")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use mediocredev/open-llama-3b-v2-instruct with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "mediocredev/open-llama-3b-v2-instruct"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mediocredev/open-llama-3b-v2-instruct",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/mediocredev/open-llama-3b-v2-instruct

SGLang

How to use mediocredev/open-llama-3b-v2-instruct 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 "mediocredev/open-llama-3b-v2-instruct" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mediocredev/open-llama-3b-v2-instruct",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

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 "mediocredev/open-llama-3b-v2-instruct" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "mediocredev/open-llama-3b-v2-instruct",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use mediocredev/open-llama-3b-v2-instruct with Docker Model Runner:
```
docker model run hf.co/mediocredev/open-llama-3b-v2-instruct
```
Browse Quantizations to use this model in llama.cpp, Ollama, LM Studio, or any compatible app.

Prerequisites

In addition to pytorch and transformers, install required packages:

pip install accelerate sentencepiece

Usage

To use, copy the following script:

from transformers import AutoTokenizer, AutoModelForCausalLM
import transformers
import torch
model_id = 'mediocredev/open-llama-3b-v2-instruct'
tokenizer_id = 'mediocredev/open-llama-3b-v2-instruct'
tokenizer = AutoTokenizer.from_pretrained(tokenizer_id)

pipeline = transformers.pipeline(
  'text-generation',
  model=model_id,
  tokenizer=tokenizer,
  torch_dtype=torch.bfloat16,
  device_map='auto',
)

system_message = 'You are a helpful assistant, who always provide explanation.'
user_message = 'How many days are there in a leap year?'

prompt = f'### System:\n{system_message}<|endoftext|>\n### User:\n{user_message}<|endoftext|>\n### Assistant:\n'
response = pipeline(
   prompt,
   max_length=1000,
   repetition_penalty=1.05,
)
response = response[0]['generated_text']
print(response)

# Assistant: A leap year has 366 days. It's an extra day added to the calendar every four years to account for the extra time it takes for Earth to complete one full orbit around the Sun.

Limitations

mediocredev/open-llama-3b-v2-instruct is based on LLaMA 3B v2. It can struggle with factual accuracy, particularly when presented with conflicting information or nuanced topics. Its outputs are not deterministic and require critical evaluation to avoid relying solely on its assertions. Additionally, its generative capabilities, while promising, can sometimes produce factually incorrect or offensive content, necessitating careful curation and human oversight. As an evolving model, LLaMA is still under development, and its limitations in areas like bias mitigation and interpretability are being actively addressed. By using this model responsibly and being aware of its shortcomings, we can unlock its potential while mitigating its risks.

Contact

Welcome any feedback, questions, and discussions. Feel free to reach out: mediocredev@outlook.com

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	42.02
AI2 Reasoning Challenge (25-Shot)	38.48
HellaSwag (10-Shot)	70.24
MMLU (5-Shot)	39.69
TruthfulQA (0-shot)	37.96
Winogrande (5-shot)	65.75
GSM8k (5-shot)	0.00

Downloads last month: 207

Safetensors

Model size

3B params

Tensor type

F16

Model tree for mediocredev/open-llama-3b-v2-instruct

Quantizations

3 models

Evaluation results

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

38.480
normalized accuracy on HellaSwag (10-Shot)
validation set Open LLM Leaderboard

70.240
accuracy on MMLU (5-Shot)
test set Open LLM Leaderboard

39.690
mc2 on TruthfulQA (0-shot)
validation set Open LLM Leaderboard

37.960
accuracy on Winogrande (5-shot)
validation set Open LLM Leaderboard

65.750
accuracy on GSM8k (5-shot)
test set Open LLM Leaderboard

0.000