Hebrew Mistral 7B

🚀 Hebrew-Mistral-7B

Hebrew-Mistral-7B is an open - source Large Language Model (LLM). It is pretrained in both Hebrew and English, with 7 billion parameters, and is based on Mistral-7B-v1.0 from Mistral.

This model comes with an extended Hebrew tokenizer that has 64,000 tokens. It is continuously pretrained from Mistral-7B on tokens in both English and Hebrew. The resulting model is a powerful general - purpose language model, well - suited for a wide range of natural language processing tasks, with a particular focus on Hebrew language understanding and generation.

🚀 Quick Start

Below are some code snippets to help you quickly start running the model. First, make sure to pip install -U transformers, then copy the relevant snippet according to your use case.

💻 Usage Examples

Basic Usage

# Running on CPU
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("yam-peleg/Hebrew-Mistral-7B")
model = AutoModelForCausalLM.from_pretrained("yam-peleg/Hebrew-Mistral-7B")

input_text = "שלום! מה שלומך היום?"
input_ids = tokenizer(input_text, return_tensors="pt")

outputs = model.generate(**input_ids)
print(tokenizer.decode(outputs[0]))

Advanced Usage

Running on GPU

from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("yam-peleg/Hebrew-Mistral-7B")
model = AutoModelForCausalLM.from_pretrained("yam-peleg/Hebrew-Mistral-7B", device_map="auto")

input_text = "שלום! מה שלומך היום?"
input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")

outputs = model.generate(**input_ids)
print(tokenizer.decode(outputs[0]))

Running with 4 - Bit precision

from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig

tokenizer = AutoTokenizer.from_pretrained("yam-peleg/Hebrew-Mistral-7B")
model = AutoModelForCausalLM.from_pretrained("yam-peleg/Hebrew-Mistral-7B", quantization_config = BitsAndBytesConfig(load_in_4bit=True))

input_text = "שלום! מה שלומך היום?"
input_ids = tokenizer(input_text, return_tensors="pt").to("cuda")

outputs = model.generate(**input_ids)
print(tokenizer.decode(outputs[0]))

📄 License

The project uses the Apache-2.0 license.

🔧 Technical Details

Hebrew-Mistral-7B is a pretrained base model and therefore does not have any moderation mechanisms.

📚 Documentation

Authors

• Trained by Yam Peleg.
• In collaboration with Jonathan Rouach and Arjeo, inc.