license: apache-2.0
pipeline_tag: text-generation
language:
- en
- he
tags:
- pretrained
inference:
parameters:
temperature: 0.7
Model Card for DictaLM-2.0
The DictaLM-2.0 Large Language Model (LLM) is a pretrained generative text model with 7 billion parameters trained to specialize in Hebrew text.
For full details of this model please read our release blog post.
This is the full-precision base model.
You can view and access the full collection of base/instruct unquantized/quantized versions of DictaLM-2.0 here.
Example Code
from transformers import pipeline
import torch
# This loads the model onto the GPU in bfloat16 precision
model = pipeline('text-generation', 'dicta-il/dictalm2.0', torch_dtype=torch.bfloat16, device_map='cuda')
# Sample few shot examples
prompt = """
注讘专: 讛诇讻转讬
注转讬讚: 讗诇讱
注讘专: 砖诪专转讬
注转讬讚: 讗砖诪讜专
注讘专: 砖诪注转讬
注转讬讚: 讗砖诪注
注讘专: 讛讘谞转讬
注转讬讚:
"""
print(model(prompt.strip(), do_sample=False, max_new_tokens=8, stop_sequence='\n'))
# [{'generated_text': '注讘专: 讛诇讻转讬\n注转讬讚: 讗诇讱\n\n注讘专: 砖诪专转讬\n注转讬讚: 讗砖诪讜专\n\n注讘专: 砖诪注转讬\n注转讬讚: 讗砖诪注\n\n注讘专: 讛讘谞转讬\n注转讬讚: 讗讘讬谉\n\n'}]
Example Code - 4-Bit
There are already pre-quantized 4-bit models using the GPTQ and AWQ methods available for use: DictaLM-2.0-AWQ and DictaLM-2.0-GPTQ.
For dynamic quantization on the go, here is sample code which loads the model onto the GPU using the bitsandbytes package, requiring :
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
model = AutoModelForCausalLM.from_pretrained('dicta-il/dictalm2.0', torch_dtype=torch.bfloat16, device_map='cuda', load_in_4bit=True)
tokenizer = AutoTokenizer.from_pretrained('dicta-il/dictalm2.0')
prompt = """
注讘专: 讛诇讻转讬
注转讬讚: 讗诇讱
注讘专: 砖诪专转讬
注转讬讚: 讗砖诪讜专
注讘专: 砖诪注转讬
注转讬讚: 讗砖诪注
注讘专: 讛讘谞转讬
注转讬讚:
"""
encoded = tokenizer(prompt.strip(), return_tensors='pt').to(model.device)
print(tokenizer.batch_decode(model.generate(**encoded, do_sample=False, max_new_tokens=4)))
# ['<s> 注讘专: 讛诇讻转讬\n注转讬讚: 讗诇讱\n\n注讘专: 砖诪专转讬\n注转讬讚: 讗砖诪讜专\n\n注讘专: 砖诪注转讬\n注转讬讚: 讗砖诪注\n\n注讘专: 讛讘谞转讬\n注转讬讚: 讗讘讬谉\n\n']
Model Architecture
DictaLM-2.0 is based on the Mistral-7B-v0.1 model with the following changes:
- An extended tokenizer with 1,000 injected tokens specifically for Hebrew, increasing the compression rate from 5.78 tokens/word to 2.76 tokens/word.
- Continued pretraining on over 190B tokens of naturally occuring text, 50% Hebrew and 50% English.
Notice
DictaLM 2.0 is a pretrained base model and therefore does not have any moderation mechanisms.
Citation
If you use this model, please cite:
[Will be added soon]