35 lines
1.1 KiB
Markdown
35 lines
1.1 KiB
Markdown
---
|
|
license: bsd
|
|
datasets:
|
|
- vicgalle/alpaca-gpt4
|
|
language:
|
|
- fa
|
|
metrics:
|
|
- accuracy
|
|
pipeline_tag: text-generation
|
|
---
|
|
This model is Llama2-7B fine-tuned model which trained on farsi-wiki(approximately 180 milion token) and translated ALPACA dataset.
|
|
We extend tokenizer by 19954 new token with BPE algorithm running on persian dataset.
|
|
|
|
Use this code to run model on you input:
|
|
```
|
|
from transformers import AutoTokenizer, LlamaModelForCausalLM
|
|
|
|
model = LlamaModelForCausalLM.from_pretrained("mostafaamiri/persian_llama_7B_merged")
|
|
tokenizer = AutoTokenizer.from_pretrained(""mostafaamiri/persian_llama_7B_merged"")
|
|
|
|
instruction = "برای رفتن به کوهنوردی چه وسایلی را با خود ببرم؟"
|
|
prompt = [
|
|
"""Below is an instruction that describes a task.
|
|
Write a response that appropriately completes the request.\n\n
|
|
### Instruction:\n\n{instruction}\n\n\n### Response:\n\n\n"""
|
|
]
|
|
|
|
model.to("cuda")
|
|
generated_ids = model.generate(**tokenizer(prompt, return_tensors='pt').to("cuda"))
|
|
print(tokenizer.batch_decode(generated_ids)[0])
|
|
|
|
```
|
|
|
|
|