transformers/docs/source/en/model_doc/open-llama.md

This model was released on {release_date} and added to Hugging Face Transformers on 2023-06-20 and contributed by s-JoL.

Warning

This model is in maintenance mode only, we don’t accept any new PRs changing its code.

If you run into any issues running this model, please reinstall the last version that supported this model: v4.31.0. You can do so by running the following command: pip install -U transformers==4.31.0.

Open-Llama

Open-Llama is an open-source framework for training large language models, covering the full pipeline from dataset preparation and tokenization to pre-training, instruction tuning, LoRA fine-tuning, and RLHF. It supports HuggingFace Transformers and achieves high performance, reaching 89% of GPT-3.5 on Chinese benchmarks, with a training speed of 3620 tokens/s, surpassing the original LLaMA. The project provides both pre-trained and instruction-tuned checkpoints, trained on 330 billion tokens with a global batch size of 4 million, and includes multi-turn dialogue, programming, and mathematical abilities. Evaluation follows FastChat methods, and the codebase is designed for high-performance experimentation and deployment.

import torch
from transformers import pipeline

pipeline = pipeline(task="text-generation", model="openlm-research/open_llama_7b", dtype="auto",)
pipeline("Plants create energy through a process known as photosynthesis.")

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

tokenizer = AutoTokenizer.from_pretrained("openlm-research/open_llama_7b")
model = AutoModelForCausalLM.from_pretrained("openlm-research/open_llama_7b", dtype="auto",)

inputs = tokenizer("Plants create energy through a process known as photosynthesis.", return_tensors="pt")
outputs = model.generate(**inputs, max_length=50)
print(tokenizer.decode(outputs[0]))

OpenLlamaConfig

autodoc OpenLlamaConfig

OpenLlamaModel

autodoc OpenLlamaModel - forward

OpenLlamaForCausalLM

autodoc OpenLlamaForCausalLM - forward

OpenLlamaForSequenceClassification

autodoc OpenLlamaForSequenceClassification - forward