Downloads · 30 days
0
llm-jp/llm-jp-13b-dpo-lora-hh_rlhf_ja-v1.1
llm-jp-13b-dpo-lora-hh_rlhf_ja-v1.1 is a text generation model from llm-jp. Use it when you need the model to write or continue text. It is set up for transformers. The card lists the license as apache-2.0.
This repository provides large language models developed by LLM-jp, a collaborative project launched in Japan.
Downloads · 30 days
0
Access
Public
Updated Mar 12, 2024
Repo size
839 MB
Likes
1
Public
Click a slice to open those files.
.safetensors839 MB · 100%
From the Hugging Face model README
This repository provides large language models developed by LLM-jp, a collaborative project launched in Japan.
This model was trained on hh-rlhf-12k-ja via Direct Preference Optimization from llm-jp-13b-instruct-full-dolly_en-dolly_ja-ichikara_003_001-oasst_en-oasst_ja-v1.1.
| Pre-trained models |
| llm-jp-13b-v1.0 |
| llm-jp-1.3b-v1.0 |
| Checkpoints format: Hugging Face Transformers (Megatron-DeepSpeed format models are available here) |
import torch
from transformers import AutoTokenizer
from peft import AutoPeftModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("llm-jp/llm-jp-13b-dpo-lora-hh_rlhf_ja-v1.1")
model = AutoPeftModelForCausalLM.from_pretrained("llm-jp/llm-jp-13b-dpo-lora-hh_rlhf_ja-v1.1", device_map="auto", torch_dtype=torch.float16)
text = "以下は、タスクを説明する指示です。要求を適切に満たす応答を書きなさい。\n\n### 指示:\n{instruction}\n\n### 応答:\n".format(instruction="自然言語処理とは何か")
tokenized_input = tokenizer.encode(text, add_special_tokens=False, return_tensors="pt").to(model.device)
with torch.no_grad():
output = model.generate(
tokenized_input,
max_new_tokens=512,
do_sample=True,
top_p=0.95,
temperature=0.7,
repetition_penalty=1.1,
)[0]
print(tokenizer.decode(output))
| Model | Params | Layers | Hidden size | Heads | Context length |
|---|---|---|---|---|---|
| 13b model | 13b | 40 | 5120 | 40 | 2048 |
| 1.3b model | 1.3b | 24 | 2048 | 16 | 2048 |
Pre-training:
Instruction tuning:
The tokenizer of this model is based on huggingface/tokenizers Unigram byte-fallback model.
The vocabulary entries were converted from llm-jp-tokenizer v2.1 (50k).
Please refer to README.md of llm-ja-tokenizer for details on the vocabulary construction procedure.
tokenizers>=0.14.0The models have been pre-trained using a blend of the following datasets.
The pre-training was continuously conducted using a total of 10 folds of non-overlapping data, each consisting of approximately 27-28B tokens. We finalized the pre-training with additional (potentially) high-quality 27B tokens data obtained from the identical source datasets listed above used for the 10-fold data.
The models have been fine-tuned on the following datasets.
| Language | Dataset | description |
|---|---|---|
| Japanese | jaster | An automatically transformed data from the existing Japanese NLP datasets |
| English | databricks-dolly-15k | - |
| Japanese | databricks-dolly-15k-ja | A translated one by DeepL in LLM-jp |
| English | oasst1-21k-en | English subset of oasst1 dataset |
| Japanese | oasst1-21k-ja | A translated one by DeepL in LLM-jp |
| Japanese | ichikara_003_001 | ichikara-instruction dataset (ver.003-001) |
| Japanese | hh-rlhf-12k-ja | A translated one by DeepL in LLM-jp |
You can view the evaluation results of several LLMs on this leaderboard. We used llm-jp-eval for the evaluation.
The models released here are still in the early stages of our research and development and have not been tuned to ensure outputs align with human intent and safety considerations.
llm-jp(at)nii.ac.jp
The names are listed in alphabetical order.
Hirokazu Kiyomaru, Hiroshi Matsuda, Jun Suzuki, Namgi Han, Saku Sugawara, Shota Sasaki, Shuhei Kurita, Taishi Nakamura, Takashi Kodama, Takumi Okamoto.