metadata
license: apache-2.0
datasets:
- llm-jp/oasst1-21k-ja
- llm-jp/oasst2-33k-ja
- HachiML/Hachi-Alpaca
- Aratako/Rosebleu-1on1-Dialogues-RP
- baobab-trees/wikipedia-human-retrieval-ja
- aixsatoshi/Longcontext-aozora-summary
- aixsatoshi/Longcontext-aozora-instruction
- kunishou/amenokaku-code-instruct
- HachiML/Evol-hh-rlhf-gen3-1k
- Kendamarron/jimba-wiki-instruction-calm3
- Manual-Dataset-Creation-Project/Malum-130
- sudy-super/CoTangent
- minnade/chat-daily
Yamase-12B
Description
Yamase-12Bは、Mistral-Nemo-Instructに対して日本語能力の向上を目的として約11万件のデータでFine-tuningを行ったモデルです。
Usage
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
B_INST, E_INST = "[INST]", "[/INST]"
text = "旅行に行くと高層ビルがたくさん建っていました。これからどのようなことが推測できますか？"
model_name = "sudy-super/Yamase-12B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype=torch.bfloat16)
if torch.cuda.is_available():
model = model.to("cuda")
prompt = "{bos_token}{b_inst}{prompt}{e_inst}".format(
bos_token=tokenizer.bos_token,
b_inst=B_INST,
prompt=text,
e_inst=E_INST,
)
with torch.no_grad():
token_ids = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
output_ids = model.generate(
token_ids.to(model.device),
max_new_tokens=256,
pad_token_id=tokenizer.pad_token_id,
eos_token_id=tokenizer.eos_token_id,
)
output = tokenizer.decode(output_ids.tolist()[0][token_ids.size(1) :], skip_special_tokens=True)
print(output)
"""
"""
Chat Template
<s>[INST]明日の東京の天気は何ですか？[/INST]晴れです。</s>[INST]大阪はどうですか？[/INST]雨です。</s>
Hyperparameter
num_train_epochs: 5
per_device_train_batch_size: 2
per_device_eval_batch_size: 2
gradient_accumulation_steps: 128
learning_rate: 2e-5
lr_scheduler_kwargs={"min_lr": 2e-6}
lr_scheduler_type: "cosine_with_min_lr"
warmup_ratio: 0.1
dataloader_pin_memory: True
gradient_checkpointing: True
bf16: True
optim: "adamw_torch_fused"
weight_decay: 0.0
max_grad_norm: 1.0
adam_beta2: 0.99
label_smoothing_factor: 0.0
seed: 42