Downloads · 30 days
141
3% of all-time downloads
RichardErkhov/sbintuitions_-_sarashina2-13b-gguf
sbintuitions_-_sarashina2-13b-gguf is a machine learning model from RichardErkhov. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Downloads · 30 days
141
3% of all-time downloads
All-time downloads
4.5K
Public
Repo size
157 GB
Likes
1
Public
Click a slice to open those files.
.gguf182 GB · 100%
From the Hugging Face model README
Quantization made by Richard Erkhov.
sarashina2-13b - GGUF
| Name | Quant method | Size |
|---|---|---|
| sarashina2-13b.Q2_K.gguf | Q2_K | 4.91GB |
| sarashina2-13b.IQ3_XS.gguf | IQ3_XS | 5.41GB |
| sarashina2-13b.IQ3_S.gguf | IQ3_S | 5.69GB |
| sarashina2-13b.Q3_K_S.gguf | Q3_K_S | 5.69GB |
| sarashina2-13b.IQ3_M.gguf | IQ3_M | 5.99GB |
| sarashina2-13b.Q3_K.gguf | Q3_K | 6.32GB |
| sarashina2-13b.Q3_K_M.gguf | Q3_K_M | 6.32GB |
| sarashina2-13b.Q3_K_L.gguf | Q3_K_L | 6.87GB |
| sarashina2-13b.IQ4_XS.gguf | IQ4_XS | 6.99GB |
| sarashina2-13b.Q4_0.gguf | Q4_0 | 7.33GB |
| sarashina2-13b.IQ4_NL.gguf | IQ4_NL | 7.37GB |
| sarashina2-13b.Q4_K_S.gguf | Q4_K_S | 7.38GB |
| sarashina2-13b.Q4_K.gguf | Q4_K | 7.79GB |
| sarashina2-13b.Q4_K_M.gguf | Q4_K_M | 7.79GB |
| sarashina2-13b.Q4_1.gguf | Q4_1 | 8.09GB |
| sarashina2-13b.Q5_0.gguf | Q5_0 | 8.86GB |
| sarashina2-13b.Q5_K_S.gguf | Q5_K_S | 8.86GB |
| sarashina2-13b.Q5_K.gguf | Q5_K | 9.1GB |
| sarashina2-13b.Q5_K_M.gguf | Q5_K_M | 9.1GB |
| sarashina2-13b.Q5_1.gguf | Q5_1 | 9.63GB |
| sarashina2-13b.Q6_K.gguf | Q6_K | 10.5GB |
| sarashina2-13b.Q8_0.gguf | Q8_0 | 13.6GB |
license: mit language:
This repository provides large language models trained by SB Intuitions.
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline, set_seed
model = AutoModelForCausalLM.from_pretrained("sbintuitions/sarashina2-13b", torch_dtype=torch.bfloat16, device_map="auto")
tokenizer = AutoTokenizer.from_pretrained("sbintuitions/sarashina2-13b")
# If you want to use slow tokenizer
# tokenizer = AutoTokenizer.from_pretrained("sbintuitions/sarashina2-13b", use_fast=False)
generator = pipeline("text-generation", model=model, tokenizer=tokenizer)
set_seed(123)
text = generator(
"おはようございます、今日の天気は",
max_length=30,
do_sample=True,
pad_token_id=tokenizer.pad_token_id,
num_return_sequences=3,
)
for t in text:
print(t)
| Parameters | Vocab size | Training tokens | Architecture | Position type | Layers | Hidden dim | Attention heads |
|---|---|---|---|---|---|---|---|
| 7B | 102400 | 2.1T | Llama2 | RoPE | 32 | 4096 | 32 |
| 13B | 102400 | 2.1T | Llama2 | RoPE | 40 | 5120 | 40 |
| 70B | 102400 | 2.1T | Llama2 | RoPE | 80 | 8192 | 64 |
For our Japanese training data, we used a Japanese portion of the Common Crawl corpus, which is the largest Web corpus, as our training dataset. To clean the training corpus, we used CCNet and HojiChar. After cleaning, our Japanese training data contains about 1T tokens.
For our English training data, we extracted English documents from SlimPajama but we removed books3 corpus due to copyright infringement.
We use a sentencepiece tokenizer with a unigram language model and byte-fallback. We do not apply pre-tokenization with Japanese tokenizer. Thus, a user may directly feed raw sentences into the tokenizer.
Sarashina2 has not been tuned to follow an instruction yet. Therefore, sarashina2 might generate some meaningless sequences, some inaccurate instances or biased/objectionable outputs. Before using sarashina2, we would like developers to tune models based on human preferences and safety considerations.