metadata

library_name: transformers
tags:
  - unsloth
  - trl
  - sft
license: mit
datasets:
  - elyza/JaMARD
  - EleutherAI/hendrycks_math
language:
  - ja
  - en
base_model:
  - sbintuitions/sarashina2-13b
pipeline_tag: text-generation

Model Card for Sarashina2-13b-finetuned-v1 🧮📘

A Japanese-English bilingual language model fine-tuned on 100 randomly shuffled samples from math datasets. Based on sbintuitions/sarashina2-13b and trained using unsloth, trl, and sft.

Model Details

Description

🧠 Fine-tuned transformer model for solving math problems and answering technical questions in 🇯🇵 Japanese and 🇺🇸 English.

Base Model: sbintuitions/sarashina2-13b
License: MIT
Languages: Japanese, English
Model Type: Transformer (Causal LM)

Limitations

⚠️ Known limitations:

May generate incorrect or biased answers
May hallucinate explanations

🔍 Always verify important outputs.

Get Started

from transformers import AutoModelForCausalLM, AutoTokenizer
model = AutoModelForCausalLM.from_pretrained("Mori-kamiyama/sarashina-13B-finetuned-v1")
tokenizer = AutoTokenizer.from_pretrained("Mori-kamiyama/sarashina-13B-finetuned-v1")

Training Data

🧮 Fine-tuned using 100 randomly shuffled samples from:

elyza/JaMARD (Japanese instruction tuning)
EleutherAI/hendrycks_math (math reasoning)

Architecture

Transformer (Causal LM)
Fine-tuned with TRL + Unsloth