Downloads · 30 days
0
satoyutaka/LLM2025_SFT10-iter300
LLM2025_SFT10-iter300 is a machine learning model from satoyutaka. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
This repository provides multiple LoRA adapter checkpoints fine-tuned from Qwen/Qwen3-4B-Instruct-2507. Unlike the standard V10 release, these weights are stored in the Hugging Face PEFT format, making them directly c…
Downloads · 30 days
0
Access
Public
Updated Feb 23, 2026
Repo size
58.8 MB
Likes
0
Public
Click a slice to open those files.
.safetensors29.4 MB · 100%
From the Hugging Face model README
This repository provides multiple LoRA adapter checkpoints fine-tuned from Qwen/Qwen3-4B-Instruct-2507. Unlike the standard V10 release, these weights are stored in the Hugging Face PEFT format, making them directly compatible with Linux, vLLM, and standard Python Transformers environments.
To help find the best balance between silence and intelligence, we provide several checkpoints from the training run:
/iter300: Early stage. Low hallucination risk./iter300: Mid-early stage (Coming soon)./iter500: Mid stage (Coming soon)./iter3000: Final stage (Trained to match exact data patterns).This adapter series is specifically trained for "No-Preamble" structured outputs. The model is trained to start its response directly with { or < characters.
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base_id = "Qwen/Qwen3-4B-Instruct-2507"
adapter_id = "satoyutaka/LLM2025_SFT10_HF"
model = AutoModelForCausalLM.from_pretrained(base_id, device_map="auto")
# Specify the sub-folder for the desired iteration
model = PeftModel.from_pretrained(model, adapter_id, subfolder="iter300")
このリポジトリは、Qwen/Qwen3-4B-Instruct-2507 をベースとしてファインチューニングされた LoRA アダプターの、各チェックポイントを提供します。
Hugging Face PEFT 形式で保存されているため、vLLM や 標準の Transformers ライブラリを使用する Linux 環境等でそのままロード・マージが可能です。
「沈黙」と「知能」の最適なバランスを検証できるよう、以下の各段階の重みを用意しています。
/iter300: 学習初期。過学習(暗記)のリスクが低く、最も安全な推論が期待できます。/iter300: 中盤初期(順次追加予定)。/iter500: 中盤(順次追加予定)。/iter3000: 最終版。学習データのパターンを最も強く反映しています。**「前置きの完全排除(No-Preamble)」**に特化しています。AIは回答の冒頭から直接データ(JSON/XML等)を書き始めるように訓練されています。
from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel
base_id = "Qwen/Qwen3-4B-Instruct-2507"
adapter_id = "satoyutaka/LLM2025_SFT10_HF"
model = AutoModelForCausalLM.from_pretrained(base_id, device_map="auto")
# subfolder に評価したいイテレーションを指定してください
model = PeftModel.from_pretrained(model, adapter_id, subfolder="iter300")