【2026年最新】オープンソースLLM徹底比較：Llama 3・Mistral・Gemma・Qwenの性能と選び方

# vLLMでLlama 3.1を起動する例
from vllm import LLM, SamplingParams

llm = LLM(
    model="meta-llama/Llama-3.1-8B-Instruct",
    tensor_parallel_size=1,
    max_model_len=8192,
    gpu_memory_utilization=0.9,
)

sampling_params = SamplingParams(
    temperature=0.7,
    top_p=0.9,
    max_tokens=512,
)

prompts = ["日本の首都について教えてください。"]
outputs = llm.generate(prompts, sampling_params)

for output in outputs:
    print(output.outputs[0].text)

# Mistralモデルのロード例
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "mistralai/Mistral-Small-Instruct-2409"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto",
)

messages = [
    {"role": "system", "content": "You are a helpful assistant."},
    {"role": "user", "content": "Explain the concept of MoE in LLMs."},
]

input_ids = tokenizer.apply_chat_template(
    messages, return_tensors="pt"
).to(model.device)

outputs = model.generate(
    input_ids,
    max_new_tokens=512,
    temperature=0.7,
    do_sample=True,
)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

# Qwen 2.5のロード例
from transformers import AutoModelForCausalLM, AutoTokenizer

model_name = "Qwen/Qwen2.5-7B-Instruct"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto",
)

messages = [
    {"role": "system", "content": "あなたは優秀な日本語AIアシスタントです。"},
    {"role": "user", "content": "機械学習とディープラーニングの違いを説明してください。"},
]

text = tokenizer.apply_chat_template(
    messages, tokenize=False, add_generation_prompt=True
)
inputs = tokenizer([text], return_tensors="pt").to(model.device)

outputs = model.generate(
    **inputs,
    max_new_tokens=512,
    temperature=0.7,
    top_p=0.9,
)

response = tokenizer.decode(
    outputs[0][len(inputs.input_ids[0]):],
    skip_special_tokens=True
)
print(response)

# vLLMサーバーの起動
pip install vllm

# OpenAI互換APIサーバーとして起動
python -m vllm.entrypoints.openai.api_server \
    --model Qwen/Qwen2.5-7B-Instruct \
    --host 0.0.0.0 \
    --port 8000 \
    --max-model-len 8192 \
    --gpu-memory-utilization 0.9 \
    --tensor-parallel-size 1

# OpenAI SDKで接続
from openai import OpenAI

client = OpenAI(
    base_url="http://localhost:8000/v1",
    api_key="dummy",
)

response = client.chat.completions.create(
    model="Qwen/Qwen2.5-7B-Instruct",
    messages=[
        {"role": "system", "content": "あなたは優秀なAIアシスタントです。"},
        {"role": "user", "content": "Pythonでフィボナッチ数列を生成する関数を書いてください。"},
    ],
    temperature=0.7,
    max_tokens=512,
)

print(response.choices[0].message.content)

# Ollamaのインストール（Linux）
curl -fsSL https://ollama.ai/install.sh | sh

# モデルのダウンロードと実行
ollama pull qwen2.5:7b
ollama run qwen2.5:7b

# API経由で利用
curl http://localhost:11434/api/chat -d '{
  "model": "qwen2.5:7b",
  "messages": [
    {
      "role": "user",
      "content": "日本の四季について教えてください。"
    }
  ],
  "stream": false
}'

version: '3.8'

services:
  vllm:
    image: vllm/vllm-openai:latest
    ports:
      - "8000:8000"
    volumes:
      - ./models:/models
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: 1
              capabilities: [gpu]
    command: >
      --model Qwen/Qwen2.5-7B-Instruct
      --host 0.0.0.0
      --port 8000
      --max-model-len 8192
      --gpu-memory-utilization 0.9      

  nginx:
    image: nginx:alpine
    ports:
      - "80:80"
      - "443:443"
    volumes:
      - ./nginx.conf:/etc/nginx/nginx.conf
    depends_on:
      - vllm

  prometheus:
    image: prom/prometheus:latest
    ports:
      - "9090:9090"
    volumes:
      - ./prometheus.yml:/etc/prometheus/prometheus.yml

  grafana:
    image: grafana/grafana:latest
    ports:
      - "3000:3000"
    environment:
      - GF_SECURITY_ADMIN_PASSWORD=admin

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import LoraConfig, get_peft_model, TaskType
from trl import SFTTrainer, SFTConfig

# ベースモデルのロード
model_name = "Qwen/Qwen2.5-7B-Instruct"
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype="auto",
    device_map="auto",
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

# LoRA設定
lora_config = LoraConfig(
    task_type=TaskType.CAUSAL_LM,
    r=16,
    lora_alpha=32,
    lora_dropout=0.05,
    target_modules=["q_proj", "k_proj", "v_proj", "o_proj"],
)

# PEFTモデルの作成
model = get_peft_model(model, lora_config)
model.print_trainable_parameters()
# 出力例: trainable params: 13,631,488 || all params: 7,628,000,000 || trainable%: 0.18%

# トレーニング設定
training_config = SFTConfig(
    output_dir="./output",
    num_train_epochs=3,
    per_device_train_batch_size=4,
    gradient_accumulation_steps=4,
    learning_rate=2e-4,
    warmup_ratio=0.1,
    logging_steps=10,
    save_strategy="epoch",
    fp16=True,
    max_seq_length=2048,
)

# トレーナーの作成と実行
trainer = SFTTrainer(
    model=model,
    args=training_config,
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
    tokenizer=tokenizer,
)

trainer.train()
trainer.save_model("./finetuned-model")