Cloud Account Atlas: HuggingFace, OpenAI, Anthropic, Together, Modal, Runpod, Lambda — Which One and Why?

# HuggingFace CLI ile login (token'ı keyring'e güvenli saklar)
uv pip install huggingface_hub
huggingface-cli login
# Token'ı yapıştır → ENTER
 
# Test
huggingface-cli whoami
# Senin kullanıcı adın
 
# Bir modeli indir (gated değilse)
huggingface-cli download Qwen/Qwen2.5-7B-Instruct --local-dir ./models/qwen-7b

# .env
# OPENAI_API_KEY=sk-...
 
# Python
import os
from openai import OpenAI
 
client = OpenAI(api_key=os.environ["OPENAI_API_KEY"])
 
resp = client.chat.completions.create(
    model="gpt-5-mini",
    messages=[{"role": "user", "content": "Bir LLM Engineer'ın 3 günlük rutinini özetle."}],
    max_tokens=300,
)
print(resp.choices[0].message.content)
print(f"Maliyet (tahmini): ${(resp.usage.prompt_tokens * 0.25 + resp.usage.completion_tokens * 2.00) / 1_000_000:.6f}")

# Together AI — OpenAI uyumlu
from openai import OpenAI
import os
 
client = OpenAI(
    api_key=os.environ["TOGETHER_API_KEY"],
    base_url="https://api.together.xyz/v1",
)
 
resp = client.chat.completions.create(
    model="meta-llama/Llama-3.3-70B-Instruct-Turbo",
    messages=[{"role": "user", "content": "1+1 kaç eder?"}],
)
print(resp.choices[0].message.content)
# Maliyet: GPT-5'in ~%10'u

# llama_inference.py — Modal'da H100'de Llama 3.3 70B çalıştır
import modal
 
image = (
    modal.Image.debian_slim()
    .pip_install("vllm==0.6.4", "huggingface_hub")
    .env({"HF_HUB_ENABLE_HF_TRANSFER": "1"})
)
 
app = modal.App("llm-muh-llama-70b")
 
@app.function(
    image=image,
    gpu="H100",
    timeout=600,
    secrets=[modal.Secret.from_name("hf-token")],
)
def generate(prompt: str) -> str:
    from vllm import LLM, SamplingParams
    llm = LLM(
        model="meta-llama/Llama-3.3-70B-Instruct",
        quantization="awq",
        max_model_len=4096,
    )
    out = llm.generate([prompt], SamplingParams(max_tokens=200))
    return out[0].outputs[0].text
 
@app.local_entrypoint()
def main():
    print(generate.remote("LLM mühendisliği nedir?"))
 
# Kullanım:
# $ modal run llama_inference.py
# H100 üzerinde Llama 70B AWQ ile inference — toplam maliyet ~$0.20

# .env (örnek şablon — kursta kullanacağın anahtarlar)
# 🔒 BU DOSYAYI GIT'E COMMIT ETME — .gitignore'da olmalı
 
# === Model Hub ===
HF_TOKEN=hf_...
 
# === Closed LLM API ===
OPENAI_API_KEY=sk-proj-...
ANTHROPIC_API_KEY=sk-ant-...
 
# === Open-source LLM Inference ===
TOGETHER_API_KEY=...
FIREWORKS_API_KEY=...
GROQ_API_KEY=...
 
# === Compute ===
MODAL_TOKEN_ID=ak-...
MODAL_TOKEN_SECRET=as-...
RUNPOD_API_KEY=...
LAMBDA_API_KEY=...
 
# === Aggregator (geliştirme için pratik) ===
OPENROUTER_API_KEY=sk-or-...
 
# === Observability ===
LANGFUSE_PUBLIC_KEY=pk-lf-...
LANGFUSE_SECRET_KEY=sk-lf-...
LANGFUSE_HOST=https://cloud.langfuse.com
 
# === Vector DB (RAG capstone'da) ===
QDRANT_URL=https://...
QDRANT_API_KEY=...
PINECONE_API_KEY=...
 
# === Diğer ===
SERPER_API_KEY=...     # Web search
TAVILY_API_KEY=...     # Agentic search