import litellm

# ===========================================================================
# BEFORE: Custom deployment name — LiteLLM can't detect model type
# ===========================================================================

# ❌ This uses default Azure config because "my-gpt5-deployment" doesn't match
#    any known model pattern. max_tokens won't be mapped to max_completion_tokens,
#    and gpt-5-specific behavior is lost.

# --- SDK direct call (before) ---
response = litellm.completion(
    model="azure/gpt5_series/my-gpt5-deployment",
    messages=[{"role": "user", "content": "Hello!"}],
    max_tokens=100,
    api_key="your-azure-api-key",
    api_base="https://my-resource.openai.azure.com",
    api_version="2024-12-01-preview",
)
# Result: sends max_tokens=100 (incorrect for gpt-5, which expects max_completion_tokens)


# ===========================================================================
# AFTER: Using base_model — LiteLLM detects the correct model type
# ===========================================================================

# ✅ With base_model, LiteLLM knows this is a gpt-5 deployment.
#    max_tokens is automatically mapped to max_completion_tokens,
#    unsupported params (like temperature) are correctly rejected,
#    and the right config class (AzureOpenAIGPT5Config) is used.

# --- SDK direct call (after) ---
response = litellm.completion(
    model="azure/gpt5_series/my-gpt5-deployment",
    messages=[{"role": "user", "content": "Hello!"}],
    max_tokens=100,
    base_model="azure/gpt-5",  # <-- tells LiteLLM the underlying model
    api_key="your-azure-api-key",
    api_base="https://my-resource.openai.azure.com",
    api_version="2024-12-01-preview",
)
# Result: sends max_completion_tokens=100 (correct for gpt-5)