Session 6· 01· 10 min

Model Initialisation & Parameters

What you'll learn
  • Pass temperature, max_tokens, timeout, max_retries to init_chat_model
  • Understand which parameters are provider-agnostic
  • See the unified constructor interface

init_chat_model accepts common LLM parameters directly. Temperature, max_tokens, timeout, and max_retries work the same way across every provider — LangChain maps them to the right field in each underlying SDK.

scripts/01_model_init_and_params.py
model = init_chat_model(
    f"openai:{model_name}",                                 ①
    temperature=0.2,                                         ②
    max_tokens=120,                                          ③
    timeout=30,                                              ④
    max_retries=2,                                           ⑤
)

response = model.invoke(
    "In exactly one line, explain why model parameters matter."
)
print(response.content)
Provider prefix + model name — switch providers by changing only this string.
temperature — same 0.0 deterministic to 1.2 creative scale as Session 1.
max_tokens — cap on response length.
timeout — how long to wait for a response before giving up (seconds).
max_retries — how many times LangChain should auto-retry transient errors.
$ python scripts/01_model_init_and_params.py
max_retries is free resilience
With max_retries=2, a transient 500 error or rate-limit will be automatically retried (with exponential backoff). You get three attempts total with one config line instead of writing a retry loop.
Knowledge Check
Which constructor argument makes LangChain retry transient errors automatically?
Recap — what you just learned
  • init_chat_model accepts common LLM params directly
  • temperature, max_tokens, timeout, max_retries work across all providers
  • max_retries gives you automatic retry logic for free
Next up: 02 — Invoke & Messages