Fix: wire temperature through to generation in LocalLLM

[howardjaw]

Closes #6.

What

Fixes a bug where the temperature parameter on LocalLLM is silently ignored at generation time. Also addresses a related reproducibility issue where identical responses repeat across runs because no seed is passed.

Why

llama-cpp-python applies temperature at sampling time (per-call), not at model load. The current code passes temperature to the Llama(...) constructor, where it has no effect on subsequent __call__ invocations. As a result, readers following Lesson 01 Exercise 2 ("Change the temperature in shared/llm.py") see no change when they edit the default, which contradicts the lesson text.

Changes

Four edits in shared/llm.py: (1) store self.temperature = temperature in LocalLLM.__init__; (2) in generate(), default kwargs["temperature"] to self.temperature when no per-call override is supplied; (3) remove the now-unused temperature=temperature kwarg from the Llama(...) call; (4) add seed=-1 to the Llama(...) call so each model load uses a fresh random seed.

Testing

Ran lesson_01_basic_chat() twice with temperature=0.0 (near-identical output across runs, as expected) and twice with temperature=1.5 (clearly different output each run). Previously both temperatures produced byte-identical output across runs regardless of the value set in __init__.