fix: remove unused imports and apply ruff formatting

- Remove unused imports: json (llm.py), tempfile (stt.py), base64 (tts.py)
- Apply ruff format to all Python files

llm.py

@@ -9,10 +9,10 @@ Features:
 - Token usage and latency metrics
 - Chat history management
 """
+
 import os
 import time
 import logging
-import json
 
 import gradio as gr
 import httpx
@@ -65,7 +65,9 @@ try:
     _mlflow_run_id = _mlflow_run.info.run_id
     _mlflow_step = 0
     MLFLOW_ENABLED = True
-    logger.info("MLflow tracking enabled: experiment=%s run=%s", _experiment_id, _mlflow_run_id)
+    logger.info(
+        "MLflow tracking enabled: experiment=%s run=%s", _experiment_id, _mlflow_run_id
+    )
 except Exception as exc:
     logger.warning("MLflow tracking disabled: %s", exc)
     _mlflow_client = None
@@ -95,18 +97,25 @@ def _log_llm_metrics(
             _mlflow_run_id,
             metrics=[
                 mlflow.entities.Metric("latency_s", latency, ts, _mlflow_step),
-                mlflow.entities.Metric("prompt_tokens", prompt_tokens, ts, _mlflow_step),
-                mlflow.entities.Metric("completion_tokens", completion_tokens, ts, _mlflow_step),
+                mlflow.entities.Metric(
+                    "prompt_tokens", prompt_tokens, ts, _mlflow_step
+                ),
+                mlflow.entities.Metric(
+                    "completion_tokens", completion_tokens, ts, _mlflow_step
+                ),
                 mlflow.entities.Metric("total_tokens", total_tokens, ts, _mlflow_step),
                 mlflow.entities.Metric("tokens_per_second", tps, ts, _mlflow_step),
                 mlflow.entities.Metric("temperature", temperature, ts, _mlflow_step),
-                mlflow.entities.Metric("max_tokens_requested", max_tokens, ts, _mlflow_step),
+                mlflow.entities.Metric(
+                    "max_tokens_requested", max_tokens, ts, _mlflow_step
+                ),
                 mlflow.entities.Metric("top_p", top_p, ts, _mlflow_step),
             ],
         )
     except Exception:
         logger.debug("MLflow log failed", exc_info=True)
 
+
 DEFAULT_SYSTEM_PROMPT = (
     "You are a helpful AI assistant running on Davies Tech Labs homelab infrastructure. "
     "You are powered by Llama 3.1 70B served via vLLM on AMD Strix Halo (ROCm). "
@@ -273,10 +282,10 @@ def single_prompt(
         metrics = f"""
 **Generation Metrics:**
 - Latency: {latency:.1f}s
-- Prompt tokens: {usage.get('prompt_tokens', 'N/A')}
-- Completion tokens: {usage.get('completion_tokens', 'N/A')}
-- Total tokens: {usage.get('total_tokens', 'N/A')}
-- Model: {result.get('model', 'N/A')}
+- Prompt tokens: {usage.get("prompt_tokens", "N/A")}
+- Completion tokens: {usage.get("completion_tokens", "N/A")}
+- Total tokens: {usage.get("total_tokens", "N/A")}
+- Model: {result.get("model", "N/A")}
 """
         return text, metrics
 
@@ -360,9 +369,13 @@ Chat with **Llama 3.1 70B** (AWQ INT4) served via vLLM on AMD Strix Halo (ROCm).
 
             gr.Examples(
                 examples=[
-                    ["Summarise the key differences between CUDA and ROCm for ML workloads."],
+                    [
+                        "Summarise the key differences between CUDA and ROCm for ML workloads."
+                    ],
                     ["Write a haiku about Kubernetes."],
-                    ["Explain Ray Serve in one paragraph for someone new to ML serving."],
+                    [
+                        "Explain Ray Serve in one paragraph for someone new to ML serving."
+                    ],
                     ["List 5 creative uses for a homelab GPU cluster."],
                 ],
                 inputs=[prompt_input],