Update src/llm/local_model_client.py

Update src/llm/local_model_client.py

Update src/llm/local_model_client.py

Update src/llm/local_model_client.py

Add HuggingFace token support for gated Llama model access

Fix Be.FM to support CUDA and CPU in addition to MPS

Change model to Claude 3 Haiku for cost efficiency

Fix model 404 error: Use claude-3-5-sonnet-20240620 instead of 20241022

Fix HF Spaces initialization and add comprehensive diagnostics

Use explicit httpx client to avoid proxies parameter error

Add version compatibility handling for Anthropic client proxies error

Fix IndentationError and proxies parameter issue in AnthropicClient

Fix Anthropic client initialization error

Optimize Be.FM MPS performance - 17x speedup (5min → 18sec)

Add Edit Personas page and complete model selector integration

Fix watermark ratio error and default to MPS GPU

Apply comprehensive MPS stability improvements for M4 Mac

Merge PEFT weights to enable stable MPS GPU inference

Use CPU by default for Be.FM to avoid MPS segmentation faults

Add thread-safe inference to prevent MPS segmentation faults

Add thread-safe model loading and fix deprecation warnings

Fix MPS device loading for Be.FM model on M4 Mac

Fix lazy loading for LocalModelClient dependencies

Integrate Stanford Be.FM model for local LLM inference

Implement Phase 1: Persona-based LLM query system for urban planning