- Added proxy methods to MultiModelManager class for cache module functions
- These methods are called by images.py sd.cpp fallback path
- Fixes AttributeError: 'MultiModelManager' object has no attribute 'get_cached_model_path'

- Enhanced the HF model resolution logic in images.py sd.cpp fallback path
- Now checks for ANY cached file from the repo first (not just GGUF files)
- Falls back to checking for cached GGUF files specifically
- Last resort: downloads the first file in the repo as fallback
- Better error handling and logging throughout the resolution process
- This should resolve models that are already cached even if the exact GGUF filename isn't known

- Enhanced model resolution for sd.cpp fallback path
- Added multiple fallback strategies:
  1. Try HuggingFace GGUF resolution (existing)
  2. Fallback to direct file path check
  3. Fallback to cached model lookup
  4. Last resort: attempt download as URL
- Better error logging and handling
- Ensures model loading attempts all possible resolution paths before failing

- Added model resolution and unload logic to /v1/audio/transcriptions
- Added model resolution and unload logic to /v1/audio/speech (TTS)
- Now ALL endpoints (text, image, audio, TTS) properly handle model switching
- In ondemand mode, ANY model type switch triggers unload first (e.g., text->audio, TTS->image, etc.)

- Added resolve_model_name() to MultiModelManager to properly resolve model aliases
- Added get_currently_loaded_model_name() to track what's actually in VRAM
- Updated /v1/chat/completions, /v1/completions, and /v1/images/generations
- Now correctly compares resolved canonical names before deciding to unload
- Handles all aliases (default, image, audio, tts) and custom aliases
- Works across ALL model types: text->text2, image->image2, text->image, etc.

- Added unload_all_models() to MultiModelManager that handles ALL model types:
  ModelManager, diffusers pipelines, sd.cpp StableDiffusion, and any other objects
- Text endpoints now properly unload image models before loading text models
- Image endpoints now properly unload text models before loading image models
- The rule: in ondemand mode, if the model in VRAM differs from the requested
  model (regardless of type), fully unload before loading the new one
- Includes gc.collect(), torch.cuda.empty_cache(), and 1s settle delay

- In ondemand mode (no --load-all or --loadswap specified), when a new model
  is requested, the current model in VRAM is now fully unloaded before loading
  the new one. This ensures clean model switching.
- Added cleanup logic to both /v1/chat/completions and /v1/completions endpoints
- Added same logic to image generation endpoints (diffusers and sd.cpp paths)
- Cleanup includes: model cleanup, gc.collect(), torch.cuda.empty_cache()

Root cause: The refactored code was hardcoding torch.float16 for CUDA,
ignoring the --image-precision bf16 CLI argument. The Z-Image-Turbo model
requires bfloat16 precision - using float16 causes NaN values in the
image processor, resulting in all-black images.

Also restored the original model loading logic with:
- GGUF model detection (skip diffusers for GGUF)
- OOM retry with progressive memory optimization
- use_safetensors=True
- Sequential CPU offload support

- Changed default image size from 512x512 back to 1024x1024 to match original coderai
- Changed NaN handling from 0.5 to 0.0 to match original coderai

- Added set_global_args call for images module in main.py
- Each API module has its own global_args, so it needs to be set separately
- Added debug logging to trace global_args in images.py

- Fixed file path not being set in app.py for /v1/files endpoint
- Fixed Host header parsing to correctly extract hostname without port
- Added debug logging to trace URL construction and file serving

Fix: Use DiffusionPipeline for custom model support (ZImagePipeline) - was using it in original code

- Default load mode is now 'loadall' (preload) instead of 'ondemand'
- Only use ondemand when --nopreload is explicitly specified
- Model will now be loaded at startup by default

- get_model_for_request now triggers model loading if not already loaded
- Added _load_default_model() method to load default model on demand
- Added _load_model_by_name() method to load any model on demand
- Fixes 503 'Model not loaded' error when requesting 'default' model

- Show full request body without truncation
- Include HTTP method, URL, and headers
- Pretty-print JSON bodies

- text.py had local global_debug variable that shadowed the state module
- Changed text.py to import get_global_debug from state module
- Changed set_global_debug() in text.py to call state module's function
- Changed all 'if global_debug:' to 'if get_global_debug():' in text.py
- log.py was already using get_global_debug() correctly

- Create codai/api/state.py for shared global state functions
- images.py now imports get_load_mode from state instead of app
- app.py re-exports functions from state for backward compatibility

- Move parse_args to codai.cli
- Move main() to codai.main
- Simplify coderai to be a thin wrapper importing from codai package
- Create codai.api module with organized endpoints:
  - codai/api/app.py: FastAPI app, /v1/models, /v1/files, get_load_mode
  - codai/api/text.py: /v1/chat/completions, legacy /v1/completions
  - codai/api/images.py: /v1/images/generations
  - codai/api/transcriptions.py: /v1/audio/transcriptions
  - codai/api/tts.py: /v1/audio/speech
- coderai is now backward compatible entry point only

- Fixed AttributeError where Tool.get() was called on Pydantic model
- Added isinstance() checks to handle both dict and Pydantic Tool formats
- This fixes the error when using --force-reasoning with tools

- Added repeat_penalty, presence_penalty, frequency_penalty params to generate() and generate_stream()
- Changed from **kwargs to explicit parameters to match base class abstract methods

This fixes the TypeError when calling VulkanBackend.generate_stream() with extra params.

- Added missing parameters to generate() and generate_stream() methods
- Updated _generate_normal() and _generate_stream_normal() to use these params
- Also updated base.py abstract method signatures to match

This fixes the TypeError when using repeat_penalty with NVIDIA backend.

New patterns added to repair_broken_tool_calls():

1. Pattern 4: <tool><function>NAME</function><parameters>XML</parameters></tool>
   - Converts XML parameters to JSON format
   - Fills missing required params (e.g., path for list_files)

2. Pattern 0a: <tool><NAME><params></NAME></tool> (with closing tool name tag)
   - Handles format with closing tag for tool name

3. Expanded guard to detect known tool names used as wrapper tags
   - Now detects <fetch_instructions>, <list_files>, etc.

4. Fixed closure bug in Pattern -2 (wrong wrapper tags)
   - Used default argument to capture loop variable correctly

5. Post-processing: Fill missing required parameters
   - list_files gets path='.' if missing
   - search_files gets path='.' if missing

All 6 test cases pass:
- <tool><function>list_files</function><parameters>...</parameters></tool> -> OK
- <fetch_instructions><task>read_file</task>...</fetch_instructions> -> OK
- <tool_call><list_files></list_files></tool_call> -> OK
- <tool><list_files><path>.</path></list_files></tool> -> OK
- Valid JSON passthrough -> OK
- Missing required params auto-filled -> OK