nimcode 0.8.7__tar.gz → 0.8.8__tar.gz

{nimcode-0.8.7 → nimcode-0.8.8}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nimcode
-Version: 0.8.7
+Version: 0.8.8
 Summary: A standalone, robust coding agent for NVIDIA NIM models.
 Home-page: https://github.com/Batunash/nimcode
 Author: Autonomous Agent

nimcode-0.8.8/src/nimcode/__version__.py

@@ -0,0 +1 @@
+__version__ = "0.8.8"

nimcode-0.8.8/src/nimcode/model_registry.py

@@ -0,0 +1,129 @@
+import logging
+import os
+import json
+import urllib.request
+import threading
+import time
+
+logger = logging.getLogger(__name__)
+
+# Known NVIDIA NIM models and their context window sizes (in tokens)
+# This acts as a fallback + enrichment layer when API is unavailable
+MODEL_CONTEXT_WINDOWS = {
+    # Llama 3.1 family
+    "meta/llama-3.1-8b-instruct": 128000,
+    "meta/llama-3.3-70b-instruct": 128000,
+    "meta/llama-3.1-405b-instruct": 128000,
+    # Llama 3.2 family
+    "meta/llama-3.2-1b-instruct": 128000,
+    "meta/llama-3.2-3b-instruct": 128000,
+    "meta/llama-3.2-11b-vision-instruct": 128000,
+    "meta/llama-3.2-90b-vision-instruct": 128000,
+    # Llama 3.3 family
+    "meta/llama-3.3-70b-instruct": 128000,
+    # Nemotron
+    "nvidia/nemotron-4-340b-instruct": 4096,
+    "nvidia/llama-3.1-nemotron-70b-instruct": 128000,
+    # Mistral
+    "mistralai/mixtral-8x22b-instruct-v0.1": 65536,
+    "mistralai/mistral-large-2-instruct": 128000,
+    # DeepSeek
+    "deepseek-ai/deepseek-r1": 128000,
+    "deepseek-ai/deepseek-v4-pro": 128000,
+    # Qwen
+    "qwen/qwen2.5-72b-instruct": 128000,
+    "qwen/qwq-32b": 128000,
+}
+
+# Default context window for unknown models
+DEFAULT_CONTEXT_WINDOW = 8192
+
+# Hardcoded fallback model list (used when API is unreachable)
+FALLBACK_MODELS = [
+    "meta/llama-3.3-70b-instruct",
+    "meta/llama-3.1-8b-instruct",
+    "meta/llama-3.1-405b-instruct",
+    "meta/llama-3.3-70b-instruct",
+    "nvidia/llama-3.1-nemotron-70b-instruct",
+    "deepseek-ai/deepseek-r1",
+    "mistralai/mixtral-8x22b-instruct-v0.1",
+    "qwen/qwen2.5-72b-instruct",
+]
+
+_CACHE_FILE = os.path.expanduser("~/.nimcode/model_contexts.json")
+_CACHE_MAX_AGE_DAYS = 7
+_DYNAMIC_CACHE = {}
+_CACHE_LOADED = False
+_UPDATE_THREAD_STARTED = False
+
+def _fetch_and_cache_models():
+    """Background task to fetch latest model info from LiteLLM."""
+    url = "https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json"
+    try:
+        req = urllib.request.Request(url, headers={'User-Agent': 'Mozilla/5.0 Nimcode/1.0'})
+        with urllib.request.urlopen(req, timeout=15) as response:
+            data = json.loads(response.read().decode('utf-8'))
+            
+            processed = {}
+            for k, v in data.items():
+                if isinstance(v, dict) and "max_tokens" in v and isinstance(v["max_tokens"], int):
+                    processed[k] = v["max_tokens"]
+                    basename = k.split("/")[-1]
+                    if basename not in processed or processed[basename] < v["max_tokens"]:
+                        processed[basename] = v["max_tokens"]
+            
+            os.makedirs(os.path.dirname(_CACHE_FILE), exist_ok=True)
+            with open(_CACHE_FILE, "w", encoding="utf-8") as f:
+                json.dump(processed, f, indent=4)
+                
+            global _DYNAMIC_CACHE
+            _DYNAMIC_CACHE.update(processed)
+            logger.debug(f"Successfully cached {len(processed)} models from LiteLLM.")
+    except Exception as e:
+        logger.debug(f"Failed to fetch model contexts: {e}")
+
+def _load_or_update_cache():
+    """Loads cache from disk, spawns background update if old/missing."""
+    global _CACHE_LOADED, _DYNAMIC_CACHE, _UPDATE_THREAD_STARTED
+    if _CACHE_LOADED:
+        return
+
+    _CACHE_LOADED = True
+    needs_update = True
+    
+    if os.path.exists(_CACHE_FILE):
+        try:
+            with open(_CACHE_FILE, "r", encoding="utf-8") as f:
+                _DYNAMIC_CACHE = json.load(f)
+            
+            # Check age
+            mtime = os.path.getmtime(_CACHE_FILE)
+            age_days = (time.time() - mtime) / (24 * 3600)
+            if age_days < _CACHE_MAX_AGE_DAYS:
+                needs_update = False
+        except Exception as e:
+            logger.debug(f"Error reading cache file: {e}")
+
+    if needs_update and not _UPDATE_THREAD_STARTED:
+        _UPDATE_THREAD_STARTED = True
+        t = threading.Thread(target=_fetch_and_cache_models, daemon=True)
+        t.start()
+
+def get_context_window(model_id: str) -> int:
+    """Returns the context window size for a given model ID."""
+    _load_or_update_cache()
+    
+    # 1. Check hardcoded exact match
+    if model_id in MODEL_CONTEXT_WINDOWS:
+        return MODEL_CONTEXT_WINDOWS[model_id]
+        
+    # 2. Check dynamic cache exact match
+    if model_id in _DYNAMIC_CACHE:
+        return _DYNAMIC_CACHE[model_id]
+        
+    # 3. Check dynamic cache basename match
+    basename = model_id.split("/")[-1]
+    if basename in _DYNAMIC_CACHE:
+        return _DYNAMIC_CACHE[basename]
+        
+    return DEFAULT_CONTEXT_WINDOW

{nimcode-0.8.7 → nimcode-0.8.8}/src/nimcode.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: nimcode
-Version: 0.8.7
+Version: 0.8.8
 Summary: A standalone, robust coding agent for NVIDIA NIM models.
 Home-page: https://github.com/Batunash/nimcode
 Author: Autonomous Agent

nimcode-0.8.8/tests/test_model_registry.py

@@ -0,0 +1,90 @@
+import os
+import json
+import time
+import pytest
+from unittest.mock import patch, mock_open, MagicMock
+
+import nimcode.model_registry as mr
+
+@pytest.fixture(autouse=True)
+def reset_registry_state():
+    """Reset the module-level state before each test."""
+    mr._CACHE_LOADED = False
+    mr._UPDATE_THREAD_STARTED = False
+    mr._DYNAMIC_CACHE = {}
+    yield
+
+def test_get_context_window_hardcoded():
+    # Known model in hardcoded dictionary
+    assert mr.get_context_window("meta/llama-3.1-8b-instruct") == 128000
+    
+def test_get_context_window_unknown():
+    # Completely unknown model
+    assert mr.get_context_window("unknown/model-123") == mr.DEFAULT_CONTEXT_WINDOW
+
+@patch("os.path.exists")
+@patch("os.path.getmtime")
+@patch("builtins.open", new_callable=mock_open, read_data='{"provider/test-model": 64000, "test-model": 64000}')
+def test_get_context_window_from_cache(mock_file, mock_getmtime, mock_exists):
+    mock_exists.return_value = True
+    mock_getmtime.return_value = time.time()  # Very recent cache
+    
+    # Force reload
+    mr._load_or_update_cache()
+    
+    assert mr.get_context_window("provider/test-model") == 64000
+    assert mr.get_context_window("other_provider/test-model") == 64000
+    assert mr._UPDATE_THREAD_STARTED is False  # Cache is fresh, no thread spawned
+
+@patch("os.path.exists")
+@patch("threading.Thread")
+def test_get_context_window_spawns_thread_if_no_cache(mock_thread, mock_exists):
+    mock_exists.return_value = False
+    
+    # Should spawn thread to fetch cache
+    mr.get_context_window("some-model")
+    
+    assert mr._UPDATE_THREAD_STARTED is True
+    mock_thread.assert_called_once()
+    mock_thread.return_value.start.assert_called_once()
+
+@patch("os.path.exists")
+@patch("os.path.getmtime")
+@patch("threading.Thread")
+@patch("builtins.open", new_callable=mock_open, read_data='{"old-model": 1000}')
+def test_get_context_window_spawns_thread_if_old_cache(mock_file, mock_thread, mock_getmtime, mock_exists):
+    mock_exists.return_value = True
+    mock_getmtime.return_value = time.time() - (8 * 24 * 3600)  # 8 days old
+    
+    mr.get_context_window("old-model")
+    
+    assert mr._UPDATE_THREAD_STARTED is True
+    mock_thread.assert_called_once()
+    mock_thread.return_value.start.assert_called_once()
+
+@patch("urllib.request.urlopen")
+def test_fetch_and_cache_models(mock_urlopen, tmp_path):
+    mock_response = MagicMock()
+    mock_response.read.return_value = b'''
+    {
+        "provider/model-A": {"max_tokens": 1234},
+        "provider/model-B": {"max_tokens": 5678, "other": "ignored"},
+        "provider/model-C": {"ignored": 10}
+    }
+    '''
+    mock_response.__enter__.return_value = mock_response
+    mock_urlopen.return_value = mock_response
+    
+    with patch("nimcode.model_registry._CACHE_FILE", str(tmp_path / "cache.json")):
+        mr._fetch_and_cache_models()
+        
+        assert "provider/model-A" in mr._DYNAMIC_CACHE
+        assert mr._DYNAMIC_CACHE["provider/model-A"] == 1234
+        assert mr._DYNAMIC_CACHE["model-A"] == 1234
+        assert "provider/model-C" not in mr._DYNAMIC_CACHE
+        
+        # Verify file is written
+        with open(str(tmp_path / "cache.json"), "r") as f:
+            data = json.load(f)
+            assert data["provider/model-A"] == 1234
+            assert data["model-A"] == 1234

nimcode-0.8.7/src/nimcode/__version__.py

@@ -1 +0,0 @@
-__version__ = "0.8.7"

nimcode-0.8.7/src/nimcode/model_registry.py

@@ -1,55 +0,0 @@
-"""
-Model registry with context window sizes and metadata.
-Used for auto-compact thresholds and dynamic model listing.
-"""
-import logging
-
-logger = logging.getLogger(__name__)
-
-# Known NVIDIA NIM models and their context window sizes (in tokens)
-# This acts as a fallback + enrichment layer when API is unavailable
-MODEL_CONTEXT_WINDOWS = {
-    # Llama 3.1 family
-    "meta/llama-3.1-8b-instruct": 128000,
-    "meta/llama-3.3-70b-instruct": 128000,
-    "meta/llama-3.1-405b-instruct": 128000,
-    # Llama 3.2 family
-    "meta/llama-3.2-1b-instruct": 128000,
-    "meta/llama-3.2-3b-instruct": 128000,
-    "meta/llama-3.2-11b-vision-instruct": 128000,
-    "meta/llama-3.2-90b-vision-instruct": 128000,
-    # Llama 3.3 family
-    "meta/llama-3.3-70b-instruct": 128000,
-    # Nemotron
-    "nvidia/nemotron-4-340b-instruct": 4096,
-    "nvidia/llama-3.1-nemotron-70b-instruct": 128000,
-    # Mistral
-    "mistralai/mixtral-8x22b-instruct-v0.1": 65536,
-    "mistralai/mistral-large-2-instruct": 128000,
-    # DeepSeek
-    "deepseek-ai/deepseek-r1": 128000,
-    "deepseek-ai/deepseek-v4-pro": 128000,
-    # Qwen
-    "qwen/qwen2.5-72b-instruct": 128000,
-    "qwen/qwq-32b": 128000,
-}
-
-# Default context window for unknown models
-DEFAULT_CONTEXT_WINDOW = 8192
-
-# Hardcoded fallback model list (used when API is unreachable)
-FALLBACK_MODELS = [
-    "meta/llama-3.3-70b-instruct",
-    "meta/llama-3.1-8b-instruct",
-    "meta/llama-3.1-405b-instruct",
-    "meta/llama-3.3-70b-instruct",
-    "nvidia/llama-3.1-nemotron-70b-instruct",
-    "deepseek-ai/deepseek-r1",
-    "mistralai/mixtral-8x22b-instruct-v0.1",
-    "qwen/qwen2.5-72b-instruct",
-]
-
-
-def get_context_window(model_id: str) -> int:
-    """Returns the context window size for a given model ID."""
-    return MODEL_CONTEXT_WINDOWS.get(model_id, DEFAULT_CONTEXT_WINDOW)

nimcode-0.8.7/tests/test_model_registry.py

@@ -1,38 +0,0 @@
-"""Tests for the model_registry module."""
-import pytest
-from nimcode.model_registry import (
-    MODEL_CONTEXT_WINDOWS,
-    FALLBACK_MODELS,
-    DEFAULT_CONTEXT_WINDOW,
-    get_context_window,
-)
-
-
-def test_get_context_window_known_model():
-    """Known models should return their specific context window size."""
-    assert get_context_window("meta/llama-3.3-70b-instruct") == 128000
-    assert get_context_window("nvidia/nemotron-4-340b-instruct") == 4096
-    assert get_context_window("mistralai/mixtral-8x22b-instruct-v0.1") == 65536
-
-
-def test_get_context_window_unknown_model():
-    """Unknown models should return the default context window."""
-    assert get_context_window("some/unknown-model") == DEFAULT_CONTEXT_WINDOW
-    assert get_context_window("") == DEFAULT_CONTEXT_WINDOW
-
-
-def test_fallback_models_not_empty():
-    """Fallback model list should have entries."""
-    assert len(FALLBACK_MODELS) > 0
-
-
-def test_fallback_models_all_in_context_windows():
-    """All fallback models should have a known context window entry."""
-    for model in FALLBACK_MODELS:
-        assert model in MODEL_CONTEXT_WINDOWS, f"Fallback model {model} missing from context windows"
-
-
-def test_context_windows_positive():
-    """All context window values should be positive integers."""
-    for model, ctx in MODEL_CONTEXT_WINDOWS.items():
-        assert isinstance(ctx, int) and ctx > 0, f"Invalid context window for {model}: {ctx}"