litellm-haystack 1.1.1__tar.gz → 1.2.0__tar.gz

litellm_haystack-1.2.0/CHANGELOG.md

@@ -0,0 +1,31 @@
+# Changelog
+
+## [integrations/litellm-v1.1.2] - 2026-07-15
+
+### 🧹 Chores
+
+- Cap litellm below 1.92.0
+
+
+## [integrations/litellm-v1.1.1] - 2026-07-15
+
+### 🧹 Chores
+
+- Support sync streaming callbacks in async contexts for Haystack 2.x/3.x compatibility (#3534)
+- Update litellm dependency version in pyproject.toml (#3599)
+
+
+## [integrations/litellm-v1.1.0] - 2026-06-04
+
+### 🚀 Features
+
+- LiteLLM - accept str as ChatGenerator input (#3393)
+
+
+## [integrations/litellm-v1.0.0] - 2026-05-29
+
+### 🚀 Features
+
+- Add LiteLLM integration with LiteLLMChatGenerator component (#3257)
+
+<!-- generated by git-cliff -->

{litellm_haystack-1.1.1 → litellm_haystack-1.2.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: litellm-haystack
-Version: 1.1.1
+Version: 1.2.0
 Summary: An integration of LiteLLM AI gateway into the Haystack framework, routing to 100+ LLM providers.
 Project-URL: Documentation, https://github.com/deepset-ai/haystack-core-integrations/tree/main/integrations/litellm#readme
 Project-URL: Issues, https://github.com/deepset-ai/haystack-core-integrations/issues
@@ -20,7 +20,7 @@ Classifier: Programming Language :: Python :: 3.14
 Classifier: Programming Language :: Python :: Implementation :: CPython
 Requires-Python: >=3.10
 Requires-Dist: haystack-ai>=2.30.0
-Requires-Dist: litellm<2.0,>=1.84.0
+Requires-Dist: litellm<1.92.0,>=1.84.0
 Description-Content-Type: text/markdown
 
 # litellm-haystack

{litellm_haystack-1.1.1 → litellm_haystack-1.2.0}/pyproject.toml

@@ -22,7 +22,9 @@ classifiers = [
   "Programming Language :: Python :: 3.14",
   "Programming Language :: Python :: Implementation :: CPython",
 ]
-dependencies = ["haystack-ai>=2.30.0", "litellm>=1.84.0,<2.0"]
+# litellm capped below 1.92.0: that release ships cp310-cp313-only compiled wheels
+# (source build fails on Python 3.14) and imports fastapi (proxy extra) on any tool call
+dependencies = ["haystack-ai>=2.30.0", "litellm>=1.84.0,<1.92.0"]
 
 [project.urls]
 Documentation = "https://github.com/deepset-ai/haystack-core-integrations/tree/main/integrations/litellm#readme"

{litellm_haystack-1.1.1 → litellm_haystack-1.2.0}/src/haystack_integrations/components/generators/litellm/chat/chat_generator.py

@@ -266,16 +266,12 @@ _FINISH_REASON_MAPPING: dict[str, FinishReason] = {
 }
 
 
-def _extract_usage(obj: Any) -> dict[str, int]:
+def _extract_usage(obj: Any) -> dict[str, Any]:
     """Pull token usage off a litellm response or chunk, tolerating its absence."""
     usage = getattr(obj, "usage", None)
-    if not usage:
-        return {}
-    return {
-        "prompt_tokens": getattr(usage, "prompt_tokens", 0),
-        "completion_tokens": getattr(usage, "completion_tokens", 0),
-        "total_tokens": getattr(usage, "total_tokens", 0),
-    }
+    if usage is not None and hasattr(usage, "model_dump"):
+        return usage.model_dump(exclude_none=True)
+    return {}
 
 
 def _build_chat_message(response: Any, choice: Any) -> ChatMessage:

{litellm_haystack-1.1.1 → litellm_haystack-1.2.0}/tests/test_litellm_chat_generator.py

@@ -13,6 +13,7 @@ import pytest
 from haystack.dataclasses import ChatMessage, StreamingChunk
 from haystack.tools import Tool
 from haystack.utils.auth import Secret
+from litellm.types.utils import Usage
 
 from haystack_integrations.components.generators.litellm import LiteLLMChatGenerator
 
@@ -29,14 +30,9 @@ def _make_mock_response(content="Hello!", model="openai/gpt-4o", tool_calls=None
     choice.index = 0
     choice.finish_reason = "stop"
 
-    usage = MagicMock()
-    usage.prompt_tokens = 10
-    usage.completion_tokens = 5
-    usage.total_tokens = 15
-
     resp = MagicMock()
     resp.choices = [choice]
-    resp.usage = usage
+    resp.usage = Usage(prompt_tokens=10, completion_tokens=5, total_tokens=15)
     resp.model = model
     return resp
 
@@ -235,6 +231,30 @@ class TestRun:
             meta = result["replies"][0].meta
             assert meta["usage"]["prompt_tokens"] == 10
             assert meta["usage"]["completion_tokens"] == 5
+            assert set(meta["usage"]) == {"prompt_tokens", "completion_tokens", "total_tokens"}
+
+    def test_cache_tokens_in_meta(self):
+        gen = LiteLLMChatGenerator(model="anthropic/claude-sonnet-4-20250514")
+        mock_resp = _make_mock_response(model="anthropic/claude-sonnet-4-20250514")
+        mock_resp.usage = Usage(
+            prompt_tokens=2010,
+            completion_tokens=5,
+            total_tokens=2015,
+            cache_creation_input_tokens=1200,
+            cache_read_input_tokens=800,
+        )
+
+        fake_litellm = types.ModuleType("litellm")
+        fake_litellm.completion = MagicMock(return_value=mock_resp)
+
+        with mock.patch.dict(sys.modules, {"litellm": fake_litellm}):
+            usage = gen.run(messages=[ChatMessage.from_user("hi")])["replies"][0].meta["usage"]
+            assert usage["cache_creation_input_tokens"] == 1200
+            assert usage["cache_read_input_tokens"] == 800
+            assert usage["prompt_tokens"] == 2010
+            # litellm also normalizes cache counts into the OpenAI-style nested breakdown
+            assert usage["prompt_tokens_details"]["cached_tokens"] == 800
+            assert usage["prompt_tokens_details"]["cache_creation_tokens"] == 1200
 
     def test_tool_calls_parsed(self):
         tc = MagicMock()
@@ -258,7 +278,7 @@ class TestRun:
 
     def test_streaming_content_accumulated(self):
         gen = LiteLLMChatGenerator(model="openai/gpt-4o")
-        usage = SimpleNamespace(prompt_tokens=10, completion_tokens=5, total_tokens=15)
+        usage = Usage(prompt_tokens=10, completion_tokens=5, total_tokens=15)
         chunks = _make_mock_stream_chunks("Hi!", usage=usage)
 
         fake_litellm = types.ModuleType("litellm")
@@ -452,7 +472,7 @@ class TestAsync:
     @pytest.mark.asyncio
     async def test_run_async_streaming(self):
         gen = LiteLLMChatGenerator(model="openai/gpt-4o")
-        usage = SimpleNamespace(prompt_tokens=10, completion_tokens=5, total_tokens=15)
+        usage = Usage(prompt_tokens=10, completion_tokens=5, total_tokens=15)
         chunks = _make_mock_stream_chunks("Hi!", usage=usage)
 
         async def _aiter():

litellm_haystack-1.1.1/CHANGELOG.md

@@ -1,16 +0,0 @@
-# Changelog
-
-## [integrations/litellm-v1.1.0] - 2026-06-04
-
-### 🚀 Features
-
-- LiteLLM - accept str as ChatGenerator input (#3393)
-
-
-## [integrations/litellm-v1.0.0] - 2026-05-29
-
-### 🚀 Features
-
-- Add LiteLLM integration with LiteLLMChatGenerator component (#3257)
-
-<!-- generated by git-cliff -->