types: expose prompt_eval_cached_count in BaseGenerateResponse (#779)

ollama/_types.py

@@ -249,6 +249,9 @@ class BaseGenerateResponse(SubscriptableBaseModel):
   prompt_eval_count: Optional[int] = None
   'Number of tokens evaluated in the prompt.'
 
+  prompt_eval_cached_count: Optional[int] = None
+  'Number of prompt tokens served from the prompt cache.'
+
   prompt_eval_duration: Optional[int] = None
   'Duration of evaluating the prompt in nanoseconds.'
 

tests/test_client.py

@@ -62,6 +62,36 @@ def test_client_chat(httpserver: HTTPServer):
   assert response['message']['content'] == "I don't know."
 
 
+def test_client_chat_with_prompt_eval_cached_count(httpserver: HTTPServer):
+  httpserver.expect_ordered_request(
+    '/api/chat',
+    method='POST',
+    json={
+      'model': 'dummy',
+      'messages': [{'role': 'user', 'content': 'Hello'}],
+      'tools': [],
+      'stream': False,
+    },
+  ).respond_with_json(
+    {
+      'model': 'dummy',
+      'message': {
+        'role': 'assistant',
+        'content': 'Hi!',
+      },
+      'prompt_eval_count': 10,
+      'prompt_eval_cached_count': 8,
+    }
+  )
+
+  client = Client(httpserver.url_for('/'))
+  response = client.chat('dummy', messages=[{'role': 'user', 'content': 'Hello'}])
+  assert response['prompt_eval_count'] == 10
+  assert response['prompt_eval_cached_count'] == 8
+  assert response.prompt_eval_cached_count == 8
+
+
+
 def test_client_chat_with_logprobs(httpserver: HTTPServer):
   httpserver.expect_ordered_request(
     '/api/chat',

tests/test_type_serialization.py

@@ -4,7 +4,7 @@
 
 import pytest
 
-from ollama._types import ChatRequest, CreateRequest, GenerateRequest, Image
+from ollama._types import ChatRequest, ChatResponse, CreateRequest, GenerateRequest, GenerateResponse, Image
 
 
 def test_image_serialization_bytes():
@@ -122,3 +122,26 @@ def test_think_boolean_serialization():
   assert ChatRequest(model='test-model', think=True).model_dump(exclude_none=True)['think'] is True
   assert ChatRequest(model='test-model', think=False).model_dump(exclude_none=True)['think'] is False
   assert 'think' not in ChatRequest(model='test-model', think=None).model_dump(exclude_none=True)
+
+
+def test_response_prompt_eval_cached_count():
+  gen_resp = GenerateResponse(model='test-model', prompt_eval_count=100, prompt_eval_cached_count=80)
+  assert gen_resp.prompt_eval_cached_count == 80
+  assert gen_resp['prompt_eval_cached_count'] == 80
+  assert gen_resp.model_dump(exclude_none=True)['prompt_eval_cached_count'] == 80
+
+  chat_resp = ChatResponse(
+    model='test-model',
+    message={'role': 'assistant', 'content': 'hi'},
+    prompt_eval_count=100,
+    prompt_eval_cached_count=80,
+  )
+  assert chat_resp.prompt_eval_cached_count == 80
+  assert chat_resp['prompt_eval_cached_count'] == 80
+  assert chat_resp.model_dump(exclude_none=True)['prompt_eval_cached_count'] == 80
+
+  default_resp = GenerateResponse(model='test-model')
+  assert default_resp.prompt_eval_cached_count is None
+  assert 'prompt_eval_cached_count' not in default_resp
+  assert 'prompt_eval_cached_count' not in default_resp.model_dump(exclude_none=True)
+