Skip to content

Drop in a screenshot and convert it to clean code (HTML/Tailwind/React/Vue)

Python

log cache hit rate

Abi Raja536a59a76db5
diff --git a/backend/agent/providers/anthropic.py b/backend/agent/providers/anthropic.py
index 2685c61d..ce7d01e8 100644
--- a/backend/agent/providers/anthropic.py
+++ b/backend/agent/providers/anthropic.py
@@ -346,10 +346,11 @@ class AnthropicProviderSession(ProviderSession):
model_name = self._model.value
pricing = MODEL_PRICING.get(model_name)
cost_str = f" cost=${u.cost(pricing):.4f}" if pricing else ""
+ cache_hit_rate_str = f" cache_hit_rate={u.cache_hit_rate_percent():.2f}%"
print(
f"[TOKEN USAGE] provider=anthropic model={model_name} | "
f"input={u.input} output={u.output} "
f"cache_read={u.cache_read} cache_write={u.cache_write} "
- f"total={u.total}{cost_str}"
+ f"total={u.total}{cache_hit_rate_str}{cost_str}"
)
await self._client.close()
diff --git a/backend/agent/providers/gemini.py b/backend/agent/providers/gemini.py
index 71515bb3..73bb088e 100644
--- a/backend/agent/providers/gemini.py
+++ b/backend/agent/providers/gemini.py
@@ -346,9 +346,10 @@ class GeminiProviderSession(ProviderSession):
model_name = _get_gemini_api_model_name(self._model)
pricing = MODEL_PRICING.get(model_name)
cost_str = f" cost=${u.cost(pricing):.4f}" if pricing else ""
+ cache_hit_rate_str = f" cache_hit_rate={u.cache_hit_rate_percent():.2f}%"
print(
f"[TOKEN USAGE] provider=gemini model={model_name} | "
f"input={u.input} output={u.output} "
f"cache_read={u.cache_read} cache_write={u.cache_write} "
- f"total={u.total}{cost_str}"
+ f"total={u.total}{cache_hit_rate_str}{cost_str}"
)
diff --git a/backend/agent/providers/openai.py b/backend/agent/providers/openai.py
index e1a3cdae..92275301 100644
--- a/backend/agent/providers/openai.py
+++ b/backend/agent/providers/openai.py
@@ -468,10 +468,11 @@ class OpenAIProviderSession(ProviderSession):
model_name = get_openai_api_name(self._model)
pricing = MODEL_PRICING.get(model_name)
cost_str = f" cost=${u.cost(pricing):.4f}" if pricing else ""
+ cache_hit_rate_str = f" cache_hit_rate={u.cache_hit_rate_percent():.2f}%"
print(
f"[TOKEN USAGE] provider=openai model={model_name} | "
f"input={u.input} output={u.output} "
f"cache_read={u.cache_read} cache_write={u.cache_write} "
- f"total={u.total}{cost_str}"
+ f"total={u.total}{cache_hit_rate_str}{cost_str}"
)
await self._client.close()
diff --git a/backend/agent/providers/token_usage.py b/backend/agent/providers/token_usage.py
index ad1ef999..2141a3c4 100644
--- a/backend/agent/providers/token_usage.py
+++ b/backend/agent/providers/token_usage.py
@@ -25,7 +25,7 @@ class TokenUsage:
total: All tokens as reported by the provider API. Equals
input + cache_read + output (+ thinking for Gemini).
- Total input sent to the model = input + cache_read.
+ Total input sent to the model = input + cache_read + cache_write.
Cost = (input * input_rate + output * output_rate
+ cache_read * cache_read_rate + cache_write * cache_write_rate)
/ 1_000_000
@@ -52,3 +52,14 @@ class TokenUsage:
+ self.cache_read * pricing.cache_read
+ self.cache_write * pricing.cache_write
) / 1_000_000
+
+ def total_input_tokens(self) -> int:
+ """All input tokens, including non-cached, cache-read, and cache-write."""
+ return self.input + self.cache_read + self.cache_write
+
+ def cache_hit_rate_percent(self) -> float:
+ """Percent of total input tokens served from cache."""
+ total_input = self.total_input_tokens()
+ if total_input == 0:
+ return 0.0
+ return (self.cache_read / total_input) * 100.0
diff --git a/backend/tests/test_token_usage.py b/backend/tests/test_token_usage.py
index a02ac007..66cf674e 100644
--- a/backend/tests/test_token_usage.py
+++ b/backend/tests/test_token_usage.py
@@ -86,6 +86,23 @@ class TestCost:
assert abs(usage.cost(pricing) - expected) < 1e-9
+class TestCacheHitRate:
+ def test_zero_total_input_is_zero_percent(self) -> None:
+ usage = TokenUsage()
+ assert usage.total_input_tokens() == 0
+ assert usage.cache_hit_rate_percent() == 0.0
+
+ def test_cache_hit_rate_without_cache_write(self) -> None:
+ usage = TokenUsage(input=300, cache_read=100)
+ assert usage.total_input_tokens() == 400
+ assert abs(usage.cache_hit_rate_percent() - 25.0) < 1e-9
+
+ def test_cache_hit_rate_includes_cache_write_in_denominator(self) -> None:
+ usage = TokenUsage(input=300, cache_read=100, cache_write=100)
+ assert usage.total_input_tokens() == 500
+ assert abs(usage.cache_hit_rate_percent() - 20.0) < 1e-9
+
+
# ---------------------------------------------------------------------------
# Gemini: _extract_usage
# ---------------------------------------------------------------------------