Skip to content

Commit f4c63f8

Browse files
No cache baseline
1 parent e62918c commit f4c63f8

1 file changed

Lines changed: 12 additions & 27 deletions

File tree

benchmarks/benchmark.py

Lines changed: 12 additions & 27 deletions
Original file line numberDiff line numberDiff line change
@@ -91,9 +91,6 @@
9191
from vcache.vcache_core.cache.eviction_policy.eviction_policy import EvictionPolicy
9292
from vcache.vcache_core.cache.eviction_policy.strategies.scu import SCUEvictionPolicy
9393
from vcache.vcache_core.similarity_evaluator import SimilarityEvaluator
94-
from vcache.vcache_core.similarity_evaluator.strategies.benchmark_comparison import (
95-
BenchmarkComparisonSimilarityEvaluator,
96-
)
9794
from vcache.vcache_core.similarity_evaluator.strategies.string_comparison import (
9895
StringComparisonSimilarityEvaluator,
9996
)
@@ -167,6 +164,12 @@ class LargeLanguageModel(Enum):
167164

168165
LLAMA_3_8B = ("response_llama_3_8b", "Llama_3_8B_Instruct", "float16", None)
169166
LLAMA_3_70B = ("response_llama_3_70b", "Llama_3_70B_Instruct", "float16", None)
167+
LLAMA_3_70B_VLLM = (
168+
"response_llama_3_70b_vllm",
169+
"Llama_3_70B_Instruct_VLLM",
170+
"float16",
171+
None,
172+
)
170173
GPT_4O_MINI = ("response_gpt-4o-mini", "GPT-4o-mini", "float16", None)
171174
GPT_4O_NANO = ("response_gpt-4.1-nano", "GPT-4.1-nano", "float16", None)
172175
GPT_4_1 = ("response_gpt-4.1", "gpt-4.1-2025-04-14", "float16", None)
@@ -194,6 +197,7 @@ class Baseline(Enum):
194197
IID = "iid"
195198
SigmoidProbability = "SigmoidProbability"
196199
SigmoidOnly = "SigmoidOnly"
200+
NoCache = "NoCache"
197201

198202

199203
class Dataset(Enum):
@@ -242,38 +246,19 @@ class GeneratePlotsOnly(Enum):
242246
int,
243247
]
244248
] = [
245-
(
246-
EmbeddingModel.E5_LARGE_V2,
247-
LargeLanguageModel.GPT_4O_MINI,
248-
Dataset.SEM_BENCHMARK_ARENA,
249-
GeneratePlotsOnly.YES,
250-
BenchmarkComparisonSimilarityEvaluator(),
251-
SCUEvictionPolicy(max_size=100000, watermark=0.99, eviction_percentage=0.1),
252-
60000,
253-
),
254249
(
255250
EmbeddingModel.GTE,
256-
LargeLanguageModel.LLAMA_3_8B,
251+
LargeLanguageModel.LLAMA_3_70B_VLLM,
257252
Dataset.SEM_BENCHMARK_CLASSIFICATION,
258-
GeneratePlotsOnly.YES,
253+
GeneratePlotsOnly.NO,
259254
StringComparisonSimilarityEvaluator(),
260255
SCUEvictionPolicy(max_size=100000, watermark=0.99, eviction_percentage=0.1),
261-
45000,
262-
),
263-
(
264-
EmbeddingModel.GTE,
265-
LargeLanguageModel.LLAMA_3_8B,
266-
Dataset.SEM_BENCHMARK_SEARCH_QUERIES,
267-
GeneratePlotsOnly.YES,
268-
StringComparisonSimilarityEvaluator(),
269-
SCUEvictionPolicy(max_size=100000, watermark=0.99, eviction_percentage=0.1),
270-
150000,
271-
),
256+
60000,
257+
)
272258
]
273259

274260
BASELINES_TO_RUN: List[Baseline] = [
275-
Baseline.SigmoidProbability,
276-
Baseline.SigmoidOnly,
261+
Baseline.VCacheLocal,
277262
]
278263

279264
STATIC_THRESHOLDS: List[float] = [0.80, 0.83, 0.86, 0.89, 0.92, 0.95, 0.97, 0.98, 0.99]

0 commit comments

Comments
 (0)