← back to Exo
Initialize inference engine session in base class
98edb393b2c67ca3645ae16348bbc3ffc42de69e · 2024-12-08 02:34:31 -0800 · Nel Nibcord
Files touched
M exo/inference/inference_engine.pyM exo/inference/mlx/sharded_inference_engine.pyM exo/inference/tinygrad/inference.py
Diff
commit 98edb393b2c67ca3645ae16348bbc3ffc42de69e
Author: Nel Nibcord <blindcrone@tuta.io>
Date: Sun Dec 8 02:34:31 2024 -0800
Initialize inference engine session in base class
---
exo/inference/inference_engine.py | 2 ++
exo/inference/mlx/sharded_inference_engine.py | 1 -
exo/inference/tinygrad/inference.py | 1 -
3 files changed, 2 insertions(+), 2 deletions(-)
diff --git a/exo/inference/inference_engine.py b/exo/inference/inference_engine.py
index 2ab5f540..ad14bf8c 100644
--- a/exo/inference/inference_engine.py
+++ b/exo/inference/inference_engine.py
@@ -8,6 +8,8 @@ from .shard import Shard
class InferenceEngine(ABC):
+ session = {}
+
@abstractmethod
async def encode(self, shard: Shard, prompt: str) -> np.ndarray:
pass
diff --git a/exo/inference/mlx/sharded_inference_engine.py b/exo/inference/mlx/sharded_inference_engine.py
index a0fe7d2d..3ba8370e 100644
--- a/exo/inference/mlx/sharded_inference_engine.py
+++ b/exo/inference/mlx/sharded_inference_engine.py
@@ -42,7 +42,6 @@ class MLXDynamicShardInferenceEngine(InferenceEngine):
self.shard_downloader = shard_downloader
self.executor = ThreadPoolExecutor(max_workers=1)
self.caches = OrderedDict()
- self.session = {}
async def poll_state(self, request_id: str, max_caches=2):
if request_id in self.caches:
diff --git a/exo/inference/tinygrad/inference.py b/exo/inference/tinygrad/inference.py
index f8503ba0..343deec3 100644
--- a/exo/inference/tinygrad/inference.py
+++ b/exo/inference/tinygrad/inference.py
@@ -65,7 +65,6 @@ class TinygradDynamicShardInferenceEngine(InferenceEngine):
self.shard_downloader = shard_downloader
self.executor = ThreadPoolExecutor(max_workers=1)
self.states = OrderedDict()
- self.session = {}
def poll_state(self, x, request_id: str, max_states=2):
if request_id not in self.states:
← bcf87e79 Okay let's turn no_grad back on. We'll worry about that when
·
back to Exo
·
Some session method cleanup b22c21ac →