update GPU config

config.py

@@ -13,13 +13,25 @@ OUTPUT_DIR = os.path.join(BASE_DIR, "outputs")
 SUBJECT_DATA_FILE = os.path.join(DATA_DIR, "subjectData.csv")
 SAMPLE_DATA_FILE = os.path.join(DATA_DIR, "sampleData.csv")
 STANDARD_NAME_MAP_DATA_FILE = os.path.join(DATA_DIR, "standardData.csv")
-# Model Names
-MODEL_NAME = "Detomo/cl-nagoya-sup-simcse-ja-nss-v_1_0_6"
-SENTENCE_EMBEDDING_FILE = os.path.join(
-    DATA_DIR,
-    "anchor_name_sentence_embeddings_cl-nagoya-sup-simcse-ja-nss-v_1_0_6.pkl",
-)
-# SENTENCE_EMBEDDING_FILE = None
+
+# Model Config
+HALF = "float16"
 MODEL_TYPE = None
 DEVICE_TYPE = "cpu"
-QINT8 = False
+MODEL_NAME = "Detomo/cl-nagoya-sup-simcse-ja-nss-v_1_0_6"
+if HALF == "float16":
+    SENTENCE_EMBEDDING_FILE = os.path.join(
+        DATA_DIR,
+        "anchor_name_sentence_embeddings_cl-nagoya-sup-simcse-ja-nss-v_1_0_6_float16.pkl",
+    )
+elif HALF == "bfloat16":
+    SENTENCE_EMBEDDING_FILE = os.path.join(
+        DATA_DIR,
+        "anchor_name_sentence_embeddings_cl-nagoya-sup-simcse-ja-nss-v_1_0_6_bfloat16.pkl",
+    )
+else:
+    SENTENCE_EMBEDDING_FILE = os.path.join(
+        DATA_DIR,
+        "anchor_name_sentence_embeddings_cl-nagoya-sup-simcse-ja-nss-v_1_0_6.pkl",
+    )
+# SENTENCE_EMBEDDING_FILE = None

data/anchor_name_sentence_embeddings_cl-nagoya-sup-simcse-ja-nss-v1_0_2.pkl

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:51d2c6c83a577d6afb5701de785405f950a5fcd4c9013ca1dfab53f7d9e5739d
-size 18373796

data/anchor_name_sentence_embeddings_cl-nagoya-sup-simcse-ja-nss-v_1_0_3.pkl

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:482862ee6138c60ce27604aee19f66ee6d2835cfd08907145d9df65d65829259
-size 19547300

data/anchor_name_sentence_embeddings_cl-nagoya-sup-simcse-ja-nss-v_1_0_5.pkl

@@ -1,3 +0,0 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:c82797adfac6a639b7566fdd424f876273177e4ebb4a5279889d73e41630b7bd
-size 21209252

data/{anchor_name_sentence_embeddings_cl-nagoya-sup-simcse-ja-nss-v0_9_13_.pkl → anchor_name_sentence_embeddings_cl-nagoya-sup-simcse-ja-nss-v_1_0_6_bfloat16.pkl}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1209c010eda0150d859ee2b3c9a889ab2bb6f69a65adafbae9ec8f7ffa813962
-size 12251300
+oid sha256:042dcf97e2e686a956e3311e897e1707e330ff1413bf3310cd8ea8b0393abf4c
+size 21888164

data/{anchor_name_sentence_embeddings_cl-nagoya-sup-simcse-ja-nss-v0_9_17.pkl → anchor_name_sentence_embeddings_cl-nagoya-sup-simcse-ja-nss-v_1_0_6_float16.pkl}

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ae5fab42d01409975e99f20e4907e42e770191ba212c9ccfcc6fe88f6db37ff0
-size 16392356
+oid sha256:a7eb117306729208293b84057ca01d8fa6238dc5be1f69c3ad452ba84fa6c6bf
+size 10944164

prepare.py

@@ -27,13 +27,16 @@ def export_model(backend="onnx", use_qint8=False):
                 model, "avx512_vnni", MODEL_NAME, push_to_hub=True
             )
         else:
-            model.push_to_hub(MODEL_NAME)
+            model.push_to_hub(MODEL_NAME, create_pr=True)
     else:
         raise ValueError(f"Invalid backend: {backend}")
 
 
 # Export all combinations
-backend = "openvino"
-use_qint8 = False
-print(f"Exporting {backend} model with QINT8={use_qint8}")
-export_model(backend=backend, use_qint8=use_qint8)
+# backend = "onnx"
+# use_qint8 = False
+# print(f"Exporting {backend} model with QINT8={use_qint8}")
+# export_model(backend=backend, use_qint8=use_qint8)
+
+# import tensorrt as trt
+# print(trt.__version__)

services/sentence_transformer_service.py

@@ -1,8 +1,8 @@
 import pickle
 from config import (
     MODEL_NAME, MODEL_TYPE, DEVICE_TYPE,
-    SENTENCE_EMBEDDING_FILE,  
-    STANDARD_NAME_MAP_DATA_FILE, SUBJECT_DATA_FILE, DATA_DIR, QINT8
+    SENTENCE_EMBEDDING_FILE,    
+    STANDARD_NAME_MAP_DATA_FILE, SUBJECT_DATA_FILE, DATA_DIR, HALF
 )
 from sentence_transformer_lib.sentence_transformer_helper import SentenceTransformerHelper
 from data_lib.subject_data import SubjectData
@@ -25,11 +25,11 @@ class SentenceTransformerService:
 
         print("Loading models and data...")
         # Load sentence transformer model
-        print(f"Loading model {MODEL_NAME} with type {MODEL_TYPE} and qint8={QINT8}")
+        print(f"Loading model {MODEL_NAME} with type {MODEL_TYPE} and half={HALF}")
         self.sentenceTransformerHelper = SentenceTransformerHelper(
             model_name=MODEL_NAME, 
             model_type=MODEL_TYPE,
-            qint8=QINT8
+            half=HALF
         )
         # Load standard subject dictionary
         self.dic_standard_subject = SubjectData.create_standard_subject_dic_from_file(SUBJECT_DATA_FILE)