Spaces:
Sleeping
Sleeping
Sushwetabm
commited on
Commit
Β·
ebe67a1
1
Parent(s):
c16d4e7
updated model.py
Browse files
model.py
CHANGED
@@ -1,21 +1,21 @@
|
|
1 |
-
#
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
|
10 |
-
|
11 |
|
12 |
-
#
|
13 |
-
|
14 |
-
|
15 |
-
|
16 |
-
|
17 |
|
18 |
-
|
19 |
# def get_model_config():
|
20 |
# """Cache model configuration"""
|
21 |
# return {
|
@@ -27,167 +27,102 @@
|
|
27 |
# "low_cpu_mem_usage": True,
|
28 |
# "use_cache": True,
|
29 |
# }
|
30 |
-
|
31 |
-
# def load_model_sync():
|
32 |
-
# """Synchronous model loading with optimizations"""
|
33 |
-
# global _tokenizer, _model, _model_loaded
|
34 |
-
|
35 |
-
# if _model_loaded:
|
36 |
-
# return _tokenizer, _model
|
37 |
-
|
38 |
-
# config = get_model_config()
|
39 |
-
# model_id = config["model_id"]
|
40 |
-
|
41 |
-
# logger.info(f"π§ Loading model {model_id}...")
|
42 |
-
|
43 |
-
# try:
|
44 |
-
# # Set cache directory to avoid re-downloading
|
45 |
-
# cache_dir = os.environ.get("TRANSFORMERS_CACHE", "./model_cache")
|
46 |
-
# os.makedirs(cache_dir, exist_ok=True)
|
47 |
-
|
48 |
-
# # Load tokenizer first (faster)
|
49 |
-
# logger.info("π Loading tokenizer...")
|
50 |
-
# _tokenizer = AutoTokenizer.from_pretrained(
|
51 |
-
# model_id,
|
52 |
-
# trust_remote_code=config["trust_remote_code"],
|
53 |
-
# cache_dir=cache_dir,
|
54 |
-
# use_fast=True, # Use fast tokenizer if available
|
55 |
-
# )
|
56 |
-
|
57 |
-
# # Load model with optimizations
|
58 |
-
# logger.info("π§ Loading model...")
|
59 |
-
# _model = AutoModelForCausalLM.from_pretrained(
|
60 |
-
# model_id,
|
61 |
-
# trust_remote_code=config["trust_remote_code"],
|
62 |
-
# torch_dtype=config["torch_dtype"],
|
63 |
-
# device_map=config["device_map"],
|
64 |
-
# low_cpu_mem_usage=config["low_cpu_mem_usage"],
|
65 |
-
# cache_dir=cache_dir,
|
66 |
-
# offload_folder="offload",
|
67 |
-
# offload_state_dict=True
|
68 |
-
# )
|
69 |
-
|
70 |
-
# # Set to evaluation mode
|
71 |
-
# _model.eval()
|
72 |
-
|
73 |
-
# _model_loaded = True
|
74 |
-
# logger.info("β
Model loaded successfully!")
|
75 |
-
# return _tokenizer, _model
|
76 |
-
|
77 |
-
# except Exception as e:
|
78 |
-
# logger.error(f"β Failed to load model: {e}")
|
79 |
-
# raise
|
80 |
-
|
81 |
-
# async def load_model_async():
|
82 |
-
# """Asynchronous model loading"""
|
83 |
-
# global _model_loading
|
84 |
-
|
85 |
-
# if _model_loaded:
|
86 |
-
# return _tokenizer, _model
|
87 |
-
|
88 |
-
# if _model_loading:
|
89 |
-
# # Wait for ongoing loading to complete
|
90 |
-
# while _model_loading and not _model_loaded:
|
91 |
-
# await asyncio.sleep(0.1)
|
92 |
-
# return _tokenizer, _model
|
93 |
-
|
94 |
-
# _model_loading = True
|
95 |
-
|
96 |
-
# try:
|
97 |
-
# # Run model loading in thread pool to avoid blocking
|
98 |
-
# loop = asyncio.get_event_loop()
|
99 |
-
# with ThreadPoolExecutor(max_workers=1) as executor:
|
100 |
-
# tokenizer, model = await loop.run_in_executor(
|
101 |
-
# executor, load_model_sync
|
102 |
-
# )
|
103 |
-
# return tokenizer, model
|
104 |
-
# finally:
|
105 |
-
# _model_loading = False
|
106 |
-
|
107 |
-
# def get_model():
|
108 |
-
# """Get the loaded model (for synchronous access)"""
|
109 |
-
# if not _model_loaded:
|
110 |
-
# return load_model_sync()
|
111 |
-
# return _tokenizer, _model
|
112 |
-
|
113 |
-
# def is_model_loaded():
|
114 |
-
# """Check if model is loaded"""
|
115 |
-
# return _model_loaded
|
116 |
-
|
117 |
-
# def get_model_info():
|
118 |
-
# """Get model information without loading"""
|
119 |
-
# config = get_model_config()
|
120 |
-
# return {
|
121 |
-
# "model_id": config["model_id"],
|
122 |
-
# "loaded": _model_loaded,
|
123 |
-
# "loading": _model_loading,
|
124 |
-
# }
|
125 |
-
|
126 |
-
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
|
127 |
-
from functools import lru_cache
|
128 |
-
import logging
|
129 |
-
import asyncio
|
130 |
-
logger = logging.getLogger(__name__)
|
131 |
-
_model_loaded = False
|
132 |
-
_tokenizer = None
|
133 |
-
_model = None
|
134 |
-
@lru_cache(maxsize=1)
|
135 |
def get_model_config():
|
136 |
return {
|
137 |
"model_id": "Salesforce/codet5p-220m",
|
138 |
"trust_remote_code": True
|
139 |
}
|
140 |
-
|
141 |
def load_model_sync():
|
|
|
142 |
global _tokenizer, _model, _model_loaded
|
143 |
-
|
144 |
if _model_loaded:
|
145 |
return _tokenizer, _model
|
146 |
-
|
147 |
config = get_model_config()
|
148 |
model_id = config["model_id"]
|
149 |
-
|
|
|
|
|
150 |
try:
|
151 |
-
|
152 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
153 |
_model.eval()
|
|
|
154 |
_model_loaded = True
|
|
|
155 |
return _tokenizer, _model
|
156 |
-
|
157 |
except Exception as e:
|
158 |
logger.error(f"β Failed to load model: {e}")
|
159 |
raise
|
160 |
|
161 |
-
|
162 |
async def load_model_async():
|
163 |
-
|
|
|
|
|
164 |
if _model_loaded:
|
165 |
-
return
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
|
|
|
|
|
|
|
|
|
|
|
170 |
try:
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
|
|
179 |
|
180 |
def get_model():
|
|
|
181 |
if not _model_loaded:
|
182 |
-
|
183 |
return _tokenizer, _model
|
184 |
|
185 |
def is_model_loaded():
|
|
|
186 |
return _model_loaded
|
187 |
|
188 |
def get_model_info():
|
|
|
|
|
189 |
return {
|
190 |
-
"model_id":
|
191 |
"loaded": _model_loaded,
|
192 |
-
"loading":
|
193 |
}
|
|
|
1 |
+
# model.py - Optimized version
|
2 |
+
from transformers import AutoTokenizer, AutoModelForCausalLM
|
3 |
+
import torch
|
4 |
+
from functools import lru_cache
|
5 |
+
import os
|
6 |
+
import asyncio
|
7 |
+
from concurrent.futures import ThreadPoolExecutor
|
8 |
+
import logging
|
9 |
|
10 |
+
logger = logging.getLogger(__name__)
|
11 |
|
12 |
+
# Global variables to store loaded model
|
13 |
+
_tokenizer = None
|
14 |
+
_model = None
|
15 |
+
_model_loading = False
|
16 |
+
_model_loaded = False
|
17 |
|
18 |
+
@lru_cache(maxsize=1)
|
19 |
# def get_model_config():
|
20 |
# """Cache model configuration"""
|
21 |
# return {
|
|
|
27 |
# "low_cpu_mem_usage": True,
|
28 |
# "use_cache": True,
|
29 |
# }
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
30 |
def get_model_config():
|
31 |
return {
|
32 |
"model_id": "Salesforce/codet5p-220m",
|
33 |
"trust_remote_code": True
|
34 |
}
|
|
|
35 |
def load_model_sync():
|
36 |
+
"""Synchronous model loading with optimizations"""
|
37 |
global _tokenizer, _model, _model_loaded
|
38 |
+
|
39 |
if _model_loaded:
|
40 |
return _tokenizer, _model
|
41 |
+
|
42 |
config = get_model_config()
|
43 |
model_id = config["model_id"]
|
44 |
+
|
45 |
+
logger.info(f"π§ Loading model {model_id}...")
|
46 |
+
|
47 |
try:
|
48 |
+
# Set cache directory to avoid re-downloading
|
49 |
+
cache_dir = os.environ.get("TRANSFORMERS_CACHE", "./model_cache")
|
50 |
+
os.makedirs(cache_dir, exist_ok=True)
|
51 |
+
|
52 |
+
# Load tokenizer first (faster)
|
53 |
+
logger.info("π Loading tokenizer...")
|
54 |
+
_tokenizer = AutoTokenizer.from_pretrained(
|
55 |
+
model_id,
|
56 |
+
trust_remote_code=config["trust_remote_code"],
|
57 |
+
cache_dir=cache_dir,
|
58 |
+
use_fast=True, # Use fast tokenizer if available
|
59 |
+
)
|
60 |
+
|
61 |
+
# Load model with optimizations
|
62 |
+
logger.info("π§ Loading model...")
|
63 |
+
_model = AutoModelForCausalLM.from_pretrained(
|
64 |
+
model_id,
|
65 |
+
trust_remote_code=config["trust_remote_code"],
|
66 |
+
torch_dtype=config["torch_dtype"],
|
67 |
+
device_map=config["device_map"],
|
68 |
+
low_cpu_mem_usage=config["low_cpu_mem_usage"],
|
69 |
+
cache_dir=cache_dir,
|
70 |
+
offload_folder="offload",
|
71 |
+
offload_state_dict=True
|
72 |
+
)
|
73 |
+
|
74 |
+
# Set to evaluation mode
|
75 |
_model.eval()
|
76 |
+
|
77 |
_model_loaded = True
|
78 |
+
logger.info("β
Model loaded successfully!")
|
79 |
return _tokenizer, _model
|
80 |
+
|
81 |
except Exception as e:
|
82 |
logger.error(f"β Failed to load model: {e}")
|
83 |
raise
|
84 |
|
|
|
85 |
async def load_model_async():
|
86 |
+
"""Asynchronous model loading"""
|
87 |
+
global _model_loading
|
88 |
+
|
89 |
if _model_loaded:
|
90 |
+
return _tokenizer, _model
|
91 |
+
|
92 |
+
if _model_loading:
|
93 |
+
# Wait for ongoing loading to complete
|
94 |
+
while _model_loading and not _model_loaded:
|
95 |
+
await asyncio.sleep(0.1)
|
96 |
+
return _tokenizer, _model
|
97 |
+
|
98 |
+
_model_loading = True
|
99 |
+
|
100 |
try:
|
101 |
+
# Run model loading in thread pool to avoid blocking
|
102 |
+
loop = asyncio.get_event_loop()
|
103 |
+
with ThreadPoolExecutor(max_workers=1) as executor:
|
104 |
+
tokenizer, model = await loop.run_in_executor(
|
105 |
+
executor, load_model_sync
|
106 |
+
)
|
107 |
+
return tokenizer, model
|
108 |
+
finally:
|
109 |
+
_model_loading = False
|
110 |
|
111 |
def get_model():
|
112 |
+
"""Get the loaded model (for synchronous access)"""
|
113 |
if not _model_loaded:
|
114 |
+
return load_model_sync()
|
115 |
return _tokenizer, _model
|
116 |
|
117 |
def is_model_loaded():
|
118 |
+
"""Check if model is loaded"""
|
119 |
return _model_loaded
|
120 |
|
121 |
def get_model_info():
|
122 |
+
"""Get model information without loading"""
|
123 |
+
config = get_model_config()
|
124 |
return {
|
125 |
+
"model_id": config["model_id"],
|
126 |
"loaded": _model_loaded,
|
127 |
+
"loading": _model_loading,
|
128 |
}
|