fix: switch to HF default cache, remove MODELS_DIR and unused import, update CTranslate2 download to include all required files
Browse files- sema_translation_api.py +22 -17
sema_translation_api.py
CHANGED
@@ -13,7 +13,7 @@ from typing import Optional
|
|
13 |
from fastapi import FastAPI, HTTPException, Request
|
14 |
from fastapi.middleware.cors import CORSMiddleware
|
15 |
from pydantic import BaseModel, Field
|
16 |
-
from huggingface_hub import hf_hub_download
|
17 |
import ctranslate2
|
18 |
import sentencepiece as spm
|
19 |
import fasttext
|
@@ -49,7 +49,6 @@ app.add_middleware(
|
|
49 |
|
50 |
# --- Global Variables ---
|
51 |
REPO_ID = "sematech/sema-utils"
|
52 |
-
MODELS_DIR = "hf_models"
|
53 |
beam_size = 1
|
54 |
device = "cpu"
|
55 |
|
@@ -71,38 +70,44 @@ def get_nairobi_time():
|
|
71 |
return full_date, curr_time
|
72 |
|
73 |
def download_models():
|
74 |
-
"""Download models from HuggingFace Hub"""
|
75 |
print("π Downloading models from sematech/sema-utils...")
|
76 |
|
77 |
-
# Ensure models directory exists
|
78 |
-
os.makedirs(MODELS_DIR, exist_ok=True)
|
79 |
-
|
80 |
try:
|
81 |
-
# Download individual files from root
|
82 |
print("π₯ Downloading SentencePiece model...")
|
83 |
spm_path = hf_hub_download(
|
84 |
repo_id=REPO_ID,
|
85 |
-
filename="spm.model"
|
86 |
-
local_dir=MODELS_DIR
|
87 |
)
|
88 |
|
89 |
print("π₯ Downloading language detection model...")
|
90 |
ft_path = hf_hub_download(
|
91 |
repo_id=REPO_ID,
|
92 |
-
filename="lid218e.bin"
|
93 |
-
local_dir=MODELS_DIR
|
94 |
)
|
95 |
|
96 |
-
# Download translation model
|
97 |
print("π₯ Downloading translation model (3.3B)...")
|
98 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
99 |
repo_id=REPO_ID,
|
100 |
-
|
101 |
-
local_dir=MODELS_DIR
|
102 |
)
|
103 |
|
104 |
-
#
|
105 |
-
ct_model_full_path = os.path.
|
106 |
|
107 |
return spm_path, ft_path, ct_model_full_path
|
108 |
|
|
|
13 |
from fastapi import FastAPI, HTTPException, Request
|
14 |
from fastapi.middleware.cors import CORSMiddleware
|
15 |
from pydantic import BaseModel, Field
|
16 |
+
from huggingface_hub import hf_hub_download
|
17 |
import ctranslate2
|
18 |
import sentencepiece as spm
|
19 |
import fasttext
|
|
|
49 |
|
50 |
# --- Global Variables ---
|
51 |
REPO_ID = "sematech/sema-utils"
|
|
|
52 |
beam_size = 1
|
53 |
device = "cpu"
|
54 |
|
|
|
70 |
return full_date, curr_time
|
71 |
|
72 |
def download_models():
|
73 |
+
"""Download models from HuggingFace Hub using default cache"""
|
74 |
print("π Downloading models from sematech/sema-utils...")
|
75 |
|
|
|
|
|
|
|
76 |
try:
|
77 |
+
# Download individual files from root (using default HF cache)
|
78 |
print("π₯ Downloading SentencePiece model...")
|
79 |
spm_path = hf_hub_download(
|
80 |
repo_id=REPO_ID,
|
81 |
+
filename="spm.model"
|
|
|
82 |
)
|
83 |
|
84 |
print("π₯ Downloading language detection model...")
|
85 |
ft_path = hf_hub_download(
|
86 |
repo_id=REPO_ID,
|
87 |
+
filename="lid218e.bin"
|
|
|
88 |
)
|
89 |
|
90 |
+
# Download translation model files individually
|
91 |
print("π₯ Downloading translation model (3.3B)...")
|
92 |
+
|
93 |
+
# Download all necessary CTranslate2 files
|
94 |
+
model_bin_path = hf_hub_download(
|
95 |
+
repo_id=REPO_ID,
|
96 |
+
filename="translation_models/sematrans-3.3B/model.bin"
|
97 |
+
)
|
98 |
+
|
99 |
+
hf_hub_download(
|
100 |
+
repo_id=REPO_ID,
|
101 |
+
filename="translation_models/sematrans-3.3B/config.json"
|
102 |
+
)
|
103 |
+
|
104 |
+
hf_hub_download(
|
105 |
repo_id=REPO_ID,
|
106 |
+
filename="translation_models/sematrans-3.3B/shared_vocabulary.txt"
|
|
|
107 |
)
|
108 |
|
109 |
+
# The model directory is the parent of the model.bin file
|
110 |
+
ct_model_full_path = os.path.dirname(model_bin_path)
|
111 |
|
112 |
return spm_path, ft_path, ct_model_full_path
|
113 |
|