Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
|
@@ -1,38 +1,49 @@
|
|
| 1 |
import os
|
| 2 |
import subprocess
|
|
|
|
| 3 |
import gradio as gr
|
| 4 |
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
|
| 5 |
import torch
|
| 6 |
from huggingface_hub import HfApi
|
| 7 |
|
| 8 |
-
# ---------------------------------------------------------
|
| 9 |
-
# 1. モデル準備 (2026年最新 hf コマンド対応)
|
| 10 |
-
# ---------------------------------------------------------
|
| 11 |
def prepare_model():
|
| 12 |
model_path = "./ni_v1_model"
|
|
|
|
|
|
|
| 13 |
token = os.getenv("HF_TOKEN")
|
| 14 |
|
| 15 |
if not os.path.exists(model_path):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 16 |
print("🚀 NI-v1 マージ開始...")
|
| 17 |
env = os.environ.copy()
|
| 18 |
if token:
|
| 19 |
env["HF_TOKEN"] = token
|
| 20 |
-
# 2026年最新の hf コマンドを使用
|
| 21 |
try:
|
| 22 |
subprocess.run(["hf", "auth", "login", "--token", token], check=True)
|
| 23 |
except:
|
| 24 |
-
|
| 25 |
-
print("⚠️ hf auth login 失敗。環境変数のみで続行します。")
|
| 26 |
|
| 27 |
try:
|
|
|
|
|
|
|
| 28 |
subprocess.run(
|
| 29 |
-
["mergekit-yaml", "config.yaml", model_path,
|
|
|
|
|
|
|
|
|
|
| 30 |
check=True,
|
| 31 |
env=env
|
| 32 |
)
|
| 33 |
-
print("✨ マージ成功。")
|
| 34 |
except:
|
| 35 |
-
raise RuntimeError("マージ失敗。設定を
|
| 36 |
|
| 37 |
print("🧠 NI-v1 ロード中...")
|
| 38 |
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
|
|
@@ -44,7 +55,6 @@ def prepare_model():
|
|
| 44 |
)
|
| 45 |
return pipeline("text-generation", model=model, tokenizer=tokenizer)
|
| 46 |
|
| 47 |
-
# 起動
|
| 48 |
try:
|
| 49 |
pipe = prepare_model()
|
| 50 |
except Exception as e:
|
|
@@ -57,21 +67,16 @@ def predict(message, history):
|
|
| 57 |
outputs = pipe(prompt, max_new_tokens=512, do_sample=True, temperature=0.7)
|
| 58 |
return outputs[0]['generated_text'].split("assistant\n")[-1].replace("<|im_end|>", "")
|
| 59 |
|
| 60 |
-
# ---------------------------------------------------------
|
| 61 |
-
# 2. UI (Gradio バージョン差異を吸収)
|
| 62 |
-
# ---------------------------------------------------------
|
| 63 |
with gr.Blocks(title="NI-v1.0") as demo:
|
| 64 |
gr.Markdown("# 🤖 Noppo-Intelligence v1.0")
|
| 65 |
|
| 66 |
with gr.Tab("チャット"):
|
| 67 |
-
# TypeError 回避: type 引数を使わず、デフォルト設定で起動
|
| 68 |
gr.ChatInterface(fn=predict)
|
| 69 |
|
| 70 |
with gr.Tab("公開"):
|
| 71 |
-
gr.Markdown("### 完成した NI-v1 を Hugging Face にアップロード")
|
| 72 |
repo_id = gr.Textbox(label="Repo ID", value="noppodev/NoppoIntelligence")
|
| 73 |
user_token = gr.Textbox(label="Write Token", type="password")
|
| 74 |
-
pub_btn = gr.Button("アップロード
|
| 75 |
status = gr.Textbox(label="Status")
|
| 76 |
|
| 77 |
def upload(r, t):
|
|
|
|
| 1 |
import os
|
| 2 |
import subprocess
|
| 3 |
+
import shutil
|
| 4 |
import gradio as gr
|
| 5 |
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
|
| 6 |
import torch
|
| 7 |
from huggingface_hub import HfApi
|
| 8 |
|
|
|
|
|
|
|
|
|
|
| 9 |
def prepare_model():
|
| 10 |
model_path = "./ni_v1_model"
|
| 11 |
+
# Hugging Faceのキャッシュディレクトリ(ここがパンクの元)
|
| 12 |
+
cache_path = os.path.expanduser("~/.cache/huggingface/hub")
|
| 13 |
token = os.getenv("HF_TOKEN")
|
| 14 |
|
| 15 |
if not os.path.exists(model_path):
|
| 16 |
+
print("🧹 ストレージ確保のため、古い残骸を掃除するぜ...")
|
| 17 |
+
# 以前のマージ失敗作があれば削除
|
| 18 |
+
if os.path.exists(model_path):
|
| 19 |
+
shutil.rmtree(model_path)
|
| 20 |
+
|
| 21 |
+
# もし容量がギリギリならキャッシュも消す(再ダウンロードになるけど背に腹は代えられない)
|
| 22 |
+
# shutil.rmtree(cache_path, ignore_errors=True)
|
| 23 |
+
|
| 24 |
print("🚀 NI-v1 マージ開始...")
|
| 25 |
env = os.environ.copy()
|
| 26 |
if token:
|
| 27 |
env["HF_TOKEN"] = token
|
|
|
|
| 28 |
try:
|
| 29 |
subprocess.run(["hf", "auth", "login", "--token", token], check=True)
|
| 30 |
except:
|
| 31 |
+
print("⚠️ ログインスキップ(環境変数で続行)")
|
|
|
|
| 32 |
|
| 33 |
try:
|
| 34 |
+
# --lazy-unpickle: メモリとディスク消費を抑える魔法の引数
|
| 35 |
+
# --low-cpu-mem: さらに負荷を減らす
|
| 36 |
subprocess.run(
|
| 37 |
+
["mergekit-yaml", "config.yaml", model_path,
|
| 38 |
+
"--allow-crimes",
|
| 39 |
+
"--lazy-unpickle",
|
| 40 |
+
"--low-cpu-mem"],
|
| 41 |
check=True,
|
| 42 |
env=env
|
| 43 |
)
|
| 44 |
+
print("✨ マージ成功。のっぽ、耐えたぜ!")
|
| 45 |
except:
|
| 46 |
+
raise RuntimeError("マージ失敗。容量か設定を見直してくれ。")
|
| 47 |
|
| 48 |
print("🧠 NI-v1 ロード中...")
|
| 49 |
tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)
|
|
|
|
| 55 |
)
|
| 56 |
return pipeline("text-generation", model=model, tokenizer=tokenizer)
|
| 57 |
|
|
|
|
| 58 |
try:
|
| 59 |
pipe = prepare_model()
|
| 60 |
except Exception as e:
|
|
|
|
| 67 |
outputs = pipe(prompt, max_new_tokens=512, do_sample=True, temperature=0.7)
|
| 68 |
return outputs[0]['generated_text'].split("assistant\n")[-1].replace("<|im_end|>", "")
|
| 69 |
|
|
|
|
|
|
|
|
|
|
| 70 |
with gr.Blocks(title="NI-v1.0") as demo:
|
| 71 |
gr.Markdown("# 🤖 Noppo-Intelligence v1.0")
|
| 72 |
|
| 73 |
with gr.Tab("チャット"):
|
|
|
|
| 74 |
gr.ChatInterface(fn=predict)
|
| 75 |
|
| 76 |
with gr.Tab("公開"):
|
|
|
|
| 77 |
repo_id = gr.Textbox(label="Repo ID", value="noppodev/NoppoIntelligence")
|
| 78 |
user_token = gr.Textbox(label="Write Token", type="password")
|
| 79 |
+
pub_btn = gr.Button("アップロード")
|
| 80 |
status = gr.Textbox(label="Status")
|
| 81 |
|
| 82 |
def upload(r, t):
|