updates
Browse files- app.py +0 -123
- results/embeddings_log.jsonl +0 -0
app.py
CHANGED
|
@@ -103,129 +103,6 @@ def get_model_info():
|
|
| 103 |
}
|
| 104 |
|
| 105 |
|
| 106 |
-
# @app.post("/predict")
|
| 107 |
-
# async def predict_malware(files: List[UploadFile] = File(...)):
|
| 108 |
-
# """批次預測惡意軟體家族"""
|
| 109 |
-
# if not files:
|
| 110 |
-
# raise HTTPException(status_code=400, detail="No files provided")
|
| 111 |
-
|
| 112 |
-
# predictions, all_embeddings = [], []
|
| 113 |
-
|
| 114 |
-
# print(f"\n📥 Received {len(files)} files")
|
| 115 |
-
|
| 116 |
-
# try:
|
| 117 |
-
# with torch.no_grad():
|
| 118 |
-
# for idx, file in enumerate(files, 1):
|
| 119 |
-
# try:
|
| 120 |
-
# content = await file.read()
|
| 121 |
-
# asm_code = content.decode("utf-8").strip()
|
| 122 |
-
# if not asm_code:
|
| 123 |
-
# continue
|
| 124 |
-
|
| 125 |
-
# print(f"📄 [{idx}/{len(files)}] {file.filename}")
|
| 126 |
-
|
| 127 |
-
# inputs = tokenizer(
|
| 128 |
-
# asm_code,
|
| 129 |
-
# return_tensors="pt",
|
| 130 |
-
# max_length=512,
|
| 131 |
-
# truncation=True,
|
| 132 |
-
# padding=True
|
| 133 |
-
# ).to(DEVICE)
|
| 134 |
-
|
| 135 |
-
# outputs = model(**inputs)
|
| 136 |
-
# logits = outputs.logits
|
| 137 |
-
# probs = torch.nn.functional.softmax(logits, dim=-1)
|
| 138 |
-
# confidence, predicted_class = torch.max(probs, dim=-1)
|
| 139 |
-
|
| 140 |
-
# predicted_id = predicted_class.item()
|
| 141 |
-
# label_name = model.config.id2label.get(predicted_id, f"class_{predicted_id}")
|
| 142 |
-
|
| 143 |
-
# # embedding
|
| 144 |
-
# if outputs.hidden_states:
|
| 145 |
-
# cls_embedding = outputs.hidden_states[-1][:, 0, :].cpu().numpy()[0]
|
| 146 |
-
# else:
|
| 147 |
-
# cls_embedding = logits.cpu().numpy()[0]
|
| 148 |
-
|
| 149 |
-
# # attention
|
| 150 |
-
# max_attention_score = 0.0
|
| 151 |
-
# if outputs.attentions:
|
| 152 |
-
# last_attention = outputs.attentions[-1].mean(dim=1)
|
| 153 |
-
# cls_attention = last_attention[0, 0, :].cpu().numpy()
|
| 154 |
-
# max_attention_score = float(cls_attention.max())
|
| 155 |
-
|
| 156 |
-
# predictions.append({
|
| 157 |
-
# "filename": file.filename,
|
| 158 |
-
# "predicted_label": label_name,
|
| 159 |
-
# "confidence": float(confidence.item()),
|
| 160 |
-
# "attention_score": max_attention_score
|
| 161 |
-
# })
|
| 162 |
-
# all_embeddings.append({
|
| 163 |
-
# "filename": file.filename,
|
| 164 |
-
# "label": label_name,
|
| 165 |
-
# "embedding": cls_embedding.tolist(),
|
| 166 |
-
# "attention_score": max_attention_score
|
| 167 |
-
# })
|
| 168 |
-
# except Exception as e:
|
| 169 |
-
# print(f"❌ {file.filename} error:", e)
|
| 170 |
-
|
| 171 |
-
# if not predictions:
|
| 172 |
-
# raise HTTPException(status_code=400, detail="No valid files processed")
|
| 173 |
-
|
| 174 |
-
# # 多數投票
|
| 175 |
-
# votes = [p["predicted_label"] for p in predictions]
|
| 176 |
-
# final_label, count = Counter(votes).most_common(1)[0]
|
| 177 |
-
# best_embedding = max(all_embeddings, key=lambda x: x["attention_score"])
|
| 178 |
-
|
| 179 |
-
# result = {
|
| 180 |
-
# "final_label": final_label,
|
| 181 |
-
# "vote_count": count,
|
| 182 |
-
# "total_segments": len(predictions),
|
| 183 |
-
# "confidence": sum(p["confidence"] for p in predictions) / len(predictions),
|
| 184 |
-
# "embedding": {
|
| 185 |
-
# "values": best_embedding["embedding"],
|
| 186 |
-
# "source_file": best_embedding["filename"],
|
| 187 |
-
# "attention_score": best_embedding["attention_score"]
|
| 188 |
-
# },
|
| 189 |
-
# "segment_predictions": predictions
|
| 190 |
-
# }
|
| 191 |
-
|
| 192 |
-
# # 寫入本地 log
|
| 193 |
-
# log_path = os.path.join("results", "embeddings_log.jsonl")
|
| 194 |
-
# with open(log_path, "a") as f:
|
| 195 |
-
# f.write(json.dumps({
|
| 196 |
-
# "label": final_label,
|
| 197 |
-
# "text" : asm_code,
|
| 198 |
-
# "embedding": best_embedding["embedding"],
|
| 199 |
-
# "source": best_embedding["filename"]
|
| 200 |
-
# }) + "\n")
|
| 201 |
-
|
| 202 |
-
# print(f"✅ Prediction complete ({final_label}), embedding logged.")
|
| 203 |
-
# return result
|
| 204 |
-
|
| 205 |
-
# except Exception as e:
|
| 206 |
-
# raise HTTPException(status_code=500, detail=f"Prediction failed: {e}")
|
| 207 |
-
|
| 208 |
-
# from huggingface_hub import HfApi
|
| 209 |
-
|
| 210 |
-
# HF_DATASET_REPO = "lyi029/malvec-embeddings"
|
| 211 |
-
|
| 212 |
-
# def upload_embeddings_to_hub(local_path: str):
|
| 213 |
-
# """將 embeddings_log.jsonl 上傳到 Hugging Face Dataset"""
|
| 214 |
-
# api = HfApi()
|
| 215 |
-
# try:
|
| 216 |
-
# api.upload_file(
|
| 217 |
-
# path_or_fileobj=local_path,
|
| 218 |
-
# path_in_repo="logs/embeddings_log.jsonl",
|
| 219 |
-
# repo_id=HF_DATASET_REPO,
|
| 220 |
-
# repo_type="dataset",
|
| 221 |
-
# token=HF_TOKEN,
|
| 222 |
-
# )
|
| 223 |
-
# print(f"☁️ Uploaded embeddings to dataset: {HF_DATASET_REPO}/logs/embeddings_log.jsonl")
|
| 224 |
-
# return True
|
| 225 |
-
# except Exception as e:
|
| 226 |
-
# print(f"⚠️ Failed to upload embeddings: {e}")
|
| 227 |
-
# return False
|
| 228 |
-
|
| 229 |
HF_DATASET_REPO = "lyi029/malvec-embeddings"
|
| 230 |
|
| 231 |
def upload_embeddings_to_hub(local_path: str):
|
|
|
|
| 103 |
}
|
| 104 |
|
| 105 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 106 |
HF_DATASET_REPO = "lyi029/malvec-embeddings"
|
| 107 |
|
| 108 |
def upload_embeddings_to_hub(local_path: str):
|
results/embeddings_log.jsonl
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|