[test] update time series test scripts

0092638_seism.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2b94653c6964b630038897a27cb6d276ff866d9ecd1f6419358b9407f0df62e
+size 72128

test_inference_ts.py

@@ -0,0 +1,78 @@
+from pathlib import Path
+import torch
+from transformers import AutoConfig, AutoModelForCausalLM, AutoProcessor
+
+
+model_path = Path(__file__).parent.resolve()
+print(f"Loading model from: {model_path}")
+
+# 加载模型配置
+config = AutoConfig.from_pretrained(model_path, trust_remote_code=True)
+print(f"Model config: {config.model_type}")
+print(f"Architecture: {config.architectures}")
+
+# 加载处理器（tokenizer + image processor + ts processor）
+print("\nLoading processor...")
+processor = AutoProcessor.from_pretrained(model_path, trust_remote_code=True)
+
+# 加载模型（使用 bfloat16 精度和自动设备映射）
+print("\nLoading model...")
+model = AutoModelForCausalLM.from_pretrained(
+    model_path,
+    dtype=torch.bfloat16,
+    device_map="auto",
+    # attn_implementation="flash_attention_2",  #时序暂不支持flash_attn，load加这行会报错
+    trust_remote_code=True
+)
+
+print(f"✓ Model loaded successfully!")
+print(f"Model type: {type(model).__name__}")
+print(f"Model device: {model.device}")
+
+# ============================================================================
+# 测试 3: 时序对话
+# ============================================================================
+print("\n" + "=" * 80)
+print("测试 3: 时序对话")
+print("=" * 80)
+
+messages = [
+    {
+        "role": "user",
+        "content": [
+            {"type": "time_series", "data": "./0092638_seism.npy", "sampling_rate": 100},
+            {"type": "text", "text": "Please determine whether an Earthquake event has occurred in the provided time-series data. If so, please specify the starting time point indices of the P-wave and S-wave in the event."},
+        ],
+    }
+]
+
+time_series_inputs = processor.time_series_preprocessor(messages)
+multimodal_inputs = processor.apply_chat_template(messages, add_generation_prompt=True, tokenize=True, return_dict=True, return_tensors="pt", enable_thinking=False, **time_series_inputs).to(model.device, dtype=torch.bfloat16)
+
+print("\n生成时序回复...")
+with torch.inference_mode():
+    multimodal_generated_ids = model.generate(
+        **multimodal_inputs,
+        max_new_tokens=200,
+        do_sample=False,
+        temperature=1.0,
+    )
+
+# 提取生成的 token（去除输入部分）
+multimodal_generated_ids_trimmed = [
+    out_ids[len(in_ids):] for in_ids, out_ids in zip(multimodal_inputs.input_ids, multimodal_generated_ids)
+]
+
+# 解码为文本
+multimodal_output = processor.batch_decode(
+    multimodal_generated_ids_trimmed,
+    skip_special_tokens=True,
+    clean_up_tokenization_spaces=False
+)
+
+print("\n" + "-" * 80)
+print("时序输出：")
+print("-" * 80)
+print(multimodal_output[0])
+print("-" * 80)
+print("\n✅ 时序功能测试完成！")