bugfix

llmengine/baichuanm2.py

@@ -0,0 +1,40 @@
+#!/share/vllm-0.8.5/bin/python
+
+# pip install accelerate
+from appPublic.worker import awaitify
+from appPublic.log import debug
+from ahserver.serverenv import get_serverenv
+from PIL import Image
+import torch
+from llmengine.base_chat_llm import BaseChatLLM, llm_register
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+class BaichuanM2LLM(BaseChatLLM):
+    def __init__(self, model_id):
+        self.tokenizer = AutoTokenizer.from_pretrained(model_id)
+        self.model = AutoModelForCausalLM.from_pretrained(
+            model_id,
+            torch_dtype="auto",
+            device_map="auto"
+        )
+        self.model_id = model_id
+
+    def build_kwargs(self, inputs, streamer):
+        generate_kwargs = dict(
+            **inputs,
+            streamer=streamer,
+            max_new_tokens=4096
+        )
+        return generate_kwargs
+
+	def _messages2inputs(self, messages):
+		text = tokenizer.apply_chat_template(
+			messages,
+			tokenize=False,
+			add_generation_prompt=True,
+			thinking_mode='on' # on/off/auto 
+		)
+		model_inputs = tokenizer([text], return_tensors="pt").to(model.device)
+		return model_inputs
+
+llm_register('Baichuan-M2', BaichuanM2LLM)

llmengine/qwen3coder.py

@@ -23,6 +23,7 @@ class Qwen3CoderLLM(BaseChatLLM):
 		generate_kwargs = dict(
 			**inputs,
 			streamer=streamer,
+			num_return_sequences=2, # 并行生成 4 个结果
 			# do_sample=True,
 			# eos_token_id=self.tokenizer.eos_token_id
 			max_new_tokens=65536

llmengine/server.py

@@ -8,6 +8,7 @@ from llmengine.gemma3_it import Gemma3LLM
 from llmengine.medgemma3_it import MedgemmaLLM
 from llmengine.qwen3 import Qwen3LLM
 from llmengine.qwen3coder import Qwen3CoderLLM
+from llmengine.baiduanm2 import BaichuanM2LLM
 
 from appPublic.registerfunction import RegisterFunction
 from appPublic.log import debug, exception