From 0d95e50df53959a92dc010c441d03eab1d209636 Mon Sep 17 00:00:00 2001
From: yumoqing <yumoqing@gmail.com>
Date: Fri, 8 Aug 2025 11:27:00 +0800
Subject: [PATCH] bugfix

---
 llmengine/gptoss.py | 41 +++++++++++++++++++++++++++++++++++++++++
 1 file changed, 41 insertions(+)
 create mode 100644 llmengine/gptoss.py

diff --git a/llmengine/gptoss.py b/llmengine/gptoss.py
new file mode 100644
index 0000000..41e4dfd
--- /dev/null
+++ b/llmengine/gptoss.py
@@ -0,0 +1,41 @@
+#!/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 GptossLLM(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=32768,
+			do_sample=True,
+			eos_token_id=self.tokenizer.eos_token_id
+		)
+		return generate_kwargs
+
+	def _messages2inputs(self, messages):
+		inputs = self.tokenizer.apply_chat_template(
+			messages, 
+			add_generation_prompt=True,
+			return_tensors="pt",
+			return_dict=True
+		).to(self.model.device)
+		return inputs
+
+llm_register("gpt-oss", GptossLLM)