From ac77a5ca5ad5d8719238d21b29635b8c02006ef2 Mon Sep 17 00:00:00 2001
From: Yule100 <chenyukun11@h-partners.com>
Date: Thu, 24 Jul 2025 11:25:03 +0800
Subject: [PATCH] =?UTF-8?q?telechat2=20mcore=20=E9=80=82=E9=85=8D?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 .../model_executor/models/mf_models/config.py | 15 +++++--
 .../models/mf_models/telechat2.py             | 45 +++++++++++++++++++
 .../model_executor/models/registry.py         |  1 +
 3 files changed, 58 insertions(+), 3 deletions(-)
 create mode 100644 vllm_mindspore/model_executor/models/mf_models/telechat2.py

diff --git a/vllm_mindspore/model_executor/models/mf_models/config.py b/vllm_mindspore/model_executor/models/mf_models/config.py
index dd11efc27..ae9c259cb 100644
--- a/vllm_mindspore/model_executor/models/mf_models/config.py
+++ b/vllm_mindspore/model_executor/models/mf_models/config.py
@@ -48,12 +48,12 @@ MODEL_COMMON_MAPPING = {
     'attention_dropout': ('model_config.hf_config.attention_dropout', None),
     'hidden_act': ('model_config.hf_config.hidden_act', None),
     'hidden_size': ('model_config.hf_config.hidden_size', None),
-    'intermediate_size': ('model_config.hf_config.intermediate_size', None),
+    #'intermediate_size': ('model_config.hf_config.intermediate_size', None),
     'max_position_embeddings':
     ('model_config.hf_config.max_position_embeddings', None),
     'num_attention_heads':
     ('model_config.hf_config.num_attention_heads', None),
-    'rms_norm_eps': ('model_config.hf_config.rms_norm_eps', None),
+    #'rms_norm_eps': ('model_config.hf_config.rms_norm_eps', None),
     'num_hidden_layers': ('model_config.hf_config.num_hidden_layers', None),
     'num_layers': ('model_config.hf_config.num_layers', None),
     'num_key_value_heads':
@@ -78,6 +78,9 @@ MODEL_COMMON_MAPPING = {
     'router_aux_loss_coef': ('model_config.hf_config.router_aux_loss_coef',
                              None),
     'mlp_only_layers': ('model_config.hf_config.mlp_only_layers', None),
+    'n_layer': ('model_config.hf_config.n_layer', None),
+    'n_head': ('model_config.hf_config.n_head', None),
+    'ffn_hidden_size': ('model_config.hf_config.ffn_hidden_size', None),
 }
 
 # model default config
@@ -90,7 +93,13 @@ MODEL_RELATED_MAPPING = {
         'layernorm_compute_dtype': 'bfloat16',
         'rotary_dtype': 'bfloat16',
         'router_dense_type': 'bfloat16',
-    }
+    },
+    'telechat': {
+        'params_dtype': 'float16',  # need an input
+        'layernorm_compute_dtype': 'float32',
+        'rotary_dtype': 'float16',
+        'compute_dtype': 'float16',
+    },
     # Add anther model type...
 }
 
diff --git a/vllm_mindspore/model_executor/models/mf_models/telechat2.py b/vllm_mindspore/model_executor/models/mf_models/telechat2.py
new file mode 100644
index 000000000..1374f66b8
--- /dev/null
+++ b/vllm_mindspore/model_executor/models/mf_models/telechat2.py
@@ -0,0 +1,45 @@
+# SPDX-License-Identifier: Apache-2.0
+
+# Copyright 2025 Huawei Technologies Co., Ltd.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+from mindformers.models.telechat2.configuration_telechat2 import Telechat2Config
+from mindformers.models.telechat2.modeling_telechat2 import (  # noqa
+    Telechat2ForCausalLM as Telechat2ForCausalLM_MF)
+from mindspore.nn.utils import no_init_parameters
+from vllm.config import VllmConfig
+from vllm.logger import init_logger
+
+from vllm_mindspore.model_executor.models.mf_models.config import (
+    gen_model_config)
+from vllm_mindspore.model_executor.models.mf_models.qwen3 import (
+    Qwen3ForCausalLM)
+
+logger = init_logger(__name__)
+
+
+class TeleChat2ForCausalLM(Qwen3ForCausalLM):
+
+    def __init__(self, *, vllm_config: VllmConfig, prefix: str = "") -> None:
+        super().__init__(vllm_config=vllm_config, prefix=prefix)
+
+    def _generate_model_config(self):
+        self.mf_model_config = gen_model_config(self.mf_config, Telechat2Config)
+        logger.debug("=====mf_model_config====\n%s", self.mf_model_config)
+
+    def _create_network(self):
+        # Initial network
+        with no_init_parameters():  # Delay initialization
+            network = Telechat2ForCausalLM_MF(self.mf_model_config)
+        return network, network.model.output_layer
diff --git a/vllm_mindspore/model_executor/models/registry.py b/vllm_mindspore/model_executor/models/registry.py
index ed7e5ceb5..1a0ebba4f 100644
--- a/vllm_mindspore/model_executor/models/registry.py
+++ b/vllm_mindspore/model_executor/models/registry.py
@@ -31,6 +31,7 @@ _MINDFORMERS_MODELS = {
     "Qwen2ForCausalLM": ("qwen2", "Qwen2ForCausalLM"),
     "Qwen3ForCausalLM": ("qwen3", "Qwen3ForCausalLM"),  # MCore
     "Qwen3MoeForCausalLM": ("qwen3_moe", "Qwen3MoeForCausalLM"),  # MCore
+    "TeleChat2ForCausalLM": ("telechat2", "TeleChat2ForCausalLM"),  # MCore
     "DeepseekV3ForCausalLM": ("deepseek_v3", "DeepseekV3ForCausalLM"),
     "DeepSeekMTPModel": ("deepseek_mtp", "DeepseekV3MTPForCausalLM"),
 }
-- 
Gitee