Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,5 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+model.safetensors.index.json filter=lfs diff=lfs merge=lfs -text
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

.mv

@@ -0,0 +1 @@
+Revision:master,CreatedAt:1746725023

README.md

@@ -0,0 +1,207 @@
+---
+license: apache-2.0
+license_link: https://huggingface.co/Qwen/Qwen3-235B-A22B/blob/main/LICENSE
+pipeline_tag: text-generation
+tags:
+- Qwen3
+- gptq
+- int8
+- 量化修复
+- vLLM
+base_model:
+  - Qwen/Qwen3-235B-A22B
+base_model_relation: quantized
+---
+# 通义千问3-235B-A22B-GPTQ-Int8
+基础型 [Qwen/Qwen3-235B-A22B](https://www.modelscope.cn/models/Qwen/Qwen3-235B-A22B)
+
+### 【模型更新日期】
+``` 
+2025-05-09
+1. 首次commit
+2. 确定支持8卡的`tensor-parallel-size` + `expert-parallel` 启动
+3. 必须 gptq_marlin 启动；不支持 Compute 7 显卡启动：vllm没有实现原生GPTQ的moe模块。
+```
+
+### 【依赖】
+
+```
+vllm==0.8.5
+transformers==4.51.3
+```
+
+<div style="
+    background: rgba(255, 193, 61, 0.15);
+    padding: 16px;
+    border-radius: 6px;
+    border: 1px solid rgba(255, 165, 0, 0.3);
+    margin: 16px 0;
+">
+### 【💡新版 VLLM MoE 注意事项💡】
+
+#### 1. 需使用V0推理模式
+启动vllm之前，先设置环境变量
+```
+export VLLM_USE_V1=0
+```
+
+#### 2. `gptq_marlin.py` 存在小bug，需要补丁
+将附件中的文件替换至
+
+```.../vllm/model_executor/layers/quantization/gptq_marlin.py```
+
+否则会出现下述报错
+```
+    raise NotImplementedError(
+NotImplementedError: Apply router weight on input is not supported forfused Marlin MoE method.
+```
+</div>
+
+<div style="
+    background: rgba(255, 0, 200, 0.15);
+    padding: 16px;
+    border-radius: 6px;
+    border: 1px solid rgba(255, 0, 200, 0.3);
+    margin: 16px 0;
+">
+### 【💡通义千问3-235B-A22B 注意事项💡】
+
+#### 1. 启动vllm的时候，要记得使用专家并行模式(`--enable-expert-parallel`)，否则不能单节点8卡启动。
+启动示例：
+```commandline
+vllm serve \
+    tclf90/Qwen3-235B-A22B-GPTQ-Int8 \
+    --served-model-name Qwen3-235B-A22B-GPTQ-Int8 \
+    --max-num-seqs 8 \
+    --max-model-len 32768 \
+    --max-seq-len-to-capture 32768 \
+    --gpu-memory-utilization 0.98 \
+    --tensor-parallel-size 8 \
+    --enable-expert-parallel \
+    --disable-log-requests \
+    --trust-remote-code
+```
+</div>
+
+
+### 【模型列表】
+
+| 文件大小    | 最近更新时间       |
+|---------|--------------|
+| `226GB` | `2025-05-09` |
+
+
+
+### 【模型下载】
+
+```python
+from modelscope import snapshot_download
+snapshot_download('tclf90/Qwen3-235B-A22B-GPTQ-Int8', cache_dir="本地路径")
+```
+
+
+### 【介绍】
+# Qwen3-235B-A22B
+<a href="https://chat.qwen.ai/" target="_blank" style="margin: 2px;">
+    <img alt="Chat" src="https://img.shields.io/badge/%F0%9F%92%9C%EF%B8%8F%20Qwen%20Chat%20-536af5" style="display: inline-block; vertical-align: middle;"/>
+</a>
+
+## Qwen3 Highlights
+
+Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Built upon extensive training, Qwen3 delivers groundbreaking advancements in reasoning, instruction-following, agent capabilities, and multilingual support, with the following key features:
+
+- **Uniquely support of seamless switching between thinking mode** (for complex logical reasoning, math, and coding) and **non-thinking mode** (for efficient, general-purpose dialogue) **within single model**, ensuring optimal performance across various scenarios.
+- **Significantly enhancement in its reasoning capabilities**, surpassing previous QwQ (in thinking mode) and Qwen2.5 instruct models (in non-thinking mode) on mathematics, code generation, and commonsense logical reasoning.
+- **Superior human preference alignment**, excelling in creative writing, role-playing, multi-turn dialogues, and instruction following, to deliver a more natural, engaging, and immersive conversational experience.
+- **Expertise in agent capabilities**, enabling precise integration with external tools in both thinking and unthinking modes and achieving leading performance among open-source models in complex agent-based tasks.
+- **Support of 100+ languages and dialects** with strong capabilities for **multilingual instruction following** and **translation**.
+
+## Model Overview
+
+**Qwen3-235B-A22B** has the following features:
+- Type: Causal Language Models
+- Training Stage: Pretraining & Post-training
+- Number of Parameters: 235B in total and 22B activated
+- Number of Paramaters (Non-Embedding): 234B
+- Number of Layers: 94
+- Number of Attention Heads (GQA): 64 for Q and 4 for KV
+- Number of Experts: 128
+- Number of Activated Experts: 8
+- Context Length: 32,768 natively and [131,072 tokens with YaRN](#processing-long-texts). 
+
+For more details, including benchmark evaluation, hardware requirements, and inference performance, please refer to our [blog](https://qwenlm.github.io/blog/qwen3/), [GitHub](https://github.com/QwenLM/Qwen3), and [Documentation](https://qwen.readthedocs.io/en/latest/).
+
+## Best Practices
+
+To achieve optimal performance, we recommend the following settings:
+
+1. **Sampling Parameters**:
+   - For thinking mode (`enable_thinking=True`), use `Temperature=0.6`, `TopP=0.95`, `TopK=20`, and `MinP=0`. **DO NOT use greedy decoding**, as it can lead to performance degradation and endless repetitions.
+   - For non-thinking mode (`enable_thinking=False`), we suggest using `Temperature=0.7`, `TopP=0.8`, `TopK=20`, and `MinP=0`.
+   - For supported frameworks, you can adjust the `presence_penalty` parameter between 0 and 2 to reduce endless repetitions. However, using a higher value may occasionally result in language mixing and a slight decrease in model performance.
+
+2. **Adequate Output Length**: We recommend using an output length of 32,768 tokens for most queries. For benchmarking on highly complex problems, such as those found in math and programming competitions, we suggest setting the max output length to 38,912 tokens. This provides the model with sufficient space to generate detailed and comprehensive responses, thereby enhancing its overall performance.
+
+3. **Standardize Output Format**: We recommend using prompts to standardize model outputs when benchmarking.
+   - **Math Problems**: Include "Please reason step by step, and put your final answer within \boxed{}." in the prompt.
+   - **Multiple-Choice Questions**: Add the following JSON structure to the prompt to standardize responses: "Please show your choice in the `answer` field with only the choice letter, e.g., `"answer": "C"`."
+
+4. **No Thinking Content in History**: In multi-turn conversations, the historical model output should only include the final output part and does not need to include the thinking content. It is implemented in the provided chat template in Jinja2. However, for frameworks that do not directly use the Jinja2 chat template, it is up to the developers to ensure that the best practice is followed.
+
+## Processing Long Texts
+
+Qwen3 natively supports context lengths of up to 32,768 tokens. For conversations where the total length (including both input and output) significantly exceeds this limit, we recommend using RoPE scaling techniques to handle long texts effectively. We have validated the model's performance on context lengths of up to 131,072 tokens using the [YaRN](https://arxiv.org/abs/2309.00071) method.
+
+YaRN is currently supported by several inference frameworks, e.g., `transformers` and `llama.cpp` for local use, `vllm` and `sglang` for deployment. In general, there are two approaches to enabling YaRN for supported frameworks:
+
+- Modifying the model files:
+  In the `config.json` file, add the `rope_scaling` fields:
+    ```json
+    {
+        ...,
+        "rope_scaling": {
+            "rope_type": "yarn",
+            "factor": 4.0,
+            "original_max_position_embeddings": 32768
+        }
+    }
+    ```
+  For `llama.cpp`, you need to regenerate the GGUF file after the modification.
+
+- Passing command line arguments:
+
+  For `vllm`, you can use
+    ```shell
+    vllm serve ... --rope-scaling '{"rope_type":"yarn","factor":4.0,"original_max_position_embeddings":32768}' --max-model-len 131072  
+    ```
+
+  For `sglang`, you can use
+    ```shell
+    python -m sglang.launch_server ... --json-model-override-args '{"rope_scaling":{"rope_type":"yarn","factor":4.0,"original_max_position_embeddings":32768}}'
+    ```
+
+  For `llama-server` from `llama.cpp`, you can use
+    ```shell
+    llama-server ... --rope-scaling yarn --rope-scale 4 --yarn-orig-ctx 32768
+    ```
+
+## To completely disable thinking, you could use a custom chat template when starting the model:
+[Vllm Guide](https://qwen.readthedocs.io/en/latest/deployment/vllm.html#thinking-non-thinking-modes)
+
+```
+vllm serve ...model_path... --chat-template ./qwen3_nonthinking.jinja
+```
+
+
+## Citation
+
+If you find our work helpful, feel free to give us a cite.
+
+```
+@misc{qwen3,
+    title  = {Qwen3},
+    url    = {https://qwenlm.github.io/blog/qwen3/},
+    author = {Qwen Team},
+    month  = {April},
+    year   = {2025}
+}

config.json

@@ -0,0 +1,49 @@
+{
+  "name_or_path": "tclf90/Qwen3-235B-A22B-GPTQ-Int8",
+  "architectures": [
+    "Qwen3MoeForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "decoder_sparse_step": 1,
+  "eos_token_id": 151645,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 12288,
+  "max_position_embeddings": 40960,
+  "max_window_layers": 94,
+  "mlp_only_layers": [],
+  "model_type": "qwen3_moe",
+  "moe_intermediate_size": 1536,
+  "norm_topk_prob": true,
+  "num_attention_heads": 64,
+  "num_experts": 128,
+  "num_experts_per_tok": 8,
+  "num_hidden_layers": 94,
+  "num_key_value_heads": 4,
+  "output_router_logits": false,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "router_aux_loss_coef": 0.001,
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float16",
+  "transformers_version": "4.51.0",
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 151936,
+  "quantization_config": {
+    "quant_method": "gptq",
+    "bits": 8,
+    "group_size": 128,
+    "sym": true,
+    "desc_act": false,
+    "block_name_to_quantize": null,
+    "module_name_preceding_first_block": null,
+    "modules_in_block_to_quantize": null
+  }
+}

configuration.json

@@ -0,0 +1 @@
+{"framework": "pytorch", "task": "others", "allow_remote": true}

generation_config.json

@@ -0,0 +1,13 @@
+{
+    "bos_token_id": 151643,
+    "do_sample": true,
+    "eos_token_id": [
+        151645,
+        151643
+    ],
+    "pad_token_id": 151643,
+    "temperature": 0.6,
+    "top_k": 20,
+    "top_p": 0.95,
+    "transformers_version": "4.51.0"
+}

gptq_marlin.py

@@ -0,0 +1,643 @@
+# SPDX-License-Identifier: Apache-2.0
+
+from typing import Any, Callable, Dict, List, Optional, Set, Union
+
+import torch
+
+import vllm.model_executor.layers.fused_moe  # noqa
+from vllm import _custom_ops as ops
+from vllm.logger import init_logger
+from vllm.model_executor.layers.fused_moe.layer import (
+    FusedMoE, FusedMoEMethodBase, FusedMoeWeightScaleSupported)
+from vllm.model_executor.layers.linear import (LinearMethodBase,
+                                               set_weight_attrs)
+from vllm.model_executor.layers.quantization.base_config import (
+    QuantizationConfig, QuantizeMethodBase)
+from vllm.model_executor.layers.quantization.kernels.mixed_precision import (
+    MPLinearLayerConfig, choose_mp_linear_kernel)
+from vllm.model_executor.layers.quantization.utils import replace_parameter
+from vllm.model_executor.layers.quantization.utils.gptq_utils import (
+    get_linear_quant_method)
+from vllm.model_executor.layers.quantization.utils.marlin_utils import (
+    check_marlin_supported, check_moe_marlin_supports_layer,
+    marlin_moe_permute_scales, marlin_repeat_scales_on_all_ranks,
+    verify_marlin_supported)
+from vllm.model_executor.parameter import (ChannelQuantScaleParameter,
+                                           GroupQuantScaleParameter,
+                                           PackedColumnParameter,
+                                           PackedvLLMParameter,
+                                           RowvLLMParameter)
+from vllm.platforms import current_platform
+from vllm.scalar_type import scalar_types
+
+logger = init_logger(__name__)
+
+
+class GPTQMarlinConfig(QuantizationConfig):
+    """Config class for GPTQ Marlin"""
+
+    # (num_bits, is_sym) -> quant_type
+    TYPE_MAP = {
+        (4, True): scalar_types.uint4b8,
+        (8, True): scalar_types.uint8b128,
+    }
+
+    def __init__(self, weight_bits: int, group_size: int, desc_act: bool,
+                 is_sym: bool, lm_head_quantized: bool,
+                 dynamic: Dict[str, Dict[str, Union[int, bool]]],
+                 full_config: Dict[str, Any]) -> None:
+        super().__init__()
+        if desc_act and group_size == -1:
+            # In this case, act_order == True is the same as act_order == False
+            # (since we have only one group per output channel)
+            desc_act = False
+
+        # GPTQModel use `dynamic` config property to allow per module
+        # quantization config so each module can be individually optimized.
+        # Format is Dict[str, Dict] where key is a regex string that can
+        # perform both positive ("+:" prefixed) or negative ("-:" prefixed)
+        # matching of a module.
+        # Default to positive match, override base quant config mode, if no
+        # prefix is used. Value is in dict format of field key and override
+        # value.
+        # Negative matching will skip quantization init for this module
+        # entirely:
+        # non-quantized inference. More details and quantization examples can be
+        # found at: https://github.com/ModelCloud/GPTQModel
+        # Example:
+        #  # last 1/2 of the layers 10-21 has 8bit vs 4bit for 0-9
+        #  # last 1/4 of the layers 16-21 has 8bit and group_size 64
+        # dynamic = {
+        #  #`.*\.` matches the layers_node prefix
+        #  # positive match layer 10-15
+        #  r"+:.*\.(?:1[0-5])\..*": {"bits": 8,},
+        #  # positive match layer 16-21
+        #  r"+:.*\.(?:1[6-9]|20|21)\..*": {"bits": 8, "group_size": 64,},
+        #  r"-:.*\.moe\..*": {}, # negative match (skip) all `moe` layers
+        # }
+        self.dynamic = dynamic
+
+        self.weight_bits = weight_bits
+        self.is_sym = is_sym
+
+        self.pack_factor = 32 // weight_bits  # packed into int32
+        self.group_size = group_size
+        self.desc_act = desc_act
+        self.lm_head_quantized = lm_head_quantized
+        self.full_config = full_config
+
+        if (weight_bits, is_sym) not in self.TYPE_MAP:
+            raise ValueError("Unsupported quantization config: "
+                             f"bits={weight_bits}, sym={is_sym}")
+
+        self.quant_type = self.TYPE_MAP[(weight_bits, is_sym)]
+
+    def __repr__(self) -> str:
+        return (f"GPTQMarlinConfig(quant_type={self.quant_type}, "
+                f"group_size={self.group_size}, "
+                f"desc_act={self.desc_act}, "
+                f"lm_head_quantized={self.lm_head_quantized}), "
+                f"dynamic={self.dynamic}")
+
+    @classmethod
+    def get_name(cls) -> str:
+        return "gptq_marlin"
+
+    @classmethod
+    def get_supported_act_dtypes(cls) -> List[torch.dtype]:
+        return [torch.half, torch.bfloat16]
+
+    @classmethod
+    def get_min_capability(cls) -> int:
+        return 80
+
+    @classmethod
+    def get_config_filenames(cls) -> List[str]:
+        return ["quantize_config.json"]
+
+    @classmethod
+    def from_config(cls, config: Dict[str, Any]) -> "GPTQMarlinConfig":
+        dynamic = cls.get_from_keys_or(config, ["dynamic"], default={})
+        dynamic = {} if dynamic is None else dynamic
+
+        weight_bits = cls.get_from_keys(config, ["bits"])
+        group_size = cls.get_from_keys(config, ["group_size"])
+        desc_act = cls.get_from_keys(config, ["desc_act"])
+        is_sym = cls.get_from_keys(config, ["sym"])
+        lm_head_quantized = cls.get_from_keys_or(config, ["lm_head"],
+                                                 default=False)
+        return cls(weight_bits, group_size, desc_act, is_sym,
+                   lm_head_quantized, dynamic, config)
+
+    @classmethod
+    def override_quantization_method(cls, hf_quant_cfg,
+                                     user_quant) -> Optional[str]:
+        can_convert = cls.is_gptq_marlin_compatible(hf_quant_cfg)
+
+        is_valid_user_quant = (user_quant is None or user_quant == "marlin"
+                               or user_quant == "gptq_marlin")
+
+        if can_convert and is_valid_user_quant:
+            msg = ("The model is convertible to {} during runtime."
+                   " Using {} kernel.".format(cls.get_name(), cls.get_name()))
+            logger.info(msg)
+            return cls.get_name()
+
+        if can_convert and user_quant == "gptq":
+            logger.info("Detected that the model can run with gptq_marlin"
+                        ", however you specified quantization=gptq explicitly,"
+                        " so forcing gptq. Use quantization=gptq_marlin for"
+                        " faster inference")
+        return None
+
+    def get_quant_method(self, layer: torch.nn.Module,
+                         prefix: str) -> Optional["QuantizeMethodBase"]:
+        if isinstance(layer, FusedMoE):
+            from vllm.model_executor.layers.quantization.moe_wna16 import (
+                MoeWNA16Config)
+            if not check_moe_marlin_supports_layer(layer, self.group_size):
+                logger.warning(
+                    f"Layer '{prefix}' is not supported by GPTQMoeMarlin. "
+                    "Falling back to Moe WNA16 kernels.")
+                return MoeWNA16Config.from_config(
+                    self.full_config).get_quant_method(layer, prefix)
+            return GPTQMarlinMoEMethod(self)
+        return get_linear_quant_method(self, layer, prefix,
+                                       GPTQMarlinLinearMethod)
+
+    @classmethod
+    def is_gptq_marlin_compatible(cls, quant_config: Dict[str, Any]):
+        quant_method = quant_config.get("quant_method", "").lower()
+        num_bits = quant_config.get("bits")
+        group_size = quant_config.get("group_size")
+        sym = quant_config.get("sym")
+        desc_act = quant_config.get("desc_act")
+
+        if not current_platform.is_cuda():
+            return False
+
+        if quant_method != "gptq":
+            return False
+
+        # Marlin conversion is only valid if required properties are found
+        if (num_bits is None or group_size is None or sym is None
+                or desc_act is None):
+            return False
+
+        if (num_bits, sym) not in cls.TYPE_MAP:
+            return False
+
+        return check_marlin_supported(quant_type=cls.TYPE_MAP[(num_bits, sym)],
+                                      group_size=group_size)
+
+
+class GPTQMarlinLinearMethod(LinearMethodBase):
+    """Linear method for GPTQ Marlin.
+
+    Args:
+        quant_config: The GPTQ Marlin quantization config.
+    """
+
+    _kernel_backends_being_used: Set[str] = set()
+
+    def __init__(self, quant_config: GPTQMarlinConfig) -> None:
+        self.quant_config = quant_config
+
+        # Verify supported on platform.
+        verify_marlin_supported(quant_type=self.quant_config.quant_type,
+                                group_size=self.quant_config.group_size)
+
+    def create_weights(
+        self,
+        layer: torch.nn.Module,
+        input_size_per_partition: int,
+        output_partition_sizes: List[int],
+        input_size: int,
+        output_size: int,
+        params_dtype: torch.dtype,
+        **extra_weight_attrs,
+    ) -> None:
+        output_size_per_partition = sum(output_partition_sizes)
+        is_row_parallel = input_size != input_size_per_partition
+        weight_loader = extra_weight_attrs.get("weight_loader")
+
+        mp_linear_kernel_config = MPLinearLayerConfig(
+            full_weight_shape=(input_size, output_size),
+            partition_weight_shape=\
+                (input_size_per_partition, output_size_per_partition),
+            weight_type=self.quant_config.quant_type,
+            act_type=params_dtype,
+            group_size=self.quant_config.group_size,
+            zero_points=False,
+            has_g_idx=self.quant_config.desc_act
+        )
+
+        kernel_type = choose_mp_linear_kernel(mp_linear_kernel_config)
+
+        if kernel_type.__name__ not in self._kernel_backends_being_used:
+            logger.info("Using %s for GPTQMarlinLinearMethod",
+                        kernel_type.__name__)
+            self._kernel_backends_being_used.add(kernel_type.__name__)
+
+        # Normalize group_size
+        if self.quant_config.group_size != -1:
+            group_size = self.quant_config.group_size
+        else:
+            group_size = input_size
+
+        # Determine sharding
+        if marlin_repeat_scales_on_all_ranks(self.quant_config.desc_act,
+                                             self.quant_config.group_size,
+                                             is_row_parallel):
+            # By setting scale_dim == None, weight_loader will
+            # repeat the scales on each GPU in TP>1 case.
+            scales_and_zp_input_dim = None
+            scales_and_zp_size = input_size // group_size
+        else:
+            # By setting scale_dim == 0, weight_loader will
+            # shard the scales in TP>1 case.
+            scales_and_zp_input_dim = 0
+            scales_and_zp_size = input_size_per_partition // group_size
+
+        # Quantized weights
+        qweight = PackedvLLMParameter(
+            data=torch.empty(
+                input_size_per_partition // self.quant_config.pack_factor,
+                output_size_per_partition,
+                dtype=torch.int32,
+            ),
+            input_dim=0,
+            output_dim=1,
+            packed_dim=0,
+            packed_factor=self.quant_config.pack_factor,
+            weight_loader=weight_loader)
+
+        # Activation order
+        g_idx = RowvLLMParameter(data=torch.empty(
+            input_size_per_partition,
+            dtype=torch.int32,
+        ),
+                                 input_dim=0,
+                                 weight_loader=weight_loader)
+
+        qzeros_args = {
+            "data":
+            torch.empty(
+                scales_and_zp_size,
+                output_size_per_partition // self.quant_config.pack_factor,
+                dtype=torch.int32,
+            ),
+            "weight_loader":
+            weight_loader
+        }
+        weight_scale_args = {
+            "data":
+            torch.empty(
+                scales_and_zp_size,
+                output_size_per_partition,
+                dtype=params_dtype,
+            ),
+            "weight_loader":
+            weight_loader
+        }
+
+        if scales_and_zp_input_dim is None:
+            scales = ChannelQuantScaleParameter(output_dim=1,
+                                                **weight_scale_args)
+            qzeros = PackedColumnParameter(
+                output_dim=1,
+                packed_dim=1,
+                packed_factor=self.quant_config.pack_factor,
+                **qzeros_args)
+
+        else:
+            scales = GroupQuantScaleParameter(output_dim=1,
+                                              input_dim=0,
+                                              **weight_scale_args)
+            qzeros = PackedvLLMParameter(
+                input_dim=0,
+                output_dim=1,
+                packed_dim=1,
+                packed_factor=self.quant_config.pack_factor,
+                **qzeros_args)
+
+        layer.register_parameter("qweight", qweight)
+        layer.register_parameter("g_idx", g_idx)
+        layer.register_parameter("scales", scales)
+        layer.register_parameter("qzeros", qzeros)
+
+        self.kernel = kernel_type(mp_linear_kernel_config,
+                                  w_q_param_name="qweight",
+                                  w_s_param_name="scales",
+                                  w_zp_param_name="qzeros",
+                                  w_gidx_param_name="g_idx")
+
+    def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
+        self.kernel.process_weights_after_loading(layer)
+
+    def apply(
+        self,
+        layer: torch.nn.Module,
+        x: torch.Tensor,
+        bias: Optional[torch.Tensor] = None,
+    ) -> torch.Tensor:
+        return self.kernel.apply_weights(layer, x, bias)
+
+
+class GPTQMarlinMoEMethod(FusedMoEMethodBase):
+    """MoE Marlin method with quantization."""
+
+    def __init__(self, quant_config: GPTQMarlinConfig) -> None:
+        self.quant_config = quant_config
+
+    def create_weights(
+        self,
+        layer: torch.nn.Module,
+        num_experts: int,
+        hidden_size: int,
+        intermediate_size_per_partition: int,
+        params_dtype: torch.dtype,
+        **extra_weight_attrs,
+    ):
+        intermediate_size_full = extra_weight_attrs.pop(
+            "intermediate_size_full")
+
+        self.is_k_full = (not self.quant_config.desc_act) or (
+            intermediate_size_per_partition == intermediate_size_full)
+
+        if self.quant_config.group_size != -1:
+            scales_size13 = hidden_size // self.quant_config.group_size
+            w2_scales_size = (intermediate_size_full
+                              if self.quant_config.desc_act else
+                              intermediate_size_per_partition)
+            scales_size2 = (w2_scales_size // self.quant_config.group_size)
+            strategy = FusedMoeWeightScaleSupported.GROUP.value
+        else:
+            scales_size13 = 1
+            scales_size2 = 1
+            strategy = FusedMoeWeightScaleSupported.CHANNEL.value
+
+        extra_weight_attrs.update({
+            "quant_method": strategy,
+            "is_transposed": True
+        })
+        # Fused gate_up_proj (column parallel)
+        w13_qweight = torch.nn.Parameter(
+            torch.empty(
+                num_experts,
+                hidden_size // self.quant_config.pack_factor,
+                2 * intermediate_size_per_partition,
+                dtype=torch.int32,
+            ),
+            requires_grad=False,
+        )
+        layer.register_parameter("w13_qweight", w13_qweight)
+        set_weight_attrs(w13_qweight, extra_weight_attrs)
+        # down_proj (row parallel)
+        w2_qweight = torch.nn.Parameter(
+            torch.empty(
+                num_experts,
+                intermediate_size_per_partition //
+                self.quant_config.pack_factor,
+                hidden_size,
+                dtype=torch.int32,
+            ),
+            requires_grad=False,
+        )
+        layer.register_parameter("w2_qweight", w2_qweight)
+        set_weight_attrs(w2_qweight, extra_weight_attrs)
+        # up_proj scales
+        w13_scales = torch.nn.Parameter(
+            torch.empty(num_experts,
+                        scales_size13,
+                        2 * intermediate_size_per_partition,
+                        dtype=params_dtype),
+            requires_grad=False,
+        )
+        layer.register_parameter("w13_scales", w13_scales)
+        set_weight_attrs(w13_scales, extra_weight_attrs)
+        # down_proj scales
+        w2_scales = torch.nn.Parameter(
+            torch.empty(num_experts,
+                        scales_size2,
+                        hidden_size,
+                        dtype=params_dtype),
+            requires_grad=False,
+        )
+        layer.register_parameter("w2_scales", w2_scales)
+        set_weight_attrs(w2_scales, extra_weight_attrs)
+        # dont shard the w2 scales when running act order
+        set_weight_attrs(w2_scales,
+                         {"load_full_w2": self.quant_config.desc_act})
+        # up_proj scales
+        w13_qzeros = torch.nn.Parameter(
+            torch.empty(num_experts,
+                        scales_size13,
+                        2 * intermediate_size_per_partition //
+                        self.quant_config.pack_factor,
+                        dtype=params_dtype),
+            requires_grad=False,
+        )
+        layer.register_parameter("w13_qzeros", w13_qzeros)
+        set_weight_attrs(w13_qzeros, extra_weight_attrs)
+        # down_proj scales
+        w2_qzeros = torch.nn.Parameter(
+            torch.empty(num_experts,
+                        scales_size2,
+                        hidden_size // self.quant_config.pack_factor,
+                        dtype=params_dtype),
+            requires_grad=False,
+        )
+        layer.register_parameter("w2_qzeros", w2_qzeros)
+        set_weight_attrs(w2_qzeros, extra_weight_attrs)
+        # dont shard the w2 scales when running act order
+        set_weight_attrs(w2_qzeros,
+                         {"load_full_w2": self.quant_config.desc_act})
+        w13_g_idx = torch.nn.Parameter(
+            torch.empty(
+                num_experts,
+                hidden_size,
+                dtype=torch.int32,
+            ),
+            requires_grad=False,
+        )
+        layer.register_parameter("w13_g_idx", w13_g_idx)
+        set_weight_attrs(w13_g_idx, extra_weight_attrs)
+        w2_g_idx = torch.nn.Parameter(
+            torch.empty(
+                num_experts,
+                intermediate_size_per_partition,
+                dtype=torch.int32,
+            ),
+            requires_grad=False,
+        )
+        layer.register_parameter("w2_g_idx", w2_g_idx)
+        set_weight_attrs(w2_g_idx, extra_weight_attrs)
+        w13_g_idx_sort_indices = torch.nn.Parameter(
+            torch.empty(
+                num_experts,
+                hidden_size,
+                dtype=torch.int32,
+            ),
+            requires_grad=False,
+        )
+        layer.register_parameter("w13_g_idx_sort_indices",
+                                 w13_g_idx_sort_indices)
+        set_weight_attrs(w13_g_idx_sort_indices, extra_weight_attrs)
+        w2_g_idx_sort_indices = torch.nn.Parameter(
+            torch.empty(
+                num_experts,
+                intermediate_size_per_partition,
+                dtype=torch.int32,
+            ),
+            requires_grad=False,
+        )
+        layer.register_parameter("w2_g_idx_sort_indices",
+                                 w2_g_idx_sort_indices)
+        set_weight_attrs(w2_g_idx_sort_indices, extra_weight_attrs)
+
+        device = layer.w13_qweight.device
+        sms = torch.cuda.get_device_properties(device).multi_processor_count
+        layer.workspace = torch.zeros((sms * 4, ),
+                                      dtype=torch.int,
+                                      device=device,
+                                      requires_grad=False)
+
+    def process_weights_after_loading(self, layer: torch.nn.Module) -> None:
+
+        # Process act_order
+        if self.quant_config.desc_act:
+            # Get sorting based on g_idx
+            num_experts = layer.w13_g_idx.shape[0]
+            w13_g_idx_sort_indices = torch.empty_like(layer.w13_g_idx)
+            w2_g_idx_sort_indices = torch.empty_like(layer.w2_g_idx)
+            w13_sorted_g_idx = torch.empty_like(layer.w13_g_idx)
+            w2_sorted_g_idx = torch.empty_like(layer.w2_g_idx)
+            for e in range(num_experts):
+                w13_g_idx_sort_indices[e] = torch.argsort(
+                    layer.w13_g_idx[e]).to(torch.int32)
+                w2_g_idx_sort_indices[e] = torch.argsort(layer.w2_g_idx[e]).to(
+                    torch.int32)
+                w13_sorted_g_idx[e] = layer.w13_g_idx[e][
+                    w13_g_idx_sort_indices[e]]
+                w2_sorted_g_idx[e] = layer.w2_g_idx[e][
+                    w2_g_idx_sort_indices[e]]
+            replace_parameter(layer, "w13_g_idx", w13_sorted_g_idx)
+            replace_parameter(layer, "w2_g_idx", w2_sorted_g_idx)
+            replace_parameter(layer, "w13_g_idx_sort_indices",
+                              w13_g_idx_sort_indices)
+            replace_parameter(layer, "w2_g_idx_sort_indices",
+                              w2_g_idx_sort_indices)
+        else:
+            # Reset g_idx related tensors
+            num_experts = layer.w13_g_idx.shape[0]
+            device = layer.w13_g_idx.device
+            layer.w13_g_idx = torch.nn.Parameter(
+                torch.empty((num_experts, 0), dtype=torch.int32,
+                            device=device),
+                requires_grad=False,
+            )
+            layer.w2_g_idx = torch.nn.Parameter(
+                torch.empty((num_experts, 0), dtype=torch.int32,
+                            device=device),
+                requires_grad=False,
+            )
+            layer.w13_g_idx_sort_indices = torch.nn.Parameter(
+                torch.empty((num_experts, 0), dtype=torch.int32,
+                            device=device),
+                requires_grad=False,
+            )
+            layer.w2_g_idx_sort_indices = torch.nn.Parameter(
+                torch.empty((num_experts, 0), dtype=torch.int32,
+                            device=device),
+                requires_grad=False,
+            )
+        # Repack weights
+        marlin_w13_qweight = ops.gptq_marlin_moe_repack(
+            layer.w13_qweight,
+            layer.w13_g_idx_sort_indices,
+            layer.w13_qweight.shape[1] * self.quant_config.pack_factor,
+            layer.w13_qweight.shape[2],
+            self.quant_config.quant_type.size_bits,
+        )
+        replace_parameter(layer, "w13_qweight", marlin_w13_qweight)
+        marlin_w2_qweight = ops.gptq_marlin_moe_repack(
+            layer.w2_qweight,
+            layer.w2_g_idx_sort_indices,
+            layer.w2_qweight.shape[1] * self.quant_config.pack_factor,
+            layer.w2_qweight.shape[2],
+            self.quant_config.quant_type.size_bits,
+        )
+        replace_parameter(layer, "w2_qweight", marlin_w2_qweight)
+        # Repack scales
+        marlin_w13_scales = marlin_moe_permute_scales(
+            s=layer.w13_scales,
+            size_k=layer.intermediate_size_per_partition,
+            size_n=layer.w13_scales.shape[2],
+            group_size=self.quant_config.group_size,
+        )
+        replace_parameter(layer, "w13_scales", marlin_w13_scales)
+        marlin_w2_scales = marlin_moe_permute_scales(
+            s=layer.w2_scales,
+            size_k=layer.w2_scales.shape[1] *
+            (self.quant_config.group_size if self.quant_config.group_size != -1
+             else self.quant_config.pack_factor),
+            size_n=layer.w2_scales.shape[2],
+            group_size=self.quant_config.group_size,
+        )
+        replace_parameter(layer, "w2_scales", marlin_w2_scales)
+
+    def apply(
+        self,
+        layer: torch.nn.Module,
+        x: torch.Tensor,
+        router_logits: torch.Tensor,
+        top_k: int,
+        renormalize: bool,
+        use_grouped_topk: bool = False,
+        topk_group: Optional[int] = None,
+        num_expert_group: Optional[int] = None,
+        global_num_experts: int = -1,
+        expert_map: Optional[torch.Tensor] = None,
+        custom_routing_function: Optional[Callable] = None,
+        scoring_func: str = "softmax",
+        e_score_correction_bias: Optional[torch.Tensor] = None,
+        apply_router_weight_on_input: bool = False,
+        activation: str = "silu",
+    ) -> torch.Tensor:
+        assert activation == "silu", "Only SiLU activation is supported."
+        if apply_router_weight_on_input:
+            raise NotImplementedError(
+                "Apply router weight on input is not supported for"
+                "fused Marlin MoE method.")
+
+        topk_weights, topk_ids = FusedMoE.select_experts(
+            hidden_states=x,
+            router_logits=router_logits,
+            use_grouped_topk=use_grouped_topk,
+            top_k=top_k,
+            renormalize=renormalize,
+            topk_group=topk_group,
+            num_expert_group=num_expert_group,
+            custom_routing_function=custom_routing_function,
+            scoring_func=scoring_func,
+            e_score_correction_bias=e_score_correction_bias)
+
+        return torch.ops.vllm.fused_marlin_moe(
+            x,
+            layer.w13_qweight,
+            layer.w2_qweight,
+            layer.w13_scales,
+            layer.w2_scales,
+            router_logits,
+            topk_weights,
+            topk_ids,
+            global_num_experts=global_num_experts,
+            expert_map=expert_map,
+            g_idx1=layer.w13_g_idx,
+            g_idx2=layer.w2_g_idx,
+            sort_indices1=layer.w13_g_idx_sort_indices,
+            sort_indices2=layer.w2_g_idx_sort_indices,
+            num_bits=self.quant_config.quant_type.size_bits,
+            workspace=layer.workspace,
+            is_k_full=self.is_k_full)

model.safetensors.index.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c5407ce97b8572b8d5eb6b450fdf13e8d2d96358e40f9eddeb984843930d3c88
+size 13211434

qwen3_nonthinking.jinja

@@ -0,0 +1,82 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {{- messages[0].content + '\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
+        {%- set ns.multi_step_tool = false %}
+        {%- set ns.last_query_index = index %}
+    {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+        {{- '<|im_start|>' + message.role + '\n' + message.content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set content = message.content %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is defined and message.reasoning_content is not none %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in message.content %}
+                {%- set content = message.content.split('</think>')[-1].lstrip('\n') %}
+                {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {%- if loop.last or (not loop.last and reasoning_content) %}
+                {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
+            {%- else %}
+                {{- '<|im_start|>' + message.role + '\n' + content }}
+            {%- endif %}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if (loop.first and content) or (not loop.first) %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>\n{"name": "' }}
+                {{- tool_call.name }}
+                {{- '", "arguments": ' }}
+                {%- if tool_call.arguments is string %}
+                    {{- tool_call.arguments }}
+                {%- else %}
+                    {{- tool_call.arguments | tojson }}
+                {%- endif %}
+                {{- '}\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- message.content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n<think>\n\n</think>\n\n' }}
+{%- endif %}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
+size 11422654

tokenizer_config.json

@@ -0,0 +1,239 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151665": {
+      "content": "<tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151666": {
+      "content": "</tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151667": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151668": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "chat_template": "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0].role == 'system' %}\n        {{- messages[0].content + '\\n\\n' }}\n    {%- endif %}\n    {{- \"# Tools\\n\\nYou may call one or more functions to assist with the user query.\\n\\nYou are provided with function signatures within <tools></tools> XML tags:\\n<tools>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\n</tools>\\n\\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\\n<tool_call>\\n{\\\"name\\\": <function-name>, \\\"arguments\\\": <args-json-object>}\\n</tool_call><|im_end|>\\n\" }}\n{%- else %}\n    {%- if messages[0].role == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0].content + '<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}\n{%- for message in messages[::-1] %}\n    {%- set index = (messages|length - 1) - loop.index0 %}\n    {%- if ns.multi_step_tool and message.role == \"user\" and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}\n        {%- set ns.multi_step_tool = false %}\n        {%- set ns.last_query_index = index %}\n    {%- endif %}\n{%- endfor %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {%- set content = message.content %}\n        {%- set reasoning_content = '' %}\n        {%- if message.reasoning_content is defined and message.reasoning_content is not none %}\n            {%- set reasoning_content = message.reasoning_content %}\n        {%- else %}\n            {%- if '</think>' in message.content %}\n                {%- set content = message.content.split('</think>')[-1].lstrip('\\n') %}\n                {%- set reasoning_content = message.content.split('</think>')[0].rstrip('\\n').split('<think>')[-1].lstrip('\\n') %}\n            {%- endif %}\n        {%- endif %}\n        {%- if loop.index0 > ns.last_query_index %}\n            {%- if loop.last or (not loop.last and reasoning_content) %}\n                {{- '<|im_start|>' + message.role + '\\n<think>\\n' + reasoning_content.strip('\\n') + '\\n</think>\\n\\n' + content.lstrip('\\n') }}\n            {%- else %}\n                {{- '<|im_start|>' + message.role + '\\n' + content }}\n            {%- endif %}\n        {%- else %}\n            {{- '<|im_start|>' + message.role + '\\n' + content }}\n        {%- endif %}\n        {%- if message.tool_calls %}\n            {%- for tool_call in message.tool_calls %}\n                {%- if (loop.first and content) or (not loop.first) %}\n                    {{- '\\n' }}\n                {%- endif %}\n                {%- if tool_call.function %}\n                    {%- set tool_call = tool_call.function %}\n                {%- endif %}\n                {{- '<tool_call>\\n{\"name\": \"' }}\n                {{- tool_call.name }}\n                {{- '\", \"arguments\": ' }}\n                {%- if tool_call.arguments is string %}\n                    {{- tool_call.arguments }}\n                {%- else %}\n                    {{- tool_call.arguments | tojson }}\n                {%- endif %}\n                {{- '}\\n</tool_call>' }}\n            {%- endfor %}\n        {%- endif %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if loop.first or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n    {%- if enable_thinking is defined and enable_thinking is false %}\n        {{- '<think>\\n\\n</think>\\n\\n' }}\n    {%- endif %}\n{%- endif %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}