Upload folder using huggingface_hub

.gitattributes

@@ -33,3 +33,4 @@ 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
+tokenizer.json filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,3 +1,226 @@
----
-license: apache-2.0
----
+# Klear
+
+<div align="center">
+  <img src="figures/klear-logo-02.png" width="500"/>
+  <p>
+    🤗 <a href="https://huggingface.co/Kwai-Klear">Hugging Face</a> |  📑 <a href="">Technique Report</a>  
+  <br>
+  🖥️ <a href="https://kml-dtmachine-15498-prod-1.kmlhb2az1l3-2.corp.kuaishou.com">Chat with Klear</a> | 💬 <a href="https://github.com/Kwai-Klear">Issues & Discussions</a>
+  </p>
+</div>
+
+
+## 🔥News
+
+- 2025.09.05: We released `Klear-46B-A2.5B` series. Currently, Klear-46B-A2.5B offers two versions: `a base model` and an advanced version that includes `instruction tuned` model. Additionally, `an reasoning version is currently in training`. Please stay tuned for more updates.
+
+
+## 1. Introduction
+
+
+`Klear-46B-A2.5B` is a sparse Mixture-of-Experts (MoE) large language model developed by **the Kwai-Klear Team at Kuaishou**, designed to deliver both **high performance** and **inference efficiency**. It features **256 experts**, with only **8 activated** per forward pass, resulting in **46 billion total parameters** but just **2.5 billion active** — achieving dense-level performance at a fraction of the computational cost.
+
+The model was trained on over **22 trillion tokens** using a **three-stage progressive curriculum**:
+
+**1. Foundational Knowledge Learning (12T tokens):**
+General-purpose datasets such as CommonCrawl were processed with stratified quality filters, following a curriculum learning strategy that progresses from lower to higher data quality.
+
+**2. Data Complexity Enhancement (8T tokens):**
+The proportion of mathematical, coding, and STEM-related data was gradually increased to strengthen the model's reasoning and problem-solving capabilities.
+
+**3. Reasoning Enhancement and Longcontext Stage (2T tokens):**
+Training focused on synthetic and reasoning-intensive data, combined with a fast learning rate annealing strategy to maximize data efficiency and optimize final performance.
+
+As a result, Klear-46B-A2.5B-Base matches or surpasses the performance of dense models with several times more active parameters, while offering significantly better efficiency and cost-effectiveness for real-world deployment.
+
+
+## Model Summary
+
+this repo contains the base and instruction-tuned model**. which has the following architecture:
+
+| **propoty**               | **value**                                                              |
+|---------------------------|------------------------------------------------------------------------|
+| hidden_size       | 2048                                                                     |
+| moe_intermediate_size                  | 896                                                                    |
+| n_shared_experts       | 1                                                                      |
+| num_attention_heads         | 32                                                                    |
+| num_experts     | 256                                                                    |
+| num_experts_per_tok               | 8                                                                    |
+| num_hidden_layers       | 32                                                                      |
+| num_key_value_heads               | 4                                                                   |
+| vocab_size                | 151936                                                                 |
+| tie_word_embeddings       | false                                                                  |
+| context length       | 65536                                                                  |
+
+
+### Model Downloads
+
+<div align="center">
+
+| **Model** | **#Total Params** | **#Activated Params** | **Context Length** | **Download Link** |
+| :------------: | :------------: | :------------: | :------------: | :------------: |
+| Klear-46B-A2.5B-Base | 46B | 2.5B | 64K   | [🤗 Hugging Face](https://huggingface.co/Kwai-Klear)   |
+| Klear-46B-A2.5B-Inst.  | 46B | 2.5B |  64K   | [🤗 Hugging Face](https://huggingface.co/Kwai-Klear)   |
+
+</div>
+
+
+## 2. Benchmark Evaluation
+### Klear-46B-A2.5B-Base Evaluation Results
+| Ability     | Benchmark              | Klear-46B-A2.5B-Base | MiMO-7B-Base | Qwen3-8B-BASE | Qwen3-14B-BASE | Ling-lite-1.5-Base | Qwen3-30B-A3B-BASE |
+| ----------- | ---------------------- | -------------------- | ------------ | ------------- | -------------- | ------------------ | ------------------ |
+|             | # Total Params         | 46B                  | 7B           | 8B            | 14B            | 16.8B              | 30B                |
+|             | # Activated Params     | 2.5B                 | 7B           | 8B            | 14B            | 2.75B              | 3B                 |
+| **Code**    | HumanEval (0-shot*)    | 89                   | -            | 84.1          | 87.8           | 83.5               | 90.9               |
+|             | MBPP (3-shot)          | 76                   | 55.2         | 69            | 74             | 66.6               | 75.6               |
+| **Math**    | MATH (4-shot, cot)     | 55.7                 | 36.78        | 58.4          | 57.1           | 56.98              | 57.6               |
+|             | CMATH (3-shot)         | 87.8                 | 78.5         | 88.3          | 90.7           | 85.7               | 89.7               |
+|             | GSM8K (4-shot, cot)    | 87.3                 | 78.47        | 89.4          | 90.3           | 87.6               | 91.1               |
+| **General** | MMLU-Pro (5-shot, cot) | 57.6                 | 43.1         | 55.2          | 58.1           | 49.9               | 58.8               |
+|             | MMLU (5-shot)          | 80.5                 | 69.24        | 77.1          | 80.6           | 73.7               | 80.4               |
+|             | CEval (5-shot)         | 89.8                 | 67.98        | 81.9          | 84.8           | 78.2               | 87.4               |
+|             | CMMLU (5-shot)         | 88                   | 70.79        | 82            | 85.6           | 81.2               | 87.1               |
+|             | GPQA (0-shot)          | 35.3                 | 31.03        | 33.9          | 35.7           | 30.1               | 35.5               |
+|             | AGIEval (0-shot)       | 52.3                 | 48.3*        | 51.7          | 55.7           | 54.3               | 56                 |
+|             | BBH (3-shot, cot)      | 77.9                 | 75.6         | 78.1          | 80.1           | 75.4               | 81.2               |
+| **Others**  | HellaSwag (0-shot)     | 80.5                 | 80*          | 78.7          | 81.5           | 80                 | 81.2               |
+|             | Winogrande (3-shot)    | 78.8                 | 78*          | 73.6          | 78.5           | 72.1               | 77.9               |
+|             | Triviaqa (5-shot)      | 69.6                 | 60.8*        | 56.3          | 62.1           | 60.9               | 65.6               |
+|             | Naturalqs (5-shot)     | 37.5                 | 23.46        | 25.7          | 29.1           | 28                 | 30.7               |
+|             | PIQA (0-shot)          | 81.6                 | 80.14        | 79.5          | 81.9           | 82                 | 80.7               |
+|             | SIQA (0-shot)          | 67.9                 | 51.74        | 56.2          | 58.4           | 56.3               | 56.3               |
+|             | OpenBookQA (0-shot)    | 37.8                 | 34.2         | 35            | 35.6           | 38.2               | 34.6               |
+
+Note:
+1. `*`During pretraining, we found that the HumanEval metric fluctuated significantly and was extremely sensitive to formatting. Therefore, we referred to the prompt from Ling-series paper to modify the original HumanEval. The results in the table are the evaluation metrics after this modification. 
+2. For Mimo-base-7B, the results marked with `*` are sourced from other public reports.
+
+### Klear-46B-A2.5B-Inst. Evaluation Results
+| Ability                   | Benchmark                   | Klear-46B-A2.5B-Inst. | MiniCPM4-8B | Qwen3-8B (NoThink) | gemma3-12b-it | Phi4-14B |
+| ------------------------- | --------------------------- | --------------- | ----------- | ------------------ | ------------- | -------- |
+|                           | # Total Params              | 46B             | 8B          | 8B                 | 12B           | 14B      |
+|                           | # Activated Params          | 2.5B            | 8B          | 8B                 | 12B           | 14B      |
+| **English Understanding** | MMLU-Redux                  | 82.23           | 77.63       | 79.32              | 78.39         | 83.09    |
+|                           | MMLU-Pro                    | 64.82           | 54.69       | 63.8               | 60.69         | 67.25    |
+|                           | GPQA-Diamoind               | 49.49           | 38.51       | 51.77              | 39.02         | 59.47    |
+|                           | SimpleQA                    | 5.94            | 3.51        | 5.5                | 6.22          | 3.28     |
+| **Chinese Understanding** | CLUEWSC                     | 88.82           | 81.91       | 82.89              | 91.12         | 88.16    |
+|                           | CEval                       | 84.29           | 81.78       | 81.66              | 60.81         | 64.79    |
+|                           | C-SimpleQA                  | 42.03           | 23.13       | 37.07              | 28.97         | 24.77    |
+| **Math & Reasoning**      | MATH500                     | 86.4            | 79.8        | 85                 | 86.8          | 80.6     |
+|                           | AIME24                      | 30.42           | 22.92       | 28.33              | 23.96         | 15.83    |
+|                           | AIME25                      | 21.04           | 15.21       | 20.62              | 18.33         | 18.75    |
+|                           | ZebraLogic                  | 46.4            | 8.5         | 25.7               | 18            | 30.3     |
+| **Code**                  | HumanEval                   | 89.63           | 74.39       | 83.54              | 82.32         | 85.37    |
+|                           | HumanEval+                  | 87.2            | 70.12       | 76.83              | 75.61         | 83.54    |
+|                           | MBPPEvalplus                | 79.6            | 82          | 76.2               | 85.7          | 77.5     |
+|                           | MBPPEvalplus++              | 68.5            | 69.3        | 66.1               | 74.1          | 66.7     |
+|                           | LiveCodeBench v5(2408-2501) | 29.75           | 12.19       | 27.24              | 24.73         | 23.66    |
+| **Instruction Following** | IF-Eval                     | 80.41           | 73.01       | 84.47              | 81.52         | 59.33    |
+|                           | Multi-IF(en+zh)             | 78.25           | 61.79       | 78.95              | 76.56         | 62.7     |
+| **Comprehensive Ability** | MTBench                     | 8.03            | 6.875       | 8.21               | 8.675         | 8.625    |
+|                           | MT-Eval                     | 8.1             | 6.7         | 8.18               | 8.45          | 8.12     |
+|                           | Arena-Hard v2               | 19.8            | 2.2         | 19.8               | 50            | 9.6      |
+|                           | AlignBench v1.1             | 6.8             | 5.99        | 6.95               | 6.3           | 6.33     |
+|                           | LiveBench 1125              | 48.7            | 25.5        | 52.1               | 43.1          | 40       |
+
+## 3. Quick start
+
+### Inference with huggingface
+
+You can now inference in Transformers starting from version `4.56.0`.
+
+#### Klear-46B-A2.5B-Base
+
+```python
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+model_name = "/path/to/Klear-Base"
+tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
+    
+model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", dtype=torch.bfloat16, trust_remote_code=True)
+
+text = "世界上最大的湖是"
+inputs = tokenizer(text, return_tensors="pt")
+outputs = model.generate(**inputs.to(model.device), max_new_tokens=256)
+result = tokenizer.decode(outputs[0], skip_special_tokens=True)
+print(result)
+```
+
+#### Klear-46B-A2.5B-Inst.
+
+```python
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM, GenerationConfig
+
+model_name = "/path/to/Klear-Inst."
+tokenizer = AutoTokenizer.from_pretrained(model_name)
+
+model = AutoModelForCausalLM.from_pretrained(model_name, device_map="auto", torch_dtype=torch.bfloat16)
+
+messages = [
+    {"role": "user", "content": "帮我用 python 写一个计算器的代码吧。"}
+]
+input_tensor = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")
+outputs = model.generate(input_tensor.to(model.device), max_new_tokens=1024)
+
+result = tokenizer.decode(outputs[0][input_tensor.shape[1]:], skip_special_tokens=True)
+print(result)
+```
+
+### Inference with vllm
+
+[vLLM](https://github.com/vllm-project/vllm) is a high-speed and memery-efficicent inference framework. We provide our own forked version of [vLLM](https://github.com/vllm-project/vllm) here.
+
+```shell
+git clone
+cd vllm
+pip install 
+vllm serve Klear-46B-A2.5B-inst --port 8000 --tensor-parallel-size 8 --trust-remote-code
+```
+
+An OpenAI-compatible API will be available at `http://localhost:8000/v1`.
+
+Or you can refer to the following Python script for offline inference
+```python
+from vllm import LLM, SamplingParams
+
+model_path = "/path/to/Klear"
+llm = LLM(
+    model=model_path,
+    trust_remote_code=True,
+    num_speculative_tokens=1,
+    disable_log_stats=False
+)
+sampling_params = SamplingParams(temperature=0.2)
+
+conversation = [
+    {
+        "role": "system",
+        "content": ""
+    },
+    {
+        "role": "user",
+        "content": "Please help me write a snake game code.",
+    },
+]
+
+outputs = llm.chat(conversation,
+                   sampling_params=sampling_params,
+                   use_tqdm=False)
+
+for idx, output in enumerate(outputs):
+    prompt = output.prompt
+    generated_text = output.outputs[0].text
+    print(f"==== Response #{idx} ====")
+    print(f"Prompt: {prompt}, Generated text: {generated_text}")
+
+```
+
+## Citation
+
+If you find `Klear-46B-A2.5B` is useful or want to use in your projects, please kindly cite our paper:
+
+```
+```

config.json

@@ -0,0 +1,42 @@
+{
+  "architectures": [
+    "KlearMoeForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_klear.KlearConfig",
+    "AutoModel": "modeling_klear.KlearModel",
+    "AutoModelForCausalLM": "modeling_klear.KlearMoeForCausalLM"
+  },
+  "decoder_sparse_step": 1,
+  "dtype": "bfloat16",
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 8064,
+  "max_position_embeddings": 65536,
+  "mlp_only_layers": [],
+  "model_type": "Klear",
+  "moe_aux_loss_coeff": 0.0001,
+  "moe_intermediate_size": 896,
+  "n_shared_experts": 1,
+  "norm_topk_prob": true,
+  "num_attention_heads": 32,
+  "num_experts": 256,
+  "num_experts_per_tok": 8,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 4,
+  "output_router_logits": false,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 500000.0,
+  "routed_scaling_factor": 2.5,
+  "router_aux_loss_coef": 0.001,
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.56.0",
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

configuration_klear.py

@@ -0,0 +1,232 @@
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           This file was automatically generated from src/transformers/models/Klear/modular_Klear.py.
+#               Do NOT edit this file manually as any edits will be overwritten by the generation of
+#             the file from the modular. If any change should be done, please apply the change to the
+#                          modular_Klear.py file directly. One of our CI enforces this.
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.modeling_rope_utils import rope_config_validation
+
+
+class KlearConfig(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`KlearModel`]. It is used to instantiate a
+    Klear model according to the specified arguments, defining the model architecture. Instantiating a configuration
+    with the defaults will yield a similar configuration to that of [Klear-kwaii/Klear-MoE](https://huggingface.co/Klear/Klear-MoE).
+
+    Configuration objects inherit from [`PretrainedConfig`] and can be used to control the model outputs. Read the
+    documentation from [`PretrainedConfig`] for more information.
+
+
+    Args:
+        vocab_size (`int`, *optional*, defaults to 151936):
+            Vocabulary size of the Klear model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`KlearModel`]
+        hidden_size (`int`, *optional*, defaults to 2048):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 6144):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 24):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        num_key_value_heads (`int`, *optional*, defaults to 4):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details, check out [this
+            paper](https://huggingface.co/papers/2305.13245). If it is not specified, will default to `32`.
+
+        hidden_act (`str` or `function`, *optional*, defaults to `"silu"`):
+            The non-linear activation function (function or string) in the decoder.
+        max_position_embeddings (`int`, *optional*, defaults to 32768):
+            The maximum sequence length that this model might ever be used with.
+        initializer_range (`float`, *optional*, defaults to 0.02):
+            The standard deviation of the truncated_normal_initializer for initializing all weight matrices.
+        rms_norm_eps (`float`, *optional*, defaults to 1e-06):
+            The epsilon used by the rms normalization layers.
+        use_cache (`bool`, *optional*, defaults to `True`):
+            Whether or not the model should return the last key/values attentions (not used by all models). Only
+            relevant if `config.is_decoder=True`.
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether the model's input and output word embeddings should be tied.
+        rope_theta (`float`, *optional*, defaults to 10000.0):
+            The base period of the RoPE embeddings.
+        rope_scaling (`Dict`, *optional*):
+            Dictionary containing the scaling configuration for the RoPE embeddings. NOTE: if you apply new rope type
+            and you expect the model to work on longer `max_position_embeddings`, we recommend you to update this value
+            accordingly.
+            Expected contents:
+                `rope_type` (`str`):
+                    The sub-variant of RoPE to use. Can be one of ['default', 'linear', 'dynamic', 'yarn', 'longrope',
+                    'llama3'], with 'default' being the original RoPE implementation.
+                `factor` (`float`, *optional*):
+                    Used with all rope types except 'default'. The scaling factor to apply to the RoPE embeddings. In
+                    most scaling types, a `factor` of x will enable the model to handle sequences of length x *
+                    original maximum pre-trained length.
+                `original_max_position_embeddings` (`int`, *optional*):
+                    Used with 'dynamic', 'longrope' and 'llama3'. The original max position embeddings used during
+                    pretraining.
+                `attention_factor` (`float`, *optional*):
+                    Used with 'yarn' and 'longrope'. The scaling factor to be applied on the attention
+                    computation. If unspecified, it defaults to value recommended by the implementation, using the
+                    `factor` field to infer the suggested value.
+                `beta_fast` (`float`, *optional*):
+                    Only used with 'yarn'. Parameter to set the boundary for extrapolation (only) in the linear
+                    ramp function. If unspecified, it defaults to 32.
+                `beta_slow` (`float`, *optional*):
+                    Only used with 'yarn'. Parameter to set the boundary for interpolation (only) in the linear
+                    ramp function. If unspecified, it defaults to 1.
+                `short_factor` (`list[float]`, *optional*):
+                    Only used with 'longrope'. The scaling factor to be applied to short contexts (<
+                    `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
+                    size divided by the number of attention heads divided by 2
+                `long_factor` (`list[float]`, *optional*):
+                    Only used with 'longrope'. The scaling factor to be applied to long contexts (<
+                    `original_max_position_embeddings`). Must be a list of numbers with the same length as the hidden
+                    size divided by the number of attention heads divided by 2
+                `low_freq_factor` (`float`, *optional*):
+                    Only used with 'llama3'. Scaling factor applied to low frequency components of the RoPE
+                `high_freq_factor` (`float`, *optional*):
+                    Only used with 'llama3'. Scaling factor applied to high frequency components of the RoPE
+        attention_bias (`bool`, defaults to `False`, *optional*, defaults to `False`):
+            Whether to use a bias in the query, key, value and output projection layers during self-attention.
+        use_sliding_window (`bool`, *optional*, defaults to `False`):
+            Whether to use sliding window attention.
+        sliding_window (`int`, *optional*, defaults to 4096):
+            Sliding window attention (SWA) window size. If not specified, will default to `4096`.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        decoder_sparse_step (`int`, *optional*, defaults to 1):
+            The frequency of the MoE layer.
+        moe_intermediate_size (`int`, *optional*, defaults to 768):
+            Intermediate size of the routed expert.
+        num_experts_per_tok (`int`, *optional*, defaults to 8):
+            Number of selected experts.
+        num_experts (`int`, *optional*, defaults to 128):
+            Number of routed experts.
+        norm_topk_prob (`bool`, *optional*, defaults to `False`):
+            Whether to normalize the topk probabilities.
+        output_router_logits (`bool`, *optional*, defaults to `False`):
+            Whether or not the router logits should be returned by the model. Enabling this will also
+            allow the model to output the auxiliary loss, including load balancing loss and router z-loss.
+        router_aux_loss_coef (`float`, *optional*, defaults to 0.001):
+            The aux loss factor for the total loss.
+        mlp_only_layers (`list[int]`, *optional*, defaults to `[]`):
+            Indicate which layers use KlearMLP rather than KlearSparseMoeBlock
+            The list contains layer index, from 0 to num_layers-1 if we have num_layers layers
+            If `mlp_only_layers` is empty, `decoder_sparse_step` is used to determine the sparsity.
+
+    ```python
+    >>> from transformers import KlearModel, KlearConfig
+
+    >>> # Initializing a Klear style configuration
+    >>> configuration = KlearConfig()
+
+    >>> # Initializing a model from the Klear-MoE" style configuration
+    >>> model = KlearModel(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "Klear"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    # Default tensor parallel plan for base model `Klear`
+    base_model_tp_plan = {
+        "layers.*.self_attn.q_proj": "colwise",
+        "layers.*.self_attn.k_proj": "colwise",
+        "layers.*.self_attn.v_proj": "colwise",
+        "layers.*.self_attn.o_proj": "rowwise",
+        "layers.*.mlp.experts.*.gate_proj": "colwise",
+        "layers.*.mlp.experts.*.up_proj": "colwise",
+        "layers.*.mlp.experts.*.down_proj": "rowwise",
+        "layers.*.mlp.gate_proj": "colwise",
+        "layers.*.mlp.up_proj": "colwise",
+        "layers.*.mlp.down_proj": "rowwise",
+    }
+    base_model_pp_plan = {
+        "embed_tokens": (["input_ids"], ["inputs_embeds"]),
+        "layers": (["hidden_states", "attention_mask"], ["hidden_states"]),
+        "norm": (["hidden_states"], ["hidden_states"]),
+    }
+
+    def __init__(
+        self,
+        vocab_size=151936,
+        hidden_size=2048,
+        intermediate_size=6144,
+        num_hidden_layers=24,
+        num_attention_heads=32,
+        num_key_value_heads=4,
+        hidden_act="silu",
+        max_position_embeddings=32768,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        attention_bias=False,
+        use_sliding_window=False,
+        sliding_window=4096,
+        attention_dropout=0.0,
+        decoder_sparse_step=1,
+        moe_intermediate_size=768,
+        num_experts_per_tok=8,
+        num_experts=128,
+        norm_topk_prob=True,
+        output_router_logits=False,
+        router_aux_loss_coef=0.001,
+        mlp_only_layers=None,
+        routed_scaling_factor=2.5,
+        n_shared_experts=1,
+        **kwargs,
+    ):
+        super().__init__(
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.use_sliding_window = use_sliding_window
+        self.sliding_window = sliding_window if use_sliding_window else None
+
+        self.num_key_value_heads = num_key_value_heads
+        self.hidden_act = hidden_act
+        self.initializer_range = initializer_range
+        self.rms_norm_eps = rms_norm_eps
+        self.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.rope_scaling = rope_scaling
+        self.attention_bias = attention_bias
+        self.attention_dropout = attention_dropout
+        # Validate the correctness of rotary position embeddings parameters
+        # BC: if there is a 'type' field, move it to 'rope_type'.
+        if self.rope_scaling is not None and "type" in self.rope_scaling:
+            self.rope_scaling["rope_type"] = self.rope_scaling["type"]
+        rope_config_validation(self)
+
+        # MoE arguments
+        self.decoder_sparse_step = decoder_sparse_step
+        self.moe_intermediate_size = moe_intermediate_size
+        self.num_experts_per_tok = num_experts_per_tok
+        self.num_experts = num_experts
+        self.norm_topk_prob = norm_topk_prob
+        self.output_router_logits = output_router_logits
+        self.router_aux_loss_coef = router_aux_loss_coef
+        self.mlp_only_layers = [] if mlp_only_layers is None else mlp_only_layers
+
+        self.routed_scaling_factor = routed_scaling_factor
+        self.n_shared_experts = n_shared_experts
+
+
+__all__ = ["KlearConfig"]

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "bos_token_id": 151643,
+  "eos_token_id": 151643,
+  "max_new_tokens": 4096,
+  "transformers_version": "4.56.0"
+}

model-00001-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5abcb5985bf8c9458c0c7b09ec3ab1a2cf768ccdfe6f2ed2950464a788943077
+size 4996631924

model-00002-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:34dc79fdf0169eda05906aa526489901a0478fb8a3e02c88d55ac8ce8b5b70ee
+size 4998242320

model-00003-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43bb174eeaafd3b828d78e3cd613b34b30542b37a3f8b13c5d431c6b6f77b872
+size 4998242432

model-00004-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:616a619532c27e8bfa20ef98640c92dba8563818c03004410b8e416e9da9dd96
+size 4996651044

model-00005-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:730e1e114edfacd7dfa3084a45ef92129db865627f49234909a66416bb1e3fc1
+size 4998242232

model-00006-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8d90ac87c7b8aecf88c24833d7c958a1bebd5824350163eb402390a0dc92cc5
+size 4998242568

model-00007-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9008f92d3d7e6bec6b4384f0f032759ed278295f20bdf37d155241efa9bc1caa
+size 4998243808

model-00008-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e10f16e09709fac19bd6b029b04a1e6b54b16be721e27345a87a3a2a8ccc8f35
+size 4996652420

model-00009-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99c8ec31a8b49dde25eb9e612fa699147dcdc4786e36b707e0060a076e490a78
+size 4998243608

model-00010-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8d8e6544ffb56bcc6066f62f1a4249019b98db9558f7d57e2e8beee384b8f933
+size 4998243672

model-00011-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01973a3b1b920364392af52ee39659c8260d3117f295de225f8f876e800a59d9
+size 4998243808

model-00012-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e1401c2fdc3677c9c281313d307b9a09e7db1f179e80fbd39b7275fdd34be784
+size 4996652436

model-00013-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:48ae47d7f09144ba130c57facc60cf2c89ce55815defabb85871cf3e09aa479d
+size 4998243608

model-00014-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fd105f4cd7647d0617b736975f9131608091bcdfd8e04252f7cecac1f12c629
+size 4998243656

model-00015-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:acd4705ace9a526bfdc76fe94a3bcd16f70e0477f855318a83073ab61c180963
+size 4998243808

model-00016-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e362d8add20c3581a0b7a0b42314e2fdd5f51aa15316ade8f8cc1d49241ac0f
+size 4996652444

model-00017-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c1ed424862b26ac921af22f9e7aa48f7336153d4f0c503ccfa793f1fbae4703
+size 4998243608

model-00018-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7d3abdefe2a7e54d5e7f895ecb2577e8430c8b7b612dd7fb9873c7d0e83c47d0
+size 4998243648

model-00019-of-00019.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1022f8ad9ed178de6c29c23a9341572872dfae455203f48ddac848811de443b
+size 2472102012

modeling_klear.py

@@ -0,0 +1,763 @@
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           This file was automatically generated from src/transformers/models/klear/modular_klear.py.
+#               Do NOT edit this file manually as any edits will be overwritten by the generation of
+#             the file from the modular. If any change should be done, please apply the change to the
+#                          modular_klear.py file directly. One of our CI enforces this.
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+
+from typing import Callable, Optional, Union
+
+import torch
+import torch.nn.functional as F
+from torch import nn
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.generation import GenerationMixin
+from transformers.integrations import use_kernel_forward_from_hub
+from transformers.masking_utils import create_causal_mask, create_sliding_window_causal_mask
+from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
+from transformers.modeling_layers import (
+    GenericForQuestionAnswering,
+    GenericForSequenceClassification,
+    GenericForTokenClassification,
+    GradientCheckpointingLayer,
+)
+from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
+from transformers.modeling_outputs import MoeCausalLMOutputWithPast, MoeModelOutputWithPast
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
+from transformers.processing_utils import Unpack
+from transformers.utils import TransformersKwargs, auto_docstring, can_return_tuple
+from transformers.utils.generic import OutputRecorder, check_model_inputs
+from .configuration_klear import KlearConfig
+
+
+def rotate_half(x):
+    """Rotates half the hidden dims of the input."""
+    x1 = x[..., : x.shape[-1] // 2]
+    x2 = x[..., x.shape[-1] // 2 :]
+    return torch.cat((-x2, x1), dim=-1)
+
+
+def apply_rotary_pos_emb(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
+    """Applies Rotary Position Embedding to the query and key tensors.
+
+    Args:
+        q (`torch.Tensor`): The query tensor.
+        k (`torch.Tensor`): The key tensor.
+        cos (`torch.Tensor`): The cosine part of the rotary embedding.
+        sin (`torch.Tensor`): The sine part of the rotary embedding.
+        position_ids (`torch.Tensor`, *optional*):
+            Deprecated and unused.
+        unsqueeze_dim (`int`, *optional*, defaults to 1):
+            The 'unsqueeze_dim' argument specifies the dimension along which to unsqueeze cos[position_ids] and
+            sin[position_ids] so that they can be properly broadcasted to the dimensions of q and k. For example, note
+            that cos[position_ids] and sin[position_ids] have the shape [batch_size, seq_len, head_dim]. Then, if q and
+            k have the shape [batch_size, heads, seq_len, head_dim], then setting unsqueeze_dim=1 makes
+            cos[position_ids] and sin[position_ids] broadcastable to the shapes of q and k. Similarly, if q and k have
+            the shape [batch_size, seq_len, heads, head_dim], then set unsqueeze_dim=2.
+    Returns:
+        `tuple(torch.Tensor)` comprising of the query and key tensors rotated using the Rotary Position Embedding.
+    """
+    cos = cos.unsqueeze(unsqueeze_dim)
+    sin = sin.unsqueeze(unsqueeze_dim)
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: float,
+    dropout: float = 0.0,
+    **kwargs: Unpack[TransformersKwargs],
+):
+    key_states = repeat_kv(key, module.num_key_value_groups)
+    value_states = repeat_kv(value, module.num_key_value_groups)
+
+    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
+    if attention_mask is not None:
+        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+
+    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
+    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+    attn_output = torch.matmul(attn_weights, value_states)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+class KlearAttention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: KlearConfig, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.head_dim = getattr(config, "head_dim", config.hidden_size // config.num_attention_heads)
+        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
+        self.scaling = self.head_dim**-0.5
+        self.attention_dropout = config.attention_dropout
+        self.is_causal = True
+
+        self.q_proj = nn.Linear(
+            config.hidden_size, config.num_attention_heads * self.head_dim, bias=config.attention_bias
+        )
+        self.k_proj = nn.Linear(
+            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
+        )
+        self.v_proj = nn.Linear(
+            config.hidden_size, config.num_key_value_heads * self.head_dim, bias=config.attention_bias
+        )
+        self.o_proj = nn.Linear(
+            config.num_attention_heads * self.head_dim, config.hidden_size, bias=config.attention_bias
+        )
+        self.q_norm = KlearRMSNorm(self.head_dim, eps=config.rms_norm_eps)  # unlike olmo, only on the head dim!
+        self.k_norm = KlearRMSNorm(self.head_dim, eps=config.rms_norm_eps)  # thus post q_norm does not need reshape
+        self.sliding_window = getattr(config, "sliding_window", None)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor],
+        attention_mask: Optional[torch.Tensor],
+        past_key_value: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> tuple[torch.Tensor, Optional[torch.Tensor], Optional[tuple[torch.Tensor]]]:
+        input_shape = hidden_states.shape[:-1]
+        hidden_shape = (*input_shape, -1, self.head_dim)
+
+        query_states = self.q_norm(self.q_proj(hidden_states).view(hidden_shape)).transpose(1, 2)
+        key_states = self.k_norm(self.k_proj(hidden_states).view(hidden_shape)).transpose(1, 2)
+        value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+
+        cos, sin = position_embeddings
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+        if past_key_value is not None:
+            # sin and cos are specific to RoPE models; cache_position needed for the static cache
+            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
+            key_states, value_states = past_key_value.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        attention_interface: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+
+        attn_output, attn_weights = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            sliding_window=self.sliding_window,  # diff with Llama
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+        return attn_output, attn_weights
+
+
+class KlearMLP(nn.Module):
+    def __init__(self, config, intermediate_size=None):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = intermediate_size if intermediate_size is not None else config.intermediate_size
+        self.gate_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.up_proj = nn.Linear(self.hidden_size, self.intermediate_size, bias=False)
+        self.down_proj = nn.Linear(self.intermediate_size, self.hidden_size, bias=False)
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, x):
+        down_proj = self.down_proj(self.act_fn(self.gate_proj(x)) * self.up_proj(x))
+        return down_proj
+
+
+class KlearSparseMoeBlock(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.num_experts = config.num_experts
+        self.top_k = config.num_experts_per_tok
+        self.norm_topk_prob = config.norm_topk_prob
+
+        # router
+        self.gate = nn.Linear(config.hidden_size, config.num_experts, bias=False)
+        self.experts = nn.ModuleList(
+            [KlearMLP(config, intermediate_size=config.moe_intermediate_size) for _ in range(config.num_experts)]
+        )
+        self.shared_experts = KlearMLP(
+            config=config, intermediate_size=config.moe_intermediate_size * config.n_shared_experts
+        )
+
+        self.coefficient = nn.Linear(config.hidden_size, 2)
+        self.register_buffer("expert_bias", torch.zeros(self.num_experts, dtype=torch.float32))
+
+    def forward(self, hidden_states):
+        residuals = hidden_states
+        batch_size, sequence_length, hidden_dim = hidden_states.shape
+        hidden_states = hidden_states.view(-1, hidden_dim)
+        # router_logits: (batch * sequence_length, n_experts)
+        router_logits = nn.functional.linear(hidden_states.to(torch.float32), self.gate.weight.to(torch.float32))
+
+        routing_weights = F.sigmoid(router_logits)
+        ori_routing_weights = routing_weights
+
+        # using bias
+        biasd_routing_weights = routing_weights + self.expert_bias.unsqueeze(0)
+        _, selected_experts = torch.topk(biasd_routing_weights, self.top_k, dim=-1)
+
+        # Extract corresponding original probabilities
+        ori_routing_weights = torch.gather(ori_routing_weights, dim=-1, index=selected_experts)
+
+        if self.norm_topk_prob:  # only diff with mixtral sparse moe block!
+            ori_routing_weights /= ori_routing_weights.sum(dim=-1, keepdim=True)
+        # we cast back to the input dtype
+        ori_routing_weights = ori_routing_weights.to(hidden_states.dtype)
+
+        final_hidden_states = torch.zeros(
+            (batch_size * sequence_length, hidden_dim), dtype=hidden_states.dtype, device=hidden_states.device
+        )
+
+        # One hot encode the selected experts to create an expert mask
+        # this will be used to easily index which expert is going to be sollicitated
+        expert_mask = torch.nn.functional.one_hot(selected_experts, num_classes=self.num_experts).permute(2, 1, 0)
+
+        # Loop over all available experts in the model and perform the computation on each expert
+        expert_hitted = torch.greater(expert_mask.sum(dim=(-1, -2)), 0).nonzero()
+        for expert_idx in expert_hitted:
+            expert_layer = self.experts[expert_idx]
+            idx, top_x = torch.where(expert_mask[expert_idx].squeeze(0))
+
+            # Index the correct hidden states and compute the expert hidden state for
+            # the current expert. We need to make sure to multiply the output hidden
+            # states by `ori_routing_weights` on the corresponding tokens (top-1 and top-2)
+            current_state = hidden_states[None, top_x].reshape(-1, hidden_dim)
+            current_hidden_states = expert_layer(current_state) * ori_routing_weights[top_x, idx, None]
+
+            # However `index_add_` only support torch tensors for indexing so we'll use
+            # the `top_x` tensor here.
+            final_hidden_states.index_add_(0, top_x, current_hidden_states.to(hidden_states.dtype))
+        final_hidden_states = final_hidden_states.reshape(batch_size, sequence_length, hidden_dim)
+
+        coef = self.coefficient(residuals).softmax(dim=-1)
+        final_hidden_states = final_hidden_states * coef[..., :1] + self.shared_experts(residuals) * coef[..., 1:]
+
+        return final_hidden_states, router_logits
+
+
+@use_kernel_forward_from_hub("RMSNorm")
+class KlearRMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        KlearRMSNorm is equivalent to T5LayerNorm
+        """
+        super().__init__()
+        self.weight = nn.Parameter(torch.ones(hidden_size))
+        self.variance_epsilon = eps
+
+    def forward(self, hidden_states):
+        input_dtype = hidden_states.dtype
+        hidden_states = hidden_states.to(torch.float32)
+        variance = hidden_states.pow(2).mean(-1, keepdim=True)
+        hidden_states = hidden_states * torch.rsqrt(variance + self.variance_epsilon)
+        return self.weight * hidden_states.to(input_dtype)
+
+    def extra_repr(self):
+        return f"{tuple(self.weight.shape)}, eps={self.variance_epsilon}"
+
+
+class KlearDecoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config: KlearConfig, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+
+        self.self_attn = KlearAttention(config, layer_idx)
+
+        if (layer_idx not in config.mlp_only_layers) and (
+            config.num_experts > 0 and (layer_idx + 1) % config.decoder_sparse_step == 0
+        ):
+            self.mlp = KlearSparseMoeBlock(config)
+        else:
+            self.mlp = KlearMLP(config, intermediate_size=config.intermediate_size)
+
+        self.input_layernorm = KlearRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = KlearRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[tuple[torch.Tensor]] = None,
+        output_attentions: Optional[bool] = False,
+        output_router_logits: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        cache_position: Optional[torch.LongTensor] = None,
+        position_embeddings: Optional[tuple[torch.Tensor, torch.Tensor]] = None,  # necessary, but kept here for BC
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> tuple[torch.FloatTensor, Optional[tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        """
+        Args:
+            hidden_states (`torch.FloatTensor`): input to the layer of shape `(batch, seq_len, embed_dim)`
+            attention_mask (`torch.FloatTensor`, *optional*): attention mask of size
+                `(batch, sequence_length)` where padding elements are indicated by 0.
+            output_attentions (`bool`, *optional*):
+                Whether or not to return the attentions tensors of all attention layers. See `attentions` under
+                returned tensors for more detail.
+            output_router_logits (`bool`, *optional*):
+                Whether or not to return the logits of all the routers. They are useful for computing the router loss,
+                and should not be returned during inference.
+            use_cache (`bool`, *optional*):
+                If set to `True`, `past_key_values` key value states are returned and can be used to speed up decoding
+                (see `past_key_values`).
+            past_key_value (`Tuple(torch.FloatTensor)`, *optional*): cached past key and value projection states
+            cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*):
+                Indices depicting the position of the input sequence tokens in the sequence.
+            position_embeddings (`tuple[torch.FloatTensor, torch.FloatTensor]`, *optional*):
+                Tuple containing the cosine and sine positional embeddings of shape `(batch_size, seq_len, head_dim)`,
+                with `head_dim` being the embedding dimension of each attention head.
+            kwargs (`dict`, *optional*):
+                Arbitrary kwargs to be ignored, used for FSDP and other methods that injects code
+                into the model
+        """
+
+        residual = hidden_states
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, self_attn_weights = self.self_attn(
+            hidden_states=hidden_states,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_value=past_key_value,
+            output_attentions=output_attentions,
+            use_cache=use_cache,
+            cache_position=cache_position,
+            position_embeddings=position_embeddings,
+            **kwargs,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+
+        hidden_states = self.mlp(hidden_states)
+        if isinstance(hidden_states, tuple):
+            hidden_states, router_logits = hidden_states
+        else:
+            router_logits = None
+
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        if output_router_logits:
+            outputs += (router_logits,)
+
+        return outputs
+
+
+class KlearRotaryEmbedding(nn.Module):
+    def __init__(self, config: KlearConfig, device=None):
+        super().__init__()
+        # BC: "rope_type" was originally "type"
+        if hasattr(config, "rope_scaling") and isinstance(config.rope_scaling, dict):
+            self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
+        else:
+            self.rope_type = "default"
+        self.max_seq_len_cached = config.max_position_embeddings
+        self.original_max_seq_len = config.max_position_embeddings
+
+        self.config = config
+        self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
+
+        inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.original_inv_freq = self.inv_freq
+
+    @torch.no_grad()
+    @dynamic_rope_update  # power user: used with advanced RoPE types (e.g. dynamic rope)
+    def forward(self, x, position_ids):
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(x.device)
+        position_ids_expanded = position_ids[:, None, :].float()
+
+        device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
+        with torch.autocast(device_type=device_type, enabled=False):  # Force float32
+            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
+            emb = torch.cat((freqs, freqs), dim=-1)
+            cos = emb.cos() * self.attention_scaling
+            sin = emb.sin() * self.attention_scaling
+
+        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+
+
+@auto_docstring
+class KlearPreTrainedModel(PreTrainedModel):
+    config: KlearConfig
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["KlearDecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_flash_attn = True
+    _supports_sdpa = True
+    _supports_flex_attn = True
+    _can_compile_fullgraph = False  # MoE models don't work with torch.compile (`torch.where(condition)` not supported)
+    _supports_attention_backend = True
+    _can_record_outputs = {
+        "router_logits": OutputRecorder(KlearSparseMoeBlock, index=1),
+        "hidden_states": KlearDecoderLayer,
+        "attentions": KlearAttention,
+    }
+
+
+@auto_docstring
+class KlearModel(KlearPreTrainedModel):
+    def __init__(self, config: KlearConfig):
+        super().__init__(config)
+        self.padding_idx = config.pad_token_id
+        self.vocab_size = config.vocab_size
+
+        self.embed_tokens = nn.Embedding(config.vocab_size, config.hidden_size, self.padding_idx)
+        self.layers = nn.ModuleList(
+            [KlearDecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.norm = KlearRMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = KlearRotaryEmbedding(config=config)
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @check_model_inputs
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[list[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        output_router_logits: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> MoeModelOutputWithPast:
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_router_logits = (
+            output_router_logits if output_router_logits is not None else self.config.output_router_logits
+        )
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        use_cache = use_cache if use_cache is not None else self.config.use_cache
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache()
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if cache_position is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position = torch.arange(
+                past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
+            )
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        mask_function = create_causal_mask if self.config.sliding_window is None else create_sliding_window_causal_mask
+        causal_mask = mask_function(
+            config=self.config,
+            input_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+            cache_position=cache_position,
+            past_key_values=past_key_values,
+            position_ids=position_ids,
+        )
+
+        hidden_states = inputs_embeds
+
+        # create position embeddings to be shared across the decoder layers
+        position_embeddings = self.rotary_emb(hidden_states, position_ids)
+
+        # decoder layers
+        all_hidden_states = () if output_hidden_states else None
+        all_self_attns = () if output_attentions else None
+        all_router_logits = () if output_router_logits else None
+
+        for decoder_layer in self.layers:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            layer_outputs = decoder_layer(
+                hidden_states,
+                attention_mask=causal_mask,
+                position_ids=position_ids,
+                past_key_value=past_key_values,
+                output_attentions=output_attentions,
+                output_router_logits=output_router_logits,
+                use_cache=use_cache,
+                cache_position=cache_position,
+                position_embeddings=position_embeddings,
+                **kwargs,
+            )
+
+            hidden_states = layer_outputs[0]
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+            if output_router_logits:
+                all_router_logits += (layer_outputs[-1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        return MoeModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+            router_logits=all_router_logits,
+        )
+
+
+def load_balancing_loss_func(
+    gate_logits: Union[torch.Tensor, tuple[torch.Tensor], None],
+    num_experts: Optional[int] = None,
+    top_k: int = 2,
+    attention_mask: Optional[torch.Tensor] = None,
+    moe_aux_loss_coeff: float = 1,
+) -> torch.Tensor:
+    """
+    Computes sequence-level auxiliary load balancing loss for MoE gating.
+
+    Args:
+        gate_logits: Tensor of shape [batch_size, seq_len, num_experts]
+            or a tuple of such tensors (for multiple towers).
+        num_experts: Number of experts (inferred from gate_logits if None).
+        top_k: Number of top experts chosen per token.
+        attention_mask: Optional mask [batch_size, seq_len], 1 for valid tokens, 0 for padding.
+        moe_aux_loss_coeff: Scaling coefficient for the balancing loss.
+
+    Returns:
+        A scalar tensor representing the load balancing loss.
+    """
+    # Merge towers if provided
+    if isinstance(gate_logits, tuple):
+        gate_logits = torch.cat(gate_logits, dim=0)
+
+    assert gate_logits is not None, "gate_logits must be provided"
+    batch_size, seq_len, n_experts = gate_logits.shape
+    num_experts = n_experts if num_experts is None else num_experts
+    assert num_experts == n_experts, f"num_experts ({num_experts}) != gate dimension ({n_experts})"
+
+    # Compute gating probabilities
+    gate_probs = F.softmax(gate_logits, dim=-1)
+
+    # Optionally mask padding tokens
+    if attention_mask is not None:
+        mask = attention_mask.float().unsqueeze(-1)  # [batch, seq, 1]
+    else:
+        mask = torch.ones(batch_size, seq_len, 1, device=gate_logits.device)
+
+    # Select top_k experts per token
+    topk_vals, topk_idx = torch.topk(gate_probs, top_k, dim=-1)  # both [batch, seq, top_k]
+    # Build one-hot mask of assignments
+    one_hot = F.one_hot(topk_idx, num_experts).float()  # [batch, seq, top_k, num_experts]
+    # Sum along top_k to combine multiple choices
+    expert_mask = one_hot.sum(dim=2)  # [batch, seq, num_experts]
+
+    # Apply token mask
+    expert_mask = expert_mask * mask  # zeros out padding
+    gate_probs_masked = gate_probs * mask
+
+    # Normalizer: number of valid tokens per sample
+    tokens_per_sample = mask.sum(dim=1).clamp(min=1.0)  # [batch, 1]
+
+    # Sequence-level tokens per expert: fraction of tokens routed to each expert per sample
+    tokens_per_expert = expert_mask.sum(dim=1).div_(tokens_per_sample * top_k / num_experts)  # [batch, num_experts]
+
+    # Sequence-level average probability per expert per sample
+    router_prob_per_expert = gate_probs_masked.sum(dim=1).div(tokens_per_sample)  # [batch, num_experts]
+
+    # Compute loss per sample: encourage uniform load
+    # Loss = sum_e (tokens_e * probs_e)
+    loss_per_sample = (tokens_per_expert * router_prob_per_expert).sum(dim=1)  # [batch]
+    # Average across batch and scale
+    loss = moe_aux_loss_coeff * loss_per_sample.mean()
+    return loss
+
+
+@auto_docstring
+class KlearMoeForCausalLM(KlearPreTrainedModel, GenerationMixin):
+    _tied_weights_keys = ["lm_head.weight"]
+    _tp_plan = {"lm_head": "colwise_rep"}
+    _pp_plan = {"lm_head": (["hidden_states"], ["logits"])}
+
+    def __init__(self, config):
+        super().__init__(config)
+        self.model = KlearModel(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+        self.router_aux_loss_coef = config.router_aux_loss_coef
+        self.num_experts = config.num_experts
+        self.num_experts_per_tok = config.num_experts_per_tok
+        self.moe_aux_loss_coeff = getattr(config, "moe_aux_loss_coeff", 1.0)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    @can_return_tuple
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[list[torch.FloatTensor]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        output_router_logits: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> MoeCausalLMOutputWithPast:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, KlearMoeForCausalLM
+
+        >>> model = KlearMoeForCausalLM.from_pretrained("Klear-kwaii/Klear-MoE")
+        >>> tokenizer = AutoTokenizer.from_pretrained("Klear-kwaii/Klear-MoE")
+
+        >>> prompt = "Hey, are you conscious? Can you talk to me?"
+        >>> inputs = tokenizer(prompt, return_tensors="pt")
+
+        >>> # Generate
+        >>> generate_ids = model.generate(inputs.input_ids, max_length=30)
+        >>> tokenizer.batch_decode(generate_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
+        "Hey, are you conscious? Can you talk to me?\nI'm not conscious, but I can talk to you."
+        ```"""
+
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        output_router_logits = (
+            output_router_logits if output_router_logits is not None else self.config.output_router_logits
+        )
+
+        output_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs: MoeModelOutputWithPast = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=output_hidden_states,
+            output_router_logits=output_router_logits,
+            cache_position=cache_position,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits, labels, self.vocab_size, **kwargs)
+
+        aux_loss = None
+        if output_router_logits:
+            aux_loss = load_balancing_loss_func(
+                outputs.router_logits,
+                self.num_experts,
+                self.num_experts_per_tok,
+                attention_mask,
+                self.moe_aux_loss_coeff,
+            )
+            if labels is not None:
+                loss += self.router_aux_loss_coef * aux_loss.to(loss.device)  # make sure to reside in the same device
+
+        return MoeCausalLMOutputWithPast(
+            loss=loss,
+            aux_loss=aux_loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            router_logits=outputs.router_logits,
+        )
+
+
+class KlearForSequenceClassification(GenericForSequenceClassification, KlearPreTrainedModel):
+    pass
+
+
+class KlearForTokenClassification(GenericForTokenClassification, KlearPreTrainedModel):
+    pass
+
+
+class KlearForQuestionAnswering(GenericForQuestionAnswering, KlearPreTrainedModel):
+    base_model_prefix = "transformer"  # For BC, where `transformer` was used instead of `model`
+
+
+__all__ = [
+    "KlearMoeForCausalLM",
+    "KlearForQuestionAnswering",
+    "KlearModel",
+    "KlearPreTrainedModel",
+    "KlearForSequenceClassification",
+    "KlearForTokenClassification",
+]

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": "<|endoftext|>",
+  "errors": "replace",
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}