Initial model upload with custom code

.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

@@ -0,0 +1,58 @@
+---
+license: apache-2.0
+language:
+- code
+library_name: transformers
+tags:
+- masked-diffusion
+- code-generation
+- qwen2
+---
+
+# Model: fredzzp/open-dcoder-0.5B
+
+This repository contains the weights and custom code for the **fredzzp/open-dcoder-0.5B** model, a masked diffusion model for code generation based on the Qwen2 architecture.
+
+This model uses bidirectional attention and must be used with the custom `diffusion_generate` method.
+
+## How to Use
+
+First, make sure you have the latest `transformers` library installed.
+
+```bash
+pip install transformers torch huggingface_hub
+
+You can then use the model for generation. Note: You must pass trust_remote_code=True to load the custom model architecture.
+
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM
+
+model_id = "fredzzp/open-dcoder-0.5B"
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+# trust_remote_code=True is essential
+tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(
+    model_id,
+    torch_dtype=torch.bfloat16,
+    trust_remote_code=True
+).to(device)
+
+prompt = "def fibonacci(n):"
+input_ids = tokenizer(prompt, return_tensors="pt").input_ids.to(device)
+
+# The model will use the generation_config.json from the repo by default
+# You can also override parameters here
+outputs = model.diffusion_generate(
+    inputs=input_ids,
+    max_new_tokens=100,
+    steps=16,
+    temperature=0.8
+)
+
+# Decode the output
+prompt_len = input_ids.shape[1]
+generated_text = tokenizer.decode(outputs.sequences[0][prompt_len:], skip_special_tokens=True)
+
+print("--- Generated Code ---")
+print(generated_text)

__init__.py

@@ -0,0 +1,13 @@
+# Copyright 2025 Bytedance Ltd. and/or its affiliates
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.

added_tokens.json

@@ -0,0 +1,25 @@
+{
+  "</tool_call>": 151658,
+  "<M>": 151665,
+  "<tool_call>": 151657,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

config.json

@@ -0,0 +1,32 @@
+{
+  "_attn_implementation_autoset": true,
+  "architectures": [
+    "Qwen2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "eos_token_id": 151643,
+  "hidden_act": "silu",
+  "hidden_size": 896,
+  "initializer_range": 0.02,
+  "intermediate_size": 4864,
+  "max_position_embeddings": 32768,
+  "max_window_layers": 24,
+  "model_type": "qwen2",
+  "num_attention_heads": 14,
+  "num_hidden_layers": 24,
+  "num_key_value_heads": 2,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": 32768,
+  "tie_word_embeddings": true,
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.51.3",
+  "use_cache": true,
+  "use_sliding_window": false,
+  "vocab_size": 151936,
+  "auto_map": {
+    "AutoModelForCausalLM": "modeling_qwen2.Qwen2ForCausalLM"
+  }
+}

generation_config.json

@@ -0,0 +1,9 @@
+{
+  "mask_token_id": 151643,
+  "max_new_tokens": 256,
+  "steps": 24,
+  "temperature": 0.7,
+  "top_k": 500,
+  "alg": "entropy",
+  "alg_temp": 0.6
+}

generation_utils.py

@@ -0,0 +1,249 @@
+# veomni/models/transformers/qwen2/generation_utils.py
+
+import warnings
+import copy
+from dataclasses import dataclass
+from typing import Any, Dict, Optional, Tuple, Union
+
+import torch
+import torch.distributions as dists
+from torch.nn import functional as F
+from transformers import __version__
+from transformers.generation.configuration_utils import GenerationConfig
+from transformers.utils import ModelOutput, is_torchdynamo_compiling, logging
+
+logger = logging.get_logger(__name__)
+
+
+def top_p_logits(logits, top_p=None):
+    sorted_logits, sorted_indices = torch.sort(logits, descending=True)
+    cumulative_probs = torch.cumsum(F.softmax(sorted_logits, dim=-1), dim=-1)
+    sorted_indices_to_remove = cumulative_probs > top_p
+    sorted_indices_to_remove[..., 1:] = sorted_indices_to_remove[..., :-1].clone()
+    sorted_indices_to_remove[..., 0] = 0
+    mask = torch.zeros_like(logits, dtype=torch.bool, device=logits.device)
+    mask = mask.scatter_(-1, sorted_indices, sorted_indices_to_remove)
+    logits = logits.masked_fill(mask, torch.finfo(logits.dtype).min)
+    return logits
+
+def top_k_logits(logits, top_k=None):
+    if top_k is None or top_k == 0:
+        return logits
+    top_k = min(top_k, logits.size(-1))
+    indices_to_remove = logits < torch.topk(logits, top_k)[0][..., -1, None]
+    logits = logits.masked_fill(indices_to_remove, torch.finfo(logits.dtype).min)
+    return logits
+
+def sample_tokens(logits, temperature=0.0, top_p=None, top_k=None, margin_confidence=False, neg_entropy=False):
+    if temperature > 0:
+        logits = logits / temperature
+    if top_p is not None and top_p < 1:
+        logits = top_p_logits(logits, top_p)
+    if top_k is not None:
+        logits = top_k_logits(logits, top_k)
+    probs = torch.softmax(logits.float(), dim=-1)
+    if temperature > 0:
+        x0 = dists.Categorical(probs=probs).sample()
+    else:
+        _, x0 = probs.max(dim=-1)
+    
+    confidence = torch.gather(probs, -1, x0.unsqueeze(-1)).squeeze(-1)
+
+    if margin_confidence:
+        sorted_probs, _ = torch.sort(probs, dim=-1, descending=True)
+        top1_probs = sorted_probs[..., 0]
+        top2_probs = sorted_probs[..., 1]
+        confidence = top1_probs - top2_probs
+    elif neg_entropy:
+        log_probs = torch.log(probs.clamp(min=1e-10))
+        confidence = (probs * log_probs).sum(dim=-1)
+    
+    return confidence, x0
+
+
+@dataclass
+class MDMModelOutput(ModelOutput):
+    sequences: torch.LongTensor = None
+    history: Optional[Tuple[torch.FloatTensor]] = None
+
+class MDMGenerationConfig(GenerationConfig):
+    def __init__(self, **kwargs):
+        super().__init__(**kwargs)
+        self.temperature: float = kwargs.pop("temperature", 0.0)
+        self.top_p: Optional[float] = kwargs.pop("top_p", None)
+        self.top_k: Optional[int] = kwargs.pop("top_k", None)
+        self.eps: float = kwargs.pop("eps", 1e-3)
+        self.steps: int = kwargs.pop("steps", 512)
+        self.alg: str = kwargs.pop("alg", 'entropy')
+        self.alg_temp: Optional[float] = kwargs.pop("alg_temp", 0.0)
+        self.output_history: bool = kwargs.pop("output_history", False)
+        self.mask_token_id = kwargs.pop("mask_token_id", None)
+
+
+class MDMGenerationMixin:
+    """
+    Mixin class for Masked Diffusion Model generation, adapted from the Dream model's generation utils.
+    """
+    @staticmethod
+    def _expand_inputs_for_generation(
+        expand_size: int = 1,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.LongTensor] = None
+    ) -> Tuple[torch.LongTensor, Dict[str, Any]]:
+        if expand_size == 1:
+            return input_ids, attention_mask
+        
+        if input_ids is not None:
+            input_ids = input_ids.repeat_interleave(expand_size, dim=0)
+        if attention_mask is not None:
+            attention_mask = attention_mask.repeat_interleave(expand_size, dim=0)
+        return input_ids, attention_mask
+
+    def _prepare_generation_config(
+        self, generation_config: Optional[GenerationConfig], **kwargs
+    ) -> MDMGenerationConfig:
+        if generation_config is None:
+            generation_config = self.generation_config
+        
+        # Use MDMGenerationConfig as the target class
+        if not isinstance(generation_config, MDMGenerationConfig):
+            generation_config = MDMGenerationConfig.from_dict(generation_config.to_dict())
+
+        # Update with kwargs
+        generation_config.update(**kwargs)
+        return generation_config
+
+    @torch.no_grad()
+    def diffusion_generate(
+        self,
+        inputs: Optional[torch.Tensor] = None,
+        generation_config: Optional[MDMGenerationConfig] = None,
+        **kwargs,
+    ) -> Union[MDMModelOutput, torch.LongTensor]:
+        
+        # 1. Prepare generation config
+        generation_config = self._prepare_generation_config(generation_config, **kwargs)
+
+        # 2. Prepare inputs
+        input_ids = inputs
+        attention_mask = kwargs.get("attention_mask", None)
+
+        if input_ids is None:
+            raise ValueError("`inputs` must be provided for diffusion generation.")
+
+        if generation_config.max_new_tokens is not None:
+            generation_config.max_length = input_ids.shape[-1] + generation_config.max_new_tokens
+        
+        # 3. Expand inputs for multi-sequence generation
+        input_ids, attention_mask = self._expand_inputs_for_generation(
+            expand_size=generation_config.num_return_sequences,
+            input_ids=input_ids,
+            attention_mask=attention_mask
+        )
+        # 4. Run the sampling loop
+        return self._sample(
+            input_ids,
+            attention_mask=attention_mask,
+            generation_config=generation_config
+        )
+
+    def _sample(
+        self,
+        input_ids: torch.LongTensor,
+        attention_mask: Optional[torch.LongTensor],
+        generation_config: MDMGenerationConfig
+    ) -> Union[MDMModelOutput, torch.LongTensor]:
+        
+        # Extract params from config
+        max_length = generation_config.max_length
+        mask_token_id = generation_config.mask_token_id
+        if mask_token_id is None:
+            raise ValueError("`mask_token_id` must be set in the generation config.")
+
+        steps = generation_config.steps
+        eps = generation_config.eps
+        alg = generation_config.alg
+        alg_temp = generation_config.alg_temp
+        temperature = generation_config.temperature
+        top_p = generation_config.top_p
+        top_k = generation_config.top_k
+
+        histories = [] if generation_config.output_history else None
+
+        # Pad input_ids to max_length with mask tokens
+        x = F.pad(input_ids, (0, max_length - input_ids.shape[1]), value=mask_token_id)
+
+        # The model expects a bidirectional mask, so we just use the presence of pad_token_id
+        # for the attention mask during generation.
+        gen_attention_mask = (x != self.config.pad_token_id).long() if self.config.pad_token_id is not None else None
+
+        timesteps = torch.linspace(1, eps, steps + 1, device=x.device)
+
+        for i in range(steps):
+            mask_index = (x == mask_token_id)
+            if not mask_index.any(): # Stop if no tokens are masked
+                break
+            # is_causal=False is crucial for bidirectional attention
+            outputs = self(input_ids=x, attention_mask=gen_attention_mask, is_causal=False)
+            logits = outputs.logits
+            
+            # CRITICAL: Shift logits to predict the next token, aligning with training
+            logits = torch.cat([logits[:, :1], logits[:, :-1]], dim=1)
+
+            mask_logits = logits[mask_index]
+            t = timesteps[i]
+            s = timesteps[i + 1]
+
+            if alg == 'origin':
+                p_transfer = 1 - s / t if i < steps - 1 else 1
+                x0 = torch.full_like(x[mask_index], fill_value=mask_token_id, device=self.device, dtype=torch.long)
+                transfer_index_t_s = torch.rand(*x0.shape, device=self.device) < p_transfer
+                _, sampled_tokens = sample_tokens(mask_logits[transfer_index_t_s], temperature=temperature, top_p=top_p, top_k=top_k)
+                x0[transfer_index_t_s] = sampled_tokens
+                x[mask_index] = x0
+            else:
+                # Confidence-based sampling (maskgit, entropy, etc.)
+                confidence_alg_map = {'maskgit_plus': False, 'topk_margin': True, 'entropy': True}
+                is_margin_conf = confidence_alg_map.get(alg, False)
+                is_neg_entropy = alg == 'entropy'
+                
+                confidence, x0 = sample_tokens(mask_logits, temperature, top_p, top_k, margin_confidence=is_margin_conf, neg_entropy=is_neg_entropy)
+
+                num_masked = mask_index.sum(dim=-1, keepdim=True)
+                gamma = 1 - s / t
+                num_to_unmask = (num_masked * gamma).long()
+
+                # Place confidence scores back into a full tensor to find top-k across the sequence
+                full_confidence = torch.full_like(x, -torch.inf, device=self.device, dtype=confidence.dtype)
+                full_confidence[mask_index] = confidence
+
+                if (alg_temp is not None and alg_temp > 0):
+                    # Temperature-based sampling of which tokens to unmask
+                    unmask_probs = F.softmax(full_confidence / alg_temp, dim=-1)
+                    unmask_indices = torch.multinomial(unmask_probs, num_samples=num_to_unmask.max(), replacement=False)
+                else:
+                    # Top-k confidence sampling
+                    _, unmask_indices = torch.topk(full_confidence, k=num_to_unmask.max(), dim=-1)
+
+                # Create a mask for the tokens we are going to unmask
+                rows = torch.arange(x.size(0), device=x.device).unsqueeze(1)
+                unmask_selection_mask = torch.zeros_like(x, dtype=torch.bool)
+                unmask_selection_mask[rows, unmask_indices] = True
+                
+                # Filter indices based on per-row `num_to_unmask`
+                unmask_selection_mask = unmask_selection_mask & (torch.cumsum(unmask_selection_mask.long(), dim=-1) <= num_to_unmask)
+
+                # Place the newly generated tokens (x0) into a full tensor
+                x_unmasked_proposals = torch.full_like(x, fill_value=mask_token_id)
+                x_unmasked_proposals[mask_index] = x0
+
+                # Update the main tensor `x` with the unmasked tokens
+                x[unmask_selection_mask] = x_unmasked_proposals[unmask_selection_mask]
+
+            if histories is not None:
+                histories.append(x.clone())
+
+        if generation_config.return_dict_in_generate:
+            return MDMModelOutput(sequences=x, history=histories)
+        else:
+            return x

model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d28ef84b77c6d180347b2de1fe48c92c4e487d807ae85cf217f30f3574fc9df2
+size 1260367448

modeling_qwen2.py

@@ -0,0 +1,1084 @@
+# Copyright 2024 The Qwen team, Alibaba Group and The HuggingFace Inc. team. All rights reserved.
+# Copyright 2025 Bytedance Ltd. and/or its affiliates
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+# adapted from https://github.com/huggingface/transformers/blob/v4.49.0/src/transformers/models/qwen2/modeling_qwen2.py
+from typing import Callable, List, Optional, Tuple, Union
+
+import torch
+from torch import nn
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache, SlidingWindowCache, StaticCache
+from transformers.generation import GenerationMixin
+from transformers.modeling_attn_mask_utils import AttentionMaskConverter
+from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
+from transformers.modeling_outputs import (
+    BaseModelOutputWithPast,
+    CausalLMOutputWithPast,
+)
+from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
+from transformers.models.qwen2.configuration_qwen2 import Qwen2Config
+from transformers.processing_utils import Unpack
+from transformers.utils import (
+    # LossKwargs,
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    replace_return_docstrings,
+)
+
+from veomni.models.transformers.qwen2.generation_utils import MDMGenerationMixin
+
+from ....data.constants import IGNORE_INDEX
+from ....distributed.parallel_state import get_parallel_state
+from ....distributed.sequence_parallel import (
+    gather_heads_scatter_seq,
+    gather_seq_scatter_heads,
+    reduce_sequence_parallel_loss,
+)
+from ....utils import logging
+from ....utils.import_utils import is_liger_kernel_available
+
+
+if is_liger_kernel_available():
+    from liger_kernel.transformers import LigerFusedLinearCrossEntropyLoss  # type: ignore
+    from liger_kernel.transformers.rms_norm import LigerRMSNorm
+    from liger_kernel.transformers.rope import liger_rotary_pos_emb
+    from liger_kernel.transformers.swiglu import LigerSwiGLUMLP
+
+logger = logging.get_logger(__name__)
+
+_CHECKPOINT_FOR_DOC = "meta-qwen2/Qwen2-2-7b-hf"
+_CONFIG_FOR_DOC = "Qwen2Config"
+
+
+class Qwen2MLP(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size
+        self.intermediate_size = 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
+
+
+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,
+):
+    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 Qwen2Attention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: Qwen2Config, 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.q_proj = nn.Linear(config.hidden_size, config.num_attention_heads * self.head_dim, bias=True)
+        self.k_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=True)
+        self.v_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=True)
+        self.o_proj = nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=False)
+
+    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,
+        is_causal: bool = True,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> Tuple[torch.Tensor, Optional[torch.Tensor], Optional[Tuple[torch.Tensor]]]:
+        bsz, q_len, _ = hidden_states.size()  # q_len = seq_length / sp_size
+        hidden_shape = (bsz, q_len, -1, self.head_dim)
+
+        query_states = self.q_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+        key_states = self.k_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+        value_states = self.v_proj(hidden_states).view(hidden_shape).transpose(1, 2)
+        if get_parallel_state().ulysses_enabled:
+            query_states = gather_seq_scatter_heads(query_states, seq_dim=2, head_dim=1)
+            key_states = gather_seq_scatter_heads(key_states, seq_dim=2, head_dim=1)
+            value_states = gather_seq_scatter_heads(value_states, seq_dim=2, head_dim=1)
+            # (batch_size, num_head / sp_size, seq_length, head_size)
+
+        full_q_len = query_states.size(2)  # full_q_len = seq_length
+        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)
+
+        sliding_window = None
+        if (
+            self.config.use_sliding_window
+            and getattr(self.config, "sliding_window", None) is not None
+            and self.layer_idx >= self.config.max_window_layers
+        ):
+            sliding_window = self.config.sliding_window
+
+        attention_interface: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            if self.config._attn_implementation == "sdpa" and kwargs.get("output_attentions", False):
+                logger.warning_once(
+                    "`torch.nn.functional.scaled_dot_product_attention` does not support `output_attentions=True`. Falling back to "
+                    'eager attention. This warning can be removed using the argument `attn_implementation="eager"` when loading the model.'
+                )
+            else:
+                attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+        self.is_causal = is_causal
+        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=sliding_window,  # main diff with Llama
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(bsz, full_q_len, -1, self.head_dim).contiguous()
+        if get_parallel_state().ulysses_enabled:
+            attn_output = gather_heads_scatter_seq(attn_output, head_dim=2, seq_dim=1)
+
+        attn_output = attn_output.reshape(bsz, q_len, self.config.hidden_size).contiguous()
+        attn_output = self.o_proj(attn_output)
+        return attn_output, attn_weights
+
+
+class Qwen2RMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        Qwen2RMSNorm 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 Qwen2DecoderLayer(nn.Module):
+    def __init__(self, config: Qwen2Config, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+        self.self_attn = Qwen2Attention(config=config, layer_idx=layer_idx)
+        self.mlp = Qwen2MLP(config)
+        self.input_layernorm = Qwen2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = Qwen2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        if config.sliding_window and config._attn_implementation != "flash_attention_2":
+            logger.warning_once(
+                f"Sliding Window Attention is enabled but not implemented for `{config._attn_implementation}`; "
+                "unexpected results may be encountered."
+            )
+
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_value: Optional[Cache] = None,
+        output_attentions: Optional[bool] = False,
+        use_cache: Optional[bool] = False,
+        cache_position: Optional[torch.LongTensor] = None,
+        position_embeddings: Optional[Tuple[torch.Tensor, torch.Tensor]] = None,
+        is_causal: bool = True,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> Tuple[torch.FloatTensor, Optional[Tuple[torch.FloatTensor, torch.FloatTensor]]]:
+        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,
+            is_causal=is_causal,
+            **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)
+        hidden_states = residual + hidden_states
+
+        outputs = (hidden_states,)
+        if output_attentions:
+            outputs += (self_attn_weights,)
+
+        return outputs
+
+
+class Qwen2RotaryEmbedding(nn.Module):
+    def __init__(self, config: Qwen2Config, device=None):
+        super().__init__()
+        # BC: "rope_type" was originally "type"
+        if hasattr(config, "rope_scaling") and config.rope_scaling is not None:
+            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
+
+    def _dynamic_frequency_update(self, position_ids, device):
+        """
+        dynamic RoPE layers should recompute `inv_freq` in the following situations:
+        1 - growing beyond the cached sequence length (allow scaling)
+        2 - the current sequence length is in the original scale (avoid losing precision with small sequences)
+        """
+        seq_len = torch.max(position_ids) + 1
+        if seq_len > self.max_seq_len_cached:  # growth
+            inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device, seq_len=seq_len)
+            self.register_buffer("inv_freq", inv_freq, persistent=False)  # TODO joao: may break with compilation
+            self.max_seq_len_cached = seq_len
+
+        if seq_len < self.original_max_seq_len and self.max_seq_len_cached > self.original_max_seq_len:  # reset
+            # This .to() is needed if the model has been moved to a device after being initialized (because
+            # the buffer is automatically moved, but not the original copy)
+            self.original_inv_freq = self.original_inv_freq.to(device)
+            self.register_buffer("inv_freq", self.original_inv_freq, persistent=False)
+            self.max_seq_len_cached = self.original_max_seq_len
+
+    @torch.no_grad()
+    def forward(self, x, position_ids):
+        if "dynamic" in self.rope_type:
+            self._dynamic_frequency_update(position_ids, device=x.device)
+
+        # Core RoPE block
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1)
+        position_ids_expanded = position_ids[:, None, :].float()
+        # Force float32 (see https://github.com/huggingface/transformers/pull/29285)
+        device_type = x.device.type
+        device_type = device_type if isinstance(device_type, str) and device_type != "mps" else "cpu"
+        with torch.autocast(device_type=device_type, enabled=False):
+            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
+            emb = torch.cat((freqs, freqs), dim=-1)
+            cos = emb.cos()
+            sin = emb.sin()
+
+        # Advanced RoPE types (e.g. yarn) apply a post-processing scaling factor, equivalent to scaling attention
+        cos = cos * self.attention_scaling
+        sin = sin * self.attention_scaling
+
+        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+
+
+QWEN2_START_DOCSTRING = r"""
+    This model inherits from [`PreTrainedModel`]. Check the superclass documentation for the generic methods the
+    library implements for all its model (such as downloading or saving, resizing the input embeddings, pruning heads
+    etc.)
+
+    This model is also a PyTorch [torch.nn.Module](https://pytorch.org/docs/stable/nn.html#torch.nn.Module) subclass.
+    Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage
+    and behavior.
+
+    Parameters:
+        config ([`Qwen2Config`]):
+            Model configuration class with all the parameters of the model. Initializing with a config file does not
+            load the weights associated with the model, only the configuration. Check out the
+            [`~PreTrainedModel.from_pretrained`] method to load the model weights.
+"""
+
+
+@add_start_docstrings(
+    "The bare Qwen2 Model outputting raw hidden-states without any specific head on top.",
+    QWEN2_START_DOCSTRING,
+)
+class Qwen2PreTrainedModel(PreTrainedModel):
+    config_class = Qwen2Config
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["Qwen2DecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_flash_attn_2 = True
+    _supports_sdpa = True
+    _supports_flex_attn = True
+    _supports_cache_class = True
+    _supports_quantized_cache = True
+    _supports_static_cache = True
+    _supports_attention_backend = True
+
+    def _init_weights(self, module):
+        std = self.config.initializer_range
+        if isinstance(module, nn.Linear):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.bias is not None:
+                module.bias.data.zero_()
+        elif isinstance(module, nn.Embedding):
+            module.weight.data.normal_(mean=0.0, std=std)
+            if module.padding_idx is not None:
+                module.weight.data[module.padding_idx].zero_()
+
+
+QWEN2_INPUTS_DOCSTRING = r"""
+    Args:
+        input_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`):
+            Indices of input sequence tokens in the vocabulary. Padding will be ignored by default should you provide
+            it.
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            [What are input IDs?](../glossary#input-ids)
+        attention_mask (`torch.Tensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Mask to avoid performing attention on padding token indices. Mask values selected in `[0, 1]`:
+
+            - 1 for tokens that are **not masked**,
+            - 0 for tokens that are **masked**.
+
+            [What are attention masks?](../glossary#attention-mask)
+
+            Indices can be obtained using [`AutoTokenizer`]. See [`PreTrainedTokenizer.encode`] and
+            [`PreTrainedTokenizer.__call__`] for details.
+
+            If `past_key_values` is used, optionally only the last `input_ids` have to be input (see
+            `past_key_values`).
+
+            If you want to change padding behavior, you should read [`modeling_opt._prepare_decoder_attention_mask`]
+            and modify to your needs. See diagram 1 in [the paper](https://arxiv.org/abs/1910.13461) for more
+            information on the default strategy.
+
+            - 1 indicates the head is **not masked**,
+            - 0 indicates the head is **masked**.
+        position_ids (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Indices of positions of each input sequence tokens in the position embeddings. Selected in the range `[0,
+            config.n_positions - 1]`.
+
+            [What are position IDs?](../glossary#position-ids)
+        past_key_values (`Cache` or `tuple(tuple(torch.FloatTensor))`, *optional*):
+            Pre-computed hidden-states (key and values in the self-attention blocks and in the cross-attention
+            blocks) that can be used to speed up sequential decoding. This typically consists in the `past_key_values`
+            returned by the model at a previous stage of decoding, when `use_cache=True` or `config.use_cache=True`.
+
+            Two formats are allowed:
+            - a [`~cache_utils.Cache`] instance, see our
+            [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache);
+            - Tuple of `tuple(torch.FloatTensor)` of length `config.n_layers`, with each tuple having 2 tensors of
+            shape `(batch_size, num_heads, sequence_length, embed_size_per_head)`). This is also known as the legacy
+            cache format.
+
+            The model will output the same cache format that is fed as input. If no `past_key_values` are passed, the
+            legacy cache format will be returned.
+
+            If `past_key_values` are used, the user can optionally input only the last `input_ids` (those that don't
+            have their past key value states given to this model) of shape `(batch_size, 1)` instead of all `input_ids`
+            of shape `(batch_size, sequence_length)`.
+        inputs_embeds (`torch.FloatTensor` of shape `(batch_size, sequence_length, hidden_size)`, *optional*):
+            Optionally, instead of passing `input_ids` you can choose to directly pass an embedded representation. This
+            is useful if you want more control over how to convert `input_ids` indices into associated vectors than the
+            model's internal embedding lookup matrix.
+        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`).
+        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_hidden_states (`bool`, *optional*):
+            Whether or not to return the hidden states of all layers. See `hidden_states` under returned tensors for
+            more detail.
+        return_dict (`bool`, *optional*):
+            Whether or not to return a [`~utils.ModelOutput`] instead of a plain tuple.
+        cache_position (`torch.LongTensor` of shape `(sequence_length)`, *optional*):
+            Indices depicting the position of the input sequence tokens in the sequence. Contrarily to `position_ids`,
+            this tensor is not affected by padding. It is used to update the cache in the correct position and to infer
+            the complete sequence length.
+"""
+
+
+@add_start_docstrings(
+    "The bare Qwen2 Model outputting raw hidden-states without any specific head on top.",
+    QWEN2_START_DOCSTRING,
+)
+class Qwen2Model(Qwen2PreTrainedModel):
+    """
+    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`Qwen2DecoderLayer`]
+
+    Args:
+        config: Qwen2Config
+    """
+
+    def __init__(self, config: Qwen2Config):
+        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(
+            [Qwen2DecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.norm = Qwen2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = Qwen2RotaryEmbedding(config=config)
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.embed_tokens = value
+
+    @add_start_docstrings_to_model_forward(QWEN2_INPUTS_DOCSTRING)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        is_causal: bool = True,
+        **flash_attn_kwargs: Unpack[FlashAttentionKwargs],
+    ) -> Union[Tuple, BaseModelOutputWithPast]:
+        output_attentions = output_attentions if output_attentions is not None else self.config.output_attentions
+        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
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        if self.gradient_checkpointing and self.training and use_cache:
+            logger.warning_once(
+                "`use_cache=True` is incompatible with gradient checkpointing. Setting `use_cache=False`."
+            )
+            use_cache = False
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache()
+
+        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)
+
+        causal_mask = self._update_causal_mask(
+            attention_mask, inputs_embeds, cache_position, past_key_values, output_attentions
+        )
+
+        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
+
+        for decoder_layer in self.layers[: self.config.num_hidden_layers]:
+            if output_hidden_states:
+                all_hidden_states += (hidden_states,)
+
+            if self.gradient_checkpointing and self.training:
+                layer_outputs = self._gradient_checkpointing_func(
+                    decoder_layer.__call__,
+                    hidden_states,
+                    causal_mask,
+                    position_ids,
+                    past_key_values,
+                    output_attentions,
+                    use_cache,
+                    cache_position,
+                    position_embeddings,
+                    is_causal,
+                )
+            else:
+                layer_outputs = decoder_layer(
+                    hidden_states,
+                    attention_mask=causal_mask,
+                    position_ids=position_ids,
+                    past_key_value=past_key_values,
+                    output_attentions=output_attentions,
+                    use_cache=use_cache,
+                    cache_position=cache_position,
+                    position_embeddings=position_embeddings,
+                    is_causal=is_causal,
+                    **flash_attn_kwargs,
+                )
+
+            hidden_states = layer_outputs[0]
+
+            if output_attentions:
+                all_self_attns += (layer_outputs[1],)
+
+        hidden_states = self.norm(hidden_states)
+
+        # add hidden states from the last decoder layer
+        if output_hidden_states:
+            all_hidden_states += (hidden_states,)
+
+        output = BaseModelOutputWithPast(
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values if use_cache else None,
+            hidden_states=all_hidden_states,
+            attentions=all_self_attns,
+        )
+        return output if return_dict else output.to_tuple()
+
+    def _update_causal_mask(
+        self,
+        attention_mask: torch.Tensor,
+        input_tensor: torch.Tensor,
+        cache_position: torch.Tensor,
+        past_key_values: Cache,
+        output_attentions: bool,
+    ):
+        if self.config._attn_implementation == "flash_attention_2":
+            if attention_mask is not None and past_key_values is not None:
+                is_padding_right = attention_mask[:, -1].sum().item() != input_tensor.size()[0]
+                if is_padding_right:
+                    raise ValueError(
+                        "You are attempting to perform batched generation with padding_side='right'"
+                        " this may lead to unexpected behaviour for Flash Attention version of Qwen2. Make sure to "
+                        " call `tokenizer.padding_side  = 'left'` before tokenizing the input. "
+                    )
+            if attention_mask is not None and 0.0 in attention_mask:
+                return attention_mask
+            return None
+
+        # For SDPA, when possible, we will rely on its `is_causal` argument instead of its `attn_mask` argument, in
+        # order to dispatch on Flash Attention 2. This feature is not compatible with static cache, as SDPA will fail
+        # to infer the attention mask.
+        past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+        using_static_cache = isinstance(past_key_values, StaticCache)
+        using_sliding_window_cache = isinstance(past_key_values, SlidingWindowCache)
+
+        # When output attentions is True, sdpa implementation's forward method calls the eager implementation's forward
+        if (
+            self.config._attn_implementation == "sdpa"
+            and not (using_static_cache or using_sliding_window_cache)
+            and not output_attentions
+        ):
+            if AttentionMaskConverter._ignore_causal_mask_sdpa(
+                attention_mask,
+                inputs_embeds=input_tensor,
+                past_key_values_length=past_seen_tokens,
+                sliding_window=self.config.sliding_window,
+                is_training=self.training,
+            ):
+                return None
+
+        dtype, device = input_tensor.dtype, input_tensor.device
+        min_dtype = torch.finfo(dtype).min
+        sequence_length = input_tensor.shape[1]
+        # SlidingWindowCache or StaticCache
+        if using_sliding_window_cache or using_static_cache:
+            target_length = past_key_values.get_max_cache_shape()
+        # DynamicCache or no cache
+        else:
+            target_length = (
+                attention_mask.shape[-1]
+                if isinstance(attention_mask, torch.Tensor)
+                else past_seen_tokens + sequence_length + 1
+            )
+
+        # In case the provided `attention` mask is 2D, we generate a causal mask here (4D).
+        causal_mask = self._prepare_4d_causal_attention_mask_with_cache_position(
+            attention_mask,
+            sequence_length=sequence_length,
+            target_length=target_length,
+            dtype=dtype,
+            device=device,
+            cache_position=cache_position,
+            batch_size=input_tensor.shape[0],
+            config=self.config,
+            past_key_values=past_key_values,
+        )
+
+        if (
+            self.config._attn_implementation == "sdpa"
+            and attention_mask is not None
+            and attention_mask.device.type in ["cuda", "xpu"]
+            and not output_attentions
+        ):
+            # Attend to all tokens in fully masked rows in the causal_mask, for example the relevant first rows when
+            # using left padding. This is required by F.scaled_dot_product_attention memory-efficient attention path.
+            # Details: https://github.com/pytorch/pytorch/issues/110213
+            causal_mask = AttentionMaskConverter._unmask_unattended(causal_mask, min_dtype)
+
+        return causal_mask
+
+    @staticmethod
+    def _prepare_4d_causal_attention_mask_with_cache_position(
+        attention_mask: torch.Tensor,
+        sequence_length: int,
+        target_length: int,
+        dtype: torch.dtype,
+        device: torch.device,
+        cache_position: torch.Tensor,
+        batch_size: int,
+        config: Qwen2Config,
+        past_key_values: Cache,
+    ):
+        """
+        Creates a causal 4D mask of shape `(batch_size, 1, query_length, key_value_length)` from a 2D mask of shape
+        `(batch_size, key_value_length)`, or if the input `attention_mask` is already 4D, do nothing.
+
+        Args:
+            attention_mask (`torch.Tensor`):
+                A 2D attention mask of shape `(batch_size, key_value_length)` or a 4D attention mask of shape `(batch_size, 1, query_length, key_value_length)`.
+            sequence_length (`int`):
+                The sequence length being processed.
+            target_length (`int`):
+                The target length: when generating with static cache, the mask should be as long as the static cache, to account for the 0 padding, the part of the cache that is not filled yet.
+            dtype (`torch.dtype`):
+                The dtype to use for the 4D attention mask.
+            device (`torch.device`):
+                The device to plcae the 4D attention mask on.
+            cache_position (`torch.Tensor`):
+                Indices depicting the position of the input sequence tokens in the sequence.
+            batch_size (`torch.Tensor`):
+                Batch size.
+            config (`Qwen2Config`):
+                The model's configuration class
+            past_key_values (`Cache`):
+                The cache class that is being used currently to generate
+        """
+        if attention_mask is not None and attention_mask.dim() == 4:
+            # In this case we assume that the mask comes already in inverted form and requires no inversion or slicing.
+            causal_mask = attention_mask
+        else:
+            min_dtype = torch.finfo(dtype).min
+            causal_mask = torch.full(
+                (sequence_length, target_length), fill_value=min_dtype, dtype=dtype, device=device
+            )
+            diagonal_attend_mask = torch.arange(target_length, device=device) > cache_position.reshape(-1, 1)
+            if config.sliding_window is not None:
+                # if we have sliding window, we should not attend to tokens beyond sliding window length, so we mask them out also
+                # the check is needed to verify is current checkpoint was trained with sliding window or not
+                if not isinstance(past_key_values, SlidingWindowCache) or sequence_length > target_length:
+                    sliding_attend_mask = torch.arange(target_length, device=device) <= (
+                        cache_position.reshape(-1, 1) - config.sliding_window
+                    )
+                    diagonal_attend_mask.bitwise_or_(sliding_attend_mask)
+            causal_mask *= diagonal_attend_mask
+            causal_mask = causal_mask[None, None, :, :].expand(batch_size, 1, -1, -1)
+            if attention_mask is not None:
+                causal_mask = causal_mask.clone()  # copy to contiguous memory for in-place edit
+                if attention_mask.shape[-1] > target_length:
+                    attention_mask = attention_mask[:, :target_length]
+                mask_length = attention_mask.shape[-1]
+                padding_mask = causal_mask[:, :, :, :mask_length] + attention_mask[:, None, None, :].to(
+                    causal_mask.device
+                )
+                padding_mask = padding_mask == 0
+                causal_mask[:, :, :, :mask_length] = causal_mask[:, :, :, :mask_length].masked_fill(
+                    padding_mask, min_dtype
+                )
+        return causal_mask
+
+
+class KwargsForCausalLM(FlashAttentionKwargs, ): ...
+
+
+class Qwen2ForCausalLM(Qwen2PreTrainedModel,  MDMGenerationMixin):
+    _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 = Qwen2Model(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    def get_input_embeddings(self):
+        return self.model.embed_tokens
+
+    def set_input_embeddings(self, value):
+        self.model.embed_tokens = value
+
+    def get_output_embeddings(self):
+        return self.lm_head
+
+    def set_output_embeddings(self, new_embeddings):
+        self.lm_head = new_embeddings
+
+    def set_decoder(self, decoder):
+        self.model = decoder
+
+    def get_decoder(self):
+        return self.model
+
+    @add_start_docstrings_to_model_forward(QWEN2_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Union[Cache, List[torch.FloatTensor]]] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        mask_ratio: Optional[torch.FloatTensor]=None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        is_causal: bool = True,
+        **kwargs: Unpack[KwargsForCausalLM],
+    ) -> Union[Tuple, CausalLMOutputWithPast]:
+        r"""
+        Args:
+            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]`.
+
+            logits_to_keep (`int` or `torch.Tensor`, *optional*):
+                If an `int`, compute logits for the last `logits_to_keep` tokens. If `0`, calculate logits for all
+                `input_ids` (special case). Only last token logits are needed for generation, and calculating them only for that
+                token can save memory, which becomes pretty significant for long sequences or large vocabulary size.
+                If a `torch.Tensor`, must be 1D corresponding to the indices to keep in the sequence length dimension.
+                This is useful when using packed tensor format (single dimension for batch and sequence length).
+
+        Returns:
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, Qwen2ForCausalLM
+
+        >>> model = Qwen2ForCausalLM.from_pretrained("meta-qwen2/Qwen2-2-7b-hf")
+        >>> tokenizer = AutoTokenizer.from_pretrained("meta-qwen2/Qwen2-2-7b-hf")
+
+        >>> 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_hidden_states = (
+            output_hidden_states if output_hidden_states is not None else self.config.output_hidden_states
+        )
+        return_dict = return_dict if return_dict is not None else self.config.use_return_dict
+
+        if not get_parallel_state().sp_enabled and labels is not None:
+            # Shift so that tokens < n predict n
+            labels = labels[..., 1:].contiguous()
+            labels = labels.view(-1)
+            if (
+                position_ids is not None
+                and position_ids.size(0) == 1
+                and not (torch.diff(position_ids, dim=-1) >= 0).all()
+            ):
+                position_ids_ = position_ids.flatten()
+                indices_q = torch.arange(position_ids_.size(0), device=position_ids_.device, dtype=torch.int32)
+                cu_seq_lens = torch.cat(
+                    (
+                        indices_q[position_ids_ == 0],
+                        torch.tensor(position_ids_.size(), device=position_ids_.device, dtype=torch.int32),
+                    )
+                )
+                labels[cu_seq_lens[1:-1] - 1] = IGNORE_INDEX
+        if mask_ratio is not None:
+            is_causal = False
+            mask_ratio = mask_ratio[..., 1:].contiguous()
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs = 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,
+            return_dict=return_dict,
+            cache_position=cache_position,
+            is_causal=is_causal,
+            **kwargs,
+        )
+
+        hidden_states = outputs[0]
+        # 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
+        hidden_states = hidden_states[:, slice_indices, :]
+
+        loss = None
+        logits = None
+        if labels is not None:
+            labels = labels.view(-1)  # flatten label
+            if is_liger_kernel_available():
+                if mask_ratio is not None:
+                    loss_fct = LigerFusedLinearCrossEntropyLoss(reduction="none",ignore_index=IGNORE_INDEX)
+                    if not get_parallel_state().sp_enabled:
+                        # Shift so that tokens < n predict n
+                        hidden_states = hidden_states[..., :-1, :].contiguous()
+                    loss = loss_fct(
+                        self.lm_head.weight, 
+                        hidden_states.view(-1, self.config.hidden_size),
+                        labels
+                    )
+                    path_loss = (-loss).exp().detach() * loss
+                    loss = loss + path_loss
+                    loss_mask = labels != IGNORE_INDEX
+                    loss = (loss * loss_mask * (1/mask_ratio)).sum() / (loss_mask.sum() + 1e-8)
+                else:
+                    loss_fct = LigerFusedLinearCrossEntropyLoss(reduction="mean")
+                    if not get_parallel_state().sp_enabled:
+                        # Shift so that tokens < n predict n
+                        hidden_states = hidden_states[..., :-1, :].contiguous()
+                    hidden_states = hidden_states.view(-1, self.config.hidden_size)
+                    loss = loss_fct(self.lm_head.weight, hidden_states, labels)
+            else:
+                if mask_ratio is not None:
+                    loss_fct = torch.nn.CrossEntropyLoss(reduction="none")
+                    logits = self.lm_head(hidden_states)
+                    logits = logits.view(-1, self.vocab_size)
+                    loss = loss_fct(logits, labels.view(-1))
+                    path_loss = (-loss).exp().detach() * loss
+                    loss = loss + path_loss
+                    loss_mask = labels != IGNORE_INDEX
+                    loss = (loss * loss_mask * (1/mask_ratio)).sum() / (loss_mask.sum() + 1e-8)
+                else:
+                    loss_fct = torch.nn.CrossEntropyLoss(reduction="mean")
+                    logits = self.lm_head(hidden_states)
+                    # Upcast to float if we need to compute the loss to avoid potential precision issues
+                    logits = logits.float()
+                    if not get_parallel_state().sp_enabled:
+                        # Shift so that tokens < n predict n
+                        logits = logits[..., :-1, :].contiguous()
+
+                    # Flatten the tokens
+                    logits = logits.view(-1, self.vocab_size)
+                    loss = loss_fct(logits, labels)
+
+            if get_parallel_state().sp_enabled:
+                num_valid_tokens = (labels != IGNORE_INDEX).sum()
+                loss = reduce_sequence_parallel_loss(loss, num_valid_tokens)
+        else:
+            logits = self.lm_head(hidden_states)
+
+        if not return_dict:
+            output = (logits,) + outputs[1:]
+            return (loss,) + output if loss is not None else output
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+    
+
+
+
+import torch
+from tqdm import tqdm
+from typing import Callable, Tuple, Any
+
+
+def topk_masking(scores: torch.Tensor, cutoff_len: torch.Tensor, mode: str = "lowest") -> torch.Tensor:
+    """Generate a mask selecting the top-k lowest or highest elements per row."""
+    sorted_scores = scores.sort(dim=-1, descending=(mode == "highest")).values
+    cutoff = sorted_scores.gather(dim=-1, index=cutoff_len)
+    return (scores >= cutoff) if mode == "highest" else (scores < cutoff)
+
+
+def sample_categorical(
+    logits: torch.Tensor, temperature: float = 1.0, noise_scale: float = 1.0
+) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]:
+    """
+    Sample from a categorical distribution with optional Gumbel noise.
+    Returns sampled tokens, their scores, and the noised logits.
+    """
+    logits = logits.to(torch.float64)
+    if temperature > 0:
+        gumbel_noise = -torch.log(-torch.log(torch.rand_like(logits) + 1e-8) + 1e-8)
+        logits = logits / temperature + noise_scale * gumbel_noise
+    log_probs = logits.log_softmax(dim=-1)
+    scores, tokens = log_probs.max(dim=-1)
+    return tokens, scores.to(logits.dtype), logits.to(logits.dtype)
+
+
+@torch.inference_mode()
+@torch.amp.autocast(device_type="cuda", dtype=torch.float16)
+def p2_sampling(
+    xt: torch.Tensor,
+    model: Any,
+    mask_id: int,
+    num_steps: int,
+    tau: float = 1.0,
+    kappa_fn: Callable[[float], float] = lambda t: t,
+    eta: float = 1.0,
+    **kwargs
+) -> torch.Tensor:
+    """
+    P2 Sampling implementation for discrete diffusion models.
+    Reference: https://arxiv.org/pdf/2502.03540
+    """
+    dt = 1 / num_steps
+    fix_mask = (xt != mask_id)
+
+    for i in tqdm(range(1, num_steps + 1)):
+        t = i * dt
+        kappa_t = kappa_fn(t)
+
+        logits = model(xt).double()
+        last_mask = (xt == mask_id)
+        unmask_t = ~last_mask & ~fix_mask
+
+        x0, score, _ = sample_categorical(logits, temperature=tau)
+        score = score.masked_fill(fix_mask, float("inf"))
+        score[unmask_t] *= eta
+
+        num_to_mask = ((~fix_mask).sum(dim=1, keepdim=True).float() * (1 - kappa_t)).long()
+        to_mask = topk_masking(score, num_to_mask, mode="lowest")
+
+        xt[to_mask] = mask_id
+        mask_2_x0 = last_mask & ~to_mask
+        xt[mask_2_x0] = x0[mask_2_x0]
+
+    xt[xt == mask_id] = x0[xt == mask_id]
+    return xt
+
+
+
+if is_liger_kernel_available():
+    apply_rotary_pos_emb = liger_rotary_pos_emb
+    Qwen2RMSNorm = LigerRMSNorm
+    Qwen2MLP = LigerSwiGLUMLP
+    logger.info_rank0("Apply liger kernel to Qwen2.")
+
+
+ModelClass = Qwen2ForCausalLM
+
+__all__ = ["Qwen2ForCausalLM", "Qwen2Model", "Qwen2PreTrainedModel"]

special_tokens_map.json

@@ -0,0 +1,38 @@
+{
+  "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|>"
+  ],
+  "eos_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "<M>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4f3ead5ffbb2abfb9e51757b104af5174f59c994ff3b2ca715d344dc66ee1f4e
+size 11422076

tokenizer_config.json

@@ -0,0 +1,218 @@
+{
+  "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": "<M>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "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    {%- else %}\n        {{- 'You are a helpful assistant.' }}\n    {%- endif %}\n    {{- \"\\n\\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    {%- else %}\n        {{- '<|im_start|>system\\nYou are a helpful assistant.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) 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{%- endif %}\n",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|endoftext|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "mask_token": "<M>",
+  "model_max_length": 32768,
+  "pad_token": "<|endoftext|>",
+  "padding_side": "right",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}