Add files using upload-large-folder tool

.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
+figures/joyai-logo.png filter=lfs diff=lfs merge=lfs -text

LICENSE

@@ -0,0 +1,27 @@
+Copyright (c) 2026 JD AI
+
+We offer you a license similar to the MIT License. In the event that the Software (or any derivative works
+thereof) is incorporated into 1) any of your commercial products or services; or 2) any of your products or
+services that either have more than 100 million monthly active users or generate more than 20 million US dollars
+(or equivalent in other currencies) in monthly revenue, you are required to conspicuously display "JoyAI-LLM Flash" 
+on the user interface of such product or service.
+
+================================================================================
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the “Software”), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -0,0 +1,374 @@
+<div align="center">
+  <picture>
+      <img src="figures/joyai-logo.png" width="30%" alt="JoyAI-LLM Flash">
+  </picture>
+</div>
+<hr>
+
+<div align="center" style="line-height: 1;">
+  <a href="https://huggingface.co/jdopensource" target="_blank"><img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-JD-ffc107?color=ffc107&logoColor=white"/></a>
+  <a href="LICENSE"><img alt="License" src="https://img.shields.io/badge/License-Modified_MIT-f5de53?&color=f5de53"/></a>
+</div>
+
+
+
+
+## 1. Model Introduction
+
+JoyAI-LLM Flash is a state-of-the-art medium-sized instruct language model with 3 billion activated parameters and 48 billion total parameters. JoyAI-LLM Flash was pretrained on 20 trillion text tokens using Muon optimizer, followed by large-scale supervised fine-tuning (SFT), direct preference optimization (DPO), and reinforcement learning (RL) across diverse environments. JoyAI-LLM Flash achieves strong performance across frontier knowledge, reasoning, coding tasks and agentic capabilities.
+
+### Key Features
+
+- Fiber Bundle RL: invole geometric manifold theory into reinforcement learning, proposing an innovative technique known as FiberPO. This approach is designed to address the growing trends of increasing heterogeneous agent scales.
+- Training-Inference Collaboration: apply Muon optimizer with dense MTP, develop novel optimization techniques to resolve instabilities while scaling up, delivering 1.3× to 1.7× the throughput of the non-MTP version.
+- Agentic Intelligence: designed for tool use, reasoning, and autonomous problem-solving.
+
+## 2. Model Summary
+
+|                                             |                          |
+| :-----------------------------------------: | :----------------------: |
+|              **Architecture**               | Mixture-of-Experts (MoE) |
+|            **Total Parameters**             |           48B            |
+|          **Activated Parameters**           |            3B            |
+| **Number of Layers** (Dense layer included) |            40            |
+|         **Number of Dense Layers**          |            1             |
+|       **Attention Hidden Dimension**        |           2048           |
+|    **MoE Hidden Dimension** (per Expert)    |           768            |
+|        **Number of Attention Heads**        |            32            |
+|            **Number of Experts**            |           256            |
+|       **Selected Experts per Token**        |            8             |
+|        **Number of Shared Experts**         |            1             |
+|             **Vocabulary Size**             |           129K           |
+|             **Context Length**              |           128K           |
+|           **Attention Mechanism**           |           MLA            |
+|           **Activation Function**           |          SwiGLU          |
+|                   </div>                    |                          |
+
+
+## 3. Evaluation Results
+
+<table>
+<thead>
+<tr>
+<th align="center">Benchmark</th>
+<th align="center"><sup>JoyAI-LLM Flash</sup></th>
+<th align="center"><sup>Qwen3-30B-A3B-Instuct-2507</sup></th>
+<th align="center"><sup>GLM-4.7-Flash<br>(Non-thinking)</sup></th>
+</tr>
+</thead>
+<tbody>
+
+
+<tr>
+<td align="center" colspan=8><strong>Knowledge &amp; Alignment</strong></td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">MMLU</td>
+<td align="center" style="vertical-align: middle"><strong>89.50</strong></td>
+<td align="center" style="vertical-align: middle">86.87</td>
+<td align="center" style="vertical-align: middle">80.53</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">MMLU-Pro</td>
+<td align="center" style="vertical-align: middle"><strong>81.02</strong></td>
+<td align="center" style="vertical-align: middle">73.88</td>
+<td align="center" style="vertical-align: middle">63.62</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">CMMLU</td>
+<td align="center" style="vertical-align: middle"><strong>87.03</strong></td>
+<td align="center" style="vertical-align: middle">85.88</td>
+<td align="center" style="vertical-align: middle">75.85</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">GPQA-Diamond</td>
+<td align="center" style="vertical-align: middle"><strong>74.43</strong></td>
+<td align="center" style="vertical-align: middle">68.69</td>
+<td align="center" style="vertical-align: middle">39.90</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">SuperGPQA</td>
+<td align="center" style="vertical-align: middle"><strong>55.00</strong></td>
+<td align="center" style="vertical-align: middle">52.00</td>
+<td align="center" style="vertical-align: middle">32.00</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">LiveBench</td>
+<td align="center" style="vertical-align: middle"><strong>72.90</strong></td>
+<td align="center" style="vertical-align: middle">59.70</td>
+<td align="center" style="vertical-align: middle">43.10</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">IFEval</td>
+<td align="center" style="vertical-align: middle"><strong>86.69</strong></td>
+<td align="center" style="vertical-align: middle">83.18</td>
+<td align="center" style="vertical-align: middle">82.44</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">AlignBench</td>
+<td align="center" style="vertical-align: middle"><strong>8.24</strong></td>
+<td align="center" style="vertical-align: middle">8.07</td>
+<td align="center" style="vertical-align: middle">6.85</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">HellaSwag</td>
+<td align="center" style="vertical-align: middle"><strong>91.79</strong></td>
+<td align="center" style="vertical-align: middle">89.90</td>
+<td align="center" style="vertical-align: middle">60.84</td>
+</tr>
+
+<tr>
+<td align="center" colspan=8><strong>Coding</strong></td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">HumanEval</td>
+<td align="center" style="vertical-align: middle"><strong>96.34</strong></td>
+<td align="center" style="vertical-align: middle">95.12</td>
+<td align="center" style="vertical-align: middle">74.39</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">LiveCodeBench</td>
+<td align="center" style="vertical-align: middle"><strong>65.60</strong></td>
+<td align="center" style="vertical-align: middle">39.71</td>
+<td align="center" style="vertical-align: middle">27.43</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">SciCode</td>
+<td align="center" style="vertical-align: middle"><strong>3.08/22.92</strong></td>
+<td align="center" style="vertical-align: middle"><strong>3.08/22.92</strong></td>
+<td align="center" style="vertical-align: middle">3.08/15.11</td>
+</tr>
+<tr>
+<td align="center" colspan=8><strong>Mathematics</strong></td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">GSM8K</td>
+<td align="center" style="vertical-align: middle"><strong>95.83</strong></td>
+<td align="center" style="vertical-align: middle">79.83</td>
+<td align="center" style="vertical-align: middle">81.88</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">AIME2025</td>
+<td align="center" style="vertical-align: middle"><strong>65.83</strong></td>
+<td align="center" style="vertical-align: middle">62.08</td>
+<td align="center" style="vertical-align: middle">24.17</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">MATH 500</td>
+<td align="center" style="vertical-align: middle"><strong>97.10</strong></td>
+<td align="center" style="vertical-align: middle">89.80</td>
+<td align="center" style="vertical-align: middle">90.90</td>
+</tr>
+
+<tr>
+<td align="center" colspan=8><strong>Agentic</strong></td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">SWE-bench Verified</td>
+<td align="center" style="vertical-align: middle"><strong>60.60</strong></td>
+<td align="center" style="vertical-align: middle">24.44</td>
+<td align="center" style="vertical-align: middle">51.60</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">Tau2-Retail</td>
+<td align="center" style="vertical-align: middle"><strong>67.55</strong></td>
+<td align="center" style="vertical-align: middle">53.51</td>
+<td align="center" style="vertical-align: middle">62.28</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">Tau2-Airline</td>
+<td align="center" style="vertical-align: middle"><strong>54.00</strong></td>
+<td align="center" style="vertical-align: middle">32.00</td>
+<td align="center" style="vertical-align: middle">52.00</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">Tau2-Telecom</td>
+<td align="center" style="vertical-align: middle">79.83</td>
+<td align="center" style="vertical-align: middle">4.39</td>
+<td align="center" style="vertical-align: middle"><strong>88.60</strong></td>
+</tr>
+
+<tr>
+<td align="center" colspan=8><strong>Long Context</strong></td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle">RULER</td>
+<td align="center" style="vertical-align: middle"><strong>95.60</strong></td>
+<td align="center" style="vertical-align: middle">89.66</td>
+<td align="center" style="vertical-align: middle">56.12</td>
+</tr>
+</tbody>
+</table>
+
+
+## 4. Deployment
+
+> [!Note]
+> You can access JoyAI-LLM Flash API on https://docs.jdcloud.com/cn/jdaip/chat and we provide OpenAI/Anthropic-compatible API for you. 
+> Currently, JoyAI-LLM Flash is recommended to run on the following inference engines:
+
+* vLLM
+* SGLang
+
+The minimum version requirement for `transformers` is `4.57.1`.
+
+Deployment examples can be found in the [Model Deployment Guide](docs/deploy_guidance.md).
+
+
+
+## 5. Model Usage
+
+The usage demos below demonstrate how to call our official API.
+
+For third-party APIs deployed with vLLM or SGLang, please note that:
+
+> [!Note] Recommended sampling parameters: `temperature=0.6`, `top_p=1.0`
+
+### Chat Completion
+
+This is a simple chat completion script which shows how to call JoyAI-Flash API.
+
+```python
+from openai import OpenAI
+
+client = OpenAI(base_url="http://IP:PORT/v1", api_key="EMPTY")
+
+
+def simple_chat(client: OpenAI):
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {
+                    "type": "text",
+                    "text": "which one is bigger, 9.11 or 9.9? think carefully.",
+                }
+            ],
+        },
+    ]
+    model_name = client.models.list().data[0].id
+    response = client.chat.completions.create(
+        model=model_name, messages=messages, stream=False, max_tokens=4096
+    )
+    print(f"response: {response.choices[0].message.content}")
+
+
+if __name__ == "__main__":
+    simple_chat(client)
+```
+
+
+### Tool call Completion
+
+This is a simple toll call completion script which shows how to call JoyAI-Flash API.
+
+```python
+import json
+
+from openai import OpenAI
+
+client = OpenAI(base_url="http://IP:PORT/v1", api_key="EMPTY")
+
+
+def my_calculator(expression: str) -> str:
+    return str(eval(expression))
+
+
+def rewrite(expression: str) -> str:
+    return str(expression)
+
+
+def simple_tool_call(client: OpenAI):
+    messages = [
+        {
+            "role": "user",
+            "content": [
+                {
+                    "type": "text",
+                    "text": "use my functions to compute the results for the equations: 6+1",
+                },
+            ],
+        },
+    ]
+    tools = [
+        {
+            "type": "function",
+            "function": {
+                "name": "my_calculator",
+                "description": "A calculator that can evaluate a mathematical equation and compute its results.",
+                "parameters": {
+                    "type": "object",
+                    "properties": {
+                        "expression": {
+                            "type": "string",
+                            "description": "The mathematical expression to evaluate.",
+                        },
+                    },
+                    "required": ["expression"],
+                },
+            },
+        },
+        {
+            "type": "function",
+            "function": {
+                "name": "rewrite",
+                "description": "Rewrite a given text for improved clarity",
+                "parameters": {
+                    "type": "object",
+                    "properties": {
+                        "text": {
+                            "type": "string",
+                            "description": "The input text to rewrite",
+                        }
+                    },
+                },
+            },
+        },
+    ]
+    model_name = client.models.list().data[0].id
+    response = client.chat.completions.create(
+        model=model_name,
+        messages=messages,
+        temperature=1.0,
+        max_tokens=1024,
+        tools=tools,
+        tool_choice="auto",
+    )
+    tool_calls = response.choices[0].message.tool_calls
+
+    results = []
+    for tool_call in tool_calls:
+        function_name = tool_call.function.name
+        function_args = tool_call.function.arguments
+        if function_name == "my_calculator":
+            result = my_calculator(**json.loads(function_args))
+            results.append(result)
+    messages.append({"role": "assistant", "tool_calls": tool_calls})
+    for tool_call, result in zip(tool_calls, results):
+        messages.append(
+            {
+                "role": "tool",
+                "tool_call_id": tool_call.id,
+                "name": tool_call.function.name,
+                "content": result,
+            }
+        )
+    response = client.chat.completions.create(
+        model=model_name,
+        messages=messages,
+        temperature=1.0,
+        max_tokens=1024,
+    )
+    print(response.choices[0].message.content)
+
+
+if __name__ == "__main__":
+    simple_tool_call(client)
+
+```
+
+---
+
+## 6. License
+
+Both the code repository and the model weights are released under the [Modified MIT License](LICENSE).

chat_template.jinja

@@ -0,0 +1,103 @@
+{%- macro render_extra_keys(json_dict, handled_keys) -%}
+    {%- if json_dict is mapping -%}
+        {%- for json_key in json_dict if json_key not in handled_keys -%}
+            {%- if json_dict[json_key] is mapping or (json_dict[json_key] is sequence and json_dict[json_key] is not string) -%}
+                {{- '\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | tojson | safe) ~ '</' ~ json_key ~ '>' -}}
+            {%- else -%}
+                {{- '\n<' ~ json_key ~ '>' ~ (json_dict[json_key] | string) ~ '</' ~ json_key ~ '>' -}}
+            {%- endif -%}
+        {%- endfor -%}
+    {%- endif -%}
+{%- endmacro -%}
+
+{%- if not add_generation_prompt is defined -%}{%- set add_generation_prompt = false -%}{%- endif -%}
+
+{%- set ns = namespace(system_prompt='', is_first_sp=true, is_last_user=false) -%}
+{%- set default_system = "You are JoyAI , a large language model trained by JD（京东）that can interact with a computer to solve tasks. Answer as concisely as possible." -%}
+{%- set ns.system_prompt = default_system -%}
+
+{%- for message in messages -%}
+    {%- if message['role'] == 'system' -%}
+        {%- if ns.is_first_sp -%}
+            {%- set ns.system_prompt = message['content'] -%}
+            {%- set ns.is_first_sp = false -%}
+        {%- else -%}
+            {%- set ns.system_prompt = ns.system_prompt + '\n\n' + message['content'] -%}
+        {%- endif -%}
+    {%- endif -%}
+{%- endfor -%}
+
+{{- bos_token -}}{{- ns.system_prompt -}}
+{%- if tools is iterable and tools | length > 0 -%}
+    {{- "\n\n# Tools\n\nYou have access to the following functions:\n\n" }}
+    {{- "<tools>" }}
+    {%- for tool in tools %}
+        {%- if tool.function is defined %}
+            {%- set tool = tool.function %}
+        {%- endif %}
+        {{- "\n<function>\n<name>" ~ tool.name ~ "</name>" }}
+        {%- if tool.description is defined %}
+            {{- '\n<description>' ~ (tool.description | trim) ~ '</description>' }}
+        {%- endif %}
+        {{- '\n<parameters>' }}
+        {%- if tool.parameters is defined and tool.parameters is mapping and tool.parameters.properties is defined and tool.parameters.properties is mapping %}
+            {%- for param_name, param_fields in tool.parameters.properties|items %}
+                {{- '\n<parameter>' }}
+                {{- '\n<name>' ~ param_name ~ '</name>' }}
+                {%- if param_fields.type is defined %}
+                    {{- '\n<type>' ~ (param_fields.type | string) ~ '</type>' }}
+                {%- endif %}
+                {%- if param_fields.description is defined %}
+                    {{- '\n<description>' ~ (param_fields.description | trim) ~ '</description>' }}
+                {%- endif %}
+                {%- set handled_keys = ['name', 'type', 'description'] %}
+                {{- render_extra_keys(param_fields, handled_keys) }}
+                {{- '\n</parameter>' }}
+            {%- endfor %}
+        {%- endif %}
+        {% set handled_keys = ['type', 'properties'] %}
+        {{- render_extra_keys(tool.parameters, handled_keys) }}
+        {{- '\n</parameters>' }}
+        {%- set handled_keys = ['type', 'name', 'description', 'parameters'] %}
+        {{- render_extra_keys(tool, handled_keys) }}
+        {{- '\n</function>' }}
+    {%- endfor %}
+    {{- "\n</tools>" }}
+    {{- '\n\nIf you choose to call a function ONLY reply in the following format with NO suffix:\n\n<tool_call>\n<function=example_function_name>\n<parameter=example_parameter_1>\nvalue_1\n</parameter>\n<parameter=example_parameter_2>\nThis is the value for the second parameter\nthat can span\nmultiple lines\n</parameter>\n</function>\n</tool_call>\n\n<IMPORTANT>\nReminder:\n- Function calls MUST follow the specified format: an inner <function=...></function> block must be nested within <tool_call></tool_call> XML tags\n- Required parameters MUST be specified\n- You may provide optional reasoning for your function call in natural language BEFORE the function call, but NOT after\n- If there is no function call available, answer the question like normal with your current knowledge and do not tell the user about function calls\n</IMPORTANT>' }}
+{%- endif %}
+{%- for message in messages -%}
+    {%- if message['role'] == 'user' -%}
+        {%- set ns.is_last_user = true -%}
+        {{- '<|User|>' + message['content'] -}}
+    {%- elif message['role'] == 'assistant' -%}
+        {%- if ns.is_last_user -%}
+            {{ '<|Assistant|>' }}
+        {%- endif -%}
+        {%- set ns.is_last_user = false -%}
+        {%- set content = message.get('content') | default('', true) -%}
+        {{ '<|end_of_thought|>' + content }}
+        {%- if message['tool_calls'] is defined and message['tool_calls'] is not none -%}
+            {%- for tool in message['tool_calls'] -%}
+                {%- if tool.function is defined %}{% set tool = tool.function %}{% endif -%}
+                {{- '\n<tool_call>\n<function=' + tool.name + '>\n' -}}
+                {%- if tool.arguments is defined -%}
+                    {%- if tool.arguments is string -%}{%- set args_data = tool.arguments | from_json -%}{%- else -%}{%- set args_data = tool.arguments -%}{%- endif -%}
+                    {%- for args_name, args_value in args_data.items() -%}
+                        {{- '<parameter=' + args_name + '>\n' -}}
+                        {%- set args_value = args_value | tojson | safe if args_value is mapping or (args_value is sequence and args_value is not string) else args_value | string -%}
+                        {{- args_value -}}{{- '\n</parameter>\n' -}}
+                    {%- endfor -%}
+                {%- endif -%}
+                {{- '</function>\n</tool_call>' -}}
+            {%- endfor -%}
+        {%- endif -%}
+        {{ '<|end▁of▁sentence|>' }}
+    {%- elif message['role'] == 'tool' -%}
+        {%- set ns.is_last_user = true -%}
+        {{ '\n<tool_response>\n' + message['content'] + '\n</tool_response>' }}
+    {%- endif -%}
+{%- endfor -%}
+
+{%- if add_generation_prompt -%}
+    {{ '<|Assistant|>' }}{{ '<|end_of_thought|>' }}
+{%- endif -%}

config.json

@@ -0,0 +1,49 @@
+{
+  "architectures": [
+    "DeepseekV3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_deepseek.DeepseekV3Config",
+    "AutoModel": "modeling_deepseek.DeepseekV3Model",
+    "AutoModelForCausalLM": "modeling_deepseek.DeepseekV3ForCausalLM"
+  },
+  "bos_token_id": 0,
+  "eos_token_id": 1,
+  "ep_size": 1,
+  "first_k_dense_replace": 1,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 7168,
+  "kv_lora_rank": 512,
+  "max_position_embeddings": 131072,
+  "model_type": "joyai_llm_flash",
+  "moe_intermediate_size": 768,
+  "moe_layer_freq": 1,
+  "n_group": 1,
+  "n_routed_experts": 256,
+  "n_shared_experts": 1,
+  "norm_topk_prob": true,
+  "num_attention_heads": 32,
+  "num_experts_per_tok": 8,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 32,
+  "num_nextn_predict_layers": 1,
+  "q_lora_rank": 1536,
+  "qk_nope_head_dim": 128,
+  "qk_rope_head_dim": 64,
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 32000000,
+  "routed_scaling_factor": 2.5,
+  "scoring_func": "sigmoid",
+  "tie_word_embeddings": false,
+  "topk_group": 1,
+  "topk_method": "noaux_tc",
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.44.2",
+  "use_cache": true,
+  "v_head_dim": 128,
+  "vocab_size": 129280
+}

configuration.json

@@ -0,0 +1 @@
+{"framework":"Pytorch","task":"text-generation"}

configuration_deepseek.py

@@ -0,0 +1,247 @@
+# coding=utf-8
+# Copyright 2025 bzantium and the HuggingFace Inc. team. All rights reserved.
+#
+# This code is based on the DeepSeekV3 implementations from the DeepSeek AI team. (https://huggingface.co/deepseek-ai/DeepSeek-V3)
+
+# 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.
+"""DeepSeekV3 model configuration"""
+
+from transformers.configuration_utils import PretrainedConfig
+from transformers.modeling_rope_utils import rope_config_validation
+
+
+DEEPSEEK_PRETRAINED_CONFIG_ARCHIVE_MAP = {}
+
+
+class DeepseekV3Config(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`DeepseekV3Model`]. It is used to instantiate an DeepSeek
+    model according to the specified arguments, defining the model architecture. Instantiating a configuration with the
+    defaults will yield a similar configuration to that of the DeepSeek-V3.
+    e.g. [bzantium/tiny-deepseek-v3](https://huggingface.co/bzantium/tiny-deepseek-v3)
+    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 129280):
+            Vocabulary size of the Deep model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`DeepseekV3Model`]
+        hidden_size (`int`, *optional*, defaults to 7168):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 18432):
+            Dimension of the MLP representations.
+        moe_intermediate_size (`int`, *optional*, defaults to 2048):
+            Dimension of the MoE representations.
+        num_hidden_layers (`int`, *optional*, defaults to 61):
+            Number of hidden layers in the Transformer decoder.
+        num_attention_heads (`int`, *optional*, defaults to 128):
+            Number of attention heads for each attention layer in the Transformer decoder.
+        num_key_value_heads (`int`, *optional*, defaults to 128):
+            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 checkout [this
+            paper](https://arxiv.org/pdf/2305.13245.pdf). If it is not specified, will default to
+            `num_attention_heads`.
+        n_shared_experts (`int`, *optional*, defaults to 1):
+            Number of shared experts.
+        n_routed_experts (`int`, *optional*, defaults to 256):
+            Number of routed experts.
+        routed_scaling_factor (`float`, *optional*, defaults to 2.5):
+            Scaling factor or routed experts.
+        kv_lora_rank (`int`, *optional*, defaults to 512):
+            Rank of the LoRA matrices for key and value projections.
+        q_lora_rank (`int`, *optional*, defaults to 1536):
+            Rank of the LoRA matrices for query projections.
+        qk_rope_head_dim (`int`, *optional*, defaults to 64):
+            Dimension of the query/key heads that use rotary position embeddings.
+        v_head_dim (`int`, *optional*, defaults to 128):
+            Dimension of the value heads.
+        qk_nope_head_dim (`int`, *optional*, defaults to 128):
+            Dimension of the query/key heads that don't use rotary position embeddings.
+        n_group (`int`, *optional*, defaults to 8):
+            Number of groups for routed experts.
+        topk_group (`int`, *optional*, defaults to 4):
+            Number of selected groups for each token(for each token, ensuring the selected experts is only within `topk_group` groups).
+        num_experts_per_tok (`int`, *optional*, defaults to 8):
+            Number of selected experts, None means dense model.
+        first_k_dense_replace (`int`, *optional*, defaults to 3):
+            Number of dense layers in shallow layers(embed->dense->dense->...->dense->moe->moe...->lm_head).
+                                                            \--k dense layers--/
+        norm_topk_prob (`bool`, *optional*, defaults to `True`):
+            Whether to normalize the weights of the routed experts.
+        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 4096):
+            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`.
+        pad_token_id (`int`, *optional*):
+            Padding token id.
+        bos_token_id (`int`, *optional*, defaults to 0):
+            Beginning of stream token id.
+        eos_token_id (`int`, *optional*, defaults to 1):
+            End of stream token id.
+        pretraining_tp (`int`, *optional*, defaults to 1):
+            Experimental feature. Tensor parallelism rank used during pretraining. Please refer to [this
+            document](https://huggingface.co/docs/transformers/parallelism) to understand more about it. This value is
+            necessary to ensure exact reproducibility of the pretraining results. Please refer to [this
+            issue](https://github.com/pytorch/pytorch/issues/76232).
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether to tie weight embeddings
+        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. Currently supports two scaling
+            strategies: linear and dynamic. Their scaling factor must be a float greater than 1. The expected format is
+            `{"type": strategy name, "factor": scaling factor}`. When using this flag, don't update
+            `max_position_embeddings` to the expected new maximum.
+        rope_interleave (`bool`, *optional*, defaults to `True`):
+            Whether to interleave the rotary position embeddings.
+        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.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+
+    ```python
+    >>> from transformers import DeepseekV3Model, DeepseekV3Config
+
+    >>> # Initializing a Deepseek-V3 style configuration
+    >>> configuration = DeepseekV3Config()
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "deepseek_v3"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    base_model_tp_plan = {  # TODO: only replicate attention layers when > first_k_dense_replace
+        "layers.*.mlp.experts.*.gate_proj": "local_colwise",
+        "layers.*.mlp.experts.*.up_proj": "local_colwise",
+        "layers.*.mlp.experts.*.down_proj": "local_rowwise",
+        "layers.*.mlp.experts.*": "local",  # each expert is wrapped in a module list
+        "layers.*.mlp.shared_experts.gate_proj": "local_colwise",
+        "layers.*.mlp.shared_experts.up_proj": "local_colwise",
+        "layers.*.mlp.shared_experts.down_proj": "local_rowwise",
+        "layers.*.mlp.shared_experts": "local",
+        "layers.*.mlp.gate_proj": "local_colwise",
+        "layers.*.mlp.up_proj": "local_colwise",
+        "layers.*.mlp.down_proj": "local_rowwise",
+        "layers.*.mlp": "gather",  # This is the only moment where results are gathered
+    }
+    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=129280,
+        hidden_size=7168,
+        intermediate_size=18432,
+        moe_intermediate_size=2048,
+        num_hidden_layers=61,
+        num_attention_heads=128,
+        num_key_value_heads=128,
+        n_shared_experts=1,
+        n_routed_experts=256,
+        routed_scaling_factor=2.5,
+        kv_lora_rank=512,
+        q_lora_rank=1536,
+        qk_rope_head_dim=64,
+        v_head_dim=128,
+        qk_nope_head_dim=128,
+        n_group=8,
+        topk_group=4,
+        num_experts_per_tok=8,
+        first_k_dense_replace=3,
+        norm_topk_prob=True,
+        hidden_act="silu",
+        max_position_embeddings=4096,
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=0,
+        eos_token_id=1,
+        pretraining_tp=1,
+        tie_word_embeddings=False,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        rope_interleave=True,
+        attention_bias=False,
+        attention_dropout=0.0,
+        **kwargs,
+    ):
+        self.vocab_size = vocab_size
+        self.max_position_embeddings = max_position_embeddings
+        self.hidden_size = hidden_size
+        self.intermediate_size = intermediate_size
+        self.moe_intermediate_size = moe_intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.n_shared_experts = n_shared_experts
+        self.n_routed_experts = n_routed_experts
+        self.routed_scaling_factor = routed_scaling_factor
+        self.kv_lora_rank = kv_lora_rank
+        self.q_lora_rank = q_lora_rank
+        self.qk_rope_head_dim = qk_rope_head_dim
+        self.v_head_dim = v_head_dim
+        self.qk_nope_head_dim = qk_nope_head_dim
+        self.qk_head_dim = qk_nope_head_dim + qk_rope_head_dim
+        self.head_dim = qk_rope_head_dim
+        self.n_group = n_group
+        self.topk_group = topk_group
+        self.num_experts_per_tok = num_experts_per_tok
+        self.first_k_dense_replace = first_k_dense_replace
+        self.norm_topk_prob = norm_topk_prob
+        self.rope_interleave = rope_interleave
+
+        # for backward compatibility
+        if num_key_value_heads is None:
+            num_key_value_heads = num_attention_heads
+
+        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.pretraining_tp = pretraining_tp
+        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, copy it 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)
+
+        super().__init__(
+            pad_token_id=pad_token_id,
+            bos_token_id=bos_token_id,
+            eos_token_id=eos_token_id,
+            tie_word_embeddings=tie_word_embeddings,
+            **kwargs,
+        )
+
+
+__all__ = ["DeepseekV3Config"]

docs/deploy_guidance.md

@@ -0,0 +1,29 @@
+# JoyAI-LLM Flash Deployment Guide
+
+> [!Note]
+> This guide offers a selection of deployment command examples for JoyAI-Flash, which may not be the optimal configuration. Given the rapid evolution of inference engines, we recommend referring to their official documentation for the latest updates to ensure peak performance.
+
+> Support for JoyAI-LLM Flash’s dense MTP architecture is currently being integrated into vLLM and SGLang. Until these PRs are merged into a stable release, please use the nightly Docker image for access to these features.
+
+## vLLM Deployment
+
+Here is the example to serve this model on a H200 single node with TP8 via vLLM:
+```bash
+vllm serve ${MODEL_PATH} --tp 8 --trust-remote-code \
+    --tool-call-parser qwen3_coder --enable-auto-tool-choice \
+    --speculative-config $'{"method": "mtp", "num_speculative_tokens": 3}'
+```
+**Key notes**
+- `--tool-call-parser qwen3_coder`: Required for enabling tool calling
+
+## SGLang Deployment
+
+Similarly, here is the example to run with TP8 on H200 in a single node via SGLang:
+```bash
+python3 -m sglang.launch_server --model-path ${MODEL_PATH} --tp-size 8 --trust-remote-code \
+  --tool-call-parser qwen3_coder \
+  --speculative-algorithm EAGLE --speculative-draft-model-path ${MTP_MODEL_PATH} \
+  --speculative-num-steps 2 --speculative-eagle-topk 2 --speculative-num-draft-tokens 3
+```
+**Key notes:**
+- `--tool-call-parser qwen3_coder`: Required when enabling tool usage.

model-1-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00342c45cd62e28fe183e2529ce61e2cecf0d8ea5451b2a8fe4137ae5e50e901
+size 140785016

model-10-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:25566dca71af5a8ad118ecff34fe03acda6d2483b9127cd26211be2082d5d6c8
+size 2479205264

model-11-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:106ca52367ca7d80b9f99553e3ad01b820ec12f28be1d2fc57233f2ce33a5199
+size 2479206048

model-12-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bc1a396967e8b7cfafe9399c031a2e869319b4bf40750a58f91b81037a36f0f2
+size 2479206048

model-13-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b813cd997285699eb10401e7b80c87773900b6c9ca48a305f228824dde553fe3
+size 2479206048

model-14-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b78f08465829cee2cf4d9a06912e33306c76a1251ac0f6637a2a505bef3376f6
+size 2479206048

model-15-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:415653c237c5189bcf725a587b7e7db5c15c858d9748c11fa9ee9a97c77ebf3e
+size 2479206048

model-16-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:edc2ec545b27884512149627d4c68c6e80a47429ab3f125e14f889aeb5c4555c
+size 2479206048

model-17-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:849209aeddc1e424dfd58c9b7bc85bb7fc26ecd79e35d0c3e0c1246ce7ea6cd8
+size 2479206048

model-18-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:37a1c9d67e32905eee90a59028f843b3e2df75b7fdf310347467525ba8af0778
+size 2479206048

model-19-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71940345d604e462c154b15eba80ce6013a9b47daaae628133c7a7b561ec0947
+size 2479206048

model-21-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36db259d412d9278343b91f5c0844c8a3c7992ceee11ebfe562494fb3934a012
+size 2479206048

model-22-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2a63e252bf2374910cf4c8a26d1c2ffe0ea88348dae23fed3cbe99c82443213b
+size 2479206048

model-23-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:490edf4128ec42e523730148c73c29445b8eea7a2d81d2e3296bc957b5d5dad7
+size 2479206048

model-24-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c82fda4cc8b9570898f355b1ca8f640262e269710d0efd914d387825b82e90f4
+size 2479206048

model-25-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:371774aa4bdfbefc9c62e4dd7a289e2cc15fe8ac8dd01f193bc3951c395149d1
+size 2479206048

model-27-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0061eca323f71053806bdd9d78e098fc28e8442b3cd27581407f56b2507e3a80
+size 2479206048

model-28-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e81ab12d6b7ab57a024daba11ccd75eaf4b3579b68aa7fc5e043a7c216df845a
+size 2479206048

model-29-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5fc269095eba7b9ffdbd51986cc5d805d73f32333b759a502f09ff55e797789a
+size 2479206048

model-3-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:734577bcc726ead7e0a9e8cad31edddb3d4ee2cc34fa5acecd37daea8332fc58
+size 2479205264

model-30-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:423416965094dc7d45422a5ca9d285cab7b12836bddd7a4c4f704ce996da3a00
+size 2479206048

model-31-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e405c1eebce9eaa75db83d8db6a108bd8d28b5e1e76903790dd20ae13297999e
+size 2479206048

model-32-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7f1b25c9b64915ba066ab35a007928d48ed3ae75fabbcc45a269b3638268c999
+size 2479206048

model-33-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f927f6e3246b676dc31fb4f227de92073ee54ea790042f7077d3d5ea5732d90f
+size 2479206048

model-34-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3582ea2db95a41584f8ceb30209fa8415e7fd254dc43be695f9b67dd28931ef2
+size 2479206048

model-35-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:feeffd5917defddd4543b8747bd33da9ee2cabfa3ff57634ebf1718e9ed46563
+size 2479206048

model-36-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:526404ba27ab66b8dcc455e1ca2234fab4843a890d7f6513e37eeebbfe2aaaf7
+size 2479206048

model-37-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4d8a43b4d41be94f6f691c254f79d8994611e9229fedd23dfcd4fcc39e0853de
+size 2479206048

model-38-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:51467503269ccea044ec96c7f94aff0efa523d586a8736dda70414fca29c5a03
+size 2479206048

model-39-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:deb76b58136b1144e79b0de587d991aed8c691fa4efcf1e54d118ec9b489c22d
+size 2479206048

model-4-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bb292fd660dfc54cf9afe89938d431bf58d8634576c8cc52bc874d4b68fadbd9
+size 2479205264

model-40-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9b09ac0b99a72696162f673ae20ad74b8ae5231fdf04f663d84c4ba981e83cf8
+size 2479206048

model-5-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f9c7bc3ad5832be99e39cdbfdc3f843d5c0d6edb9de0c0d8d34524d049b1d945
+size 2479205264

model-6-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eccfb66a683dd836293360c8c8441c466209708fd0220b04f7511814d75aed7c
+size 2479205264

model-8-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42c68787a8622f322c03190dda75801e30f65aa26997df147d9f7dec4fd264b4
+size 2479205264

model-9-of-40.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bf720fd443166d2e2a2f1a524035959febadfa079c7971c5a777642183008d6
+size 2479205264

model-non-layer.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd760d732c11c23778a0dbf2280b62431d77d1f4ebc4f01f111cf716786981f0
+size 1059066184

modeling_deepseek.py

@@ -0,0 +1,1028 @@
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           This file was automatically generated from src/transformers/models/deepseek_v3/modular_deepseek_v3.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_deepseek_v3.py file directly. One of our CI enforces this.
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+import math
+from functools import partial
+from typing import Callable, Optional, Tuple, 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, 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, dynamic_rope_update
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
+from transformers.processing_utils import Unpack
+from transformers.utils import (
+    LossKwargs,
+    add_start_docstrings,
+    add_start_docstrings_to_model_forward,
+    can_return_tuple,
+    is_torch_flex_attn_available,
+    logging,
+    replace_return_docstrings,
+)
+from transformers.utils.deprecation import deprecate_kwarg
+from .configuration_deepseek import DeepseekV3Config
+
+
+if is_torch_flex_attn_available():
+    from torch.nn.attention.flex_attention import BlockMask
+
+    from transformers.integrations.flex_attention import make_flex_block_causal_mask
+
+
+logger = logging.get_logger(__name__)
+_CONFIG_FOR_DOC = "DeepseekV3Config"
+
+
+class DeepseekV3RMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        DeepseekV3RMSNorm 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 DeepseekV3RotaryEmbedding(nn.Module):
+    def __init__(self, config: DeepseekV3Config, 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
+
+    @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)
+
+
+class DeepseekV3MLP(nn.Module):
+    def __init__(self, config, hidden_size=None, intermediate_size=None):
+        super().__init__()
+        self.config = config
+        self.hidden_size = config.hidden_size if hidden_size is None else hidden_size
+        self.intermediate_size = config.intermediate_size if intermediate_size is None else 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 DeepseekV3TopkRouter(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.top_k = config.num_experts_per_tok
+        self.n_routed_experts = config.n_routed_experts
+        self.routed_scaling_factor = config.routed_scaling_factor
+        self.n_group = config.n_group
+        self.topk_group = config.topk_group
+        self.norm_topk_prob = config.norm_topk_prob
+
+        self.weight = nn.Parameter(torch.empty((self.n_routed_experts, config.hidden_size)))
+        self.register_buffer("e_score_correction_bias", torch.zeros((self.n_routed_experts)))
+
+    @torch.no_grad()
+    def get_topk_indices(self, scores):
+        scores_for_choice = scores.view(-1, self.n_routed_experts) + self.e_score_correction_bias.unsqueeze(0)
+        group_scores = (
+            scores_for_choice.view(-1, self.n_group, self.n_routed_experts // self.n_group)
+            .topk(2, dim=-1)[0]
+            .sum(dim=-1)
+        )
+        group_idx = torch.topk(group_scores, k=self.topk_group, dim=-1, sorted=False)[1]
+        group_mask = torch.zeros_like(group_scores)
+        group_mask.scatter_(1, group_idx, 1)
+        score_mask = (
+            group_mask.unsqueeze(-1)
+            .expand(-1, self.n_group, self.n_routed_experts // self.n_group)
+            .reshape(-1, self.n_routed_experts)
+        )
+        scores_for_choice = scores_for_choice.masked_fill(~score_mask.bool(), 0.0)
+        topk_indices = torch.topk(scores_for_choice, k=self.top_k, dim=-1, sorted=False)[1]
+        return topk_indices
+
+    def forward(self, hidden_states):
+        hidden_states = hidden_states.view(-1, self.config.hidden_size)
+        router_logits = F.linear(hidden_states.type(torch.float32), self.weight.type(torch.float32))
+        scores = router_logits.sigmoid()
+        topk_indices = self.get_topk_indices(scores)
+        topk_weights = scores.gather(1, topk_indices)
+        if self.norm_topk_prob:
+            denominator = topk_weights.sum(dim=-1, keepdim=True) + 1e-20
+            topk_weights /= denominator
+        topk_weights = topk_weights * self.routed_scaling_factor
+        return topk_indices, topk_weights
+
+
+class DeepseekV3MoE(nn.Module):
+    """
+    A mixed expert module containing shared experts.
+    """
+
+    def __init__(self, config):
+        super().__init__()
+        self.config = config
+        self.experts = nn.ModuleList(
+            [
+                DeepseekV3MLP(config, intermediate_size=config.moe_intermediate_size)
+                for _ in range(config.n_routed_experts)
+            ]
+        )
+        self.gate = DeepseekV3TopkRouter(config)
+        self.shared_experts = DeepseekV3MLP(
+            config=config, intermediate_size=config.moe_intermediate_size * config.n_shared_experts
+        )
+
+    def moe(self, hidden_states: torch.Tensor, topk_indices: torch.Tensor, topk_weights: torch.Tensor):
+        r"""
+        CALL FOR CONTRIBUTION! I don't have time to optimise this right now, but expert weights need to be fused
+        to not have to do a loop here (deepseek has 256 experts soooo yeah).
+        """
+        final_hidden_states = torch.zeros_like(hidden_states, dtype=topk_weights.dtype)
+        expert_mask = torch.nn.functional.one_hot(topk_indices, num_classes=len(self.experts))
+        expert_mask = expert_mask.permute(2, 0, 1)
+
+        for expert_idx in range(len(self.experts)):
+            expert = self.experts[expert_idx]
+            mask = expert_mask[expert_idx]
+            token_indices, weight_indices = torch.where(mask)
+
+            if token_indices.numel() > 0:
+                expert_weights = topk_weights[token_indices, weight_indices]
+                expert_input = hidden_states[token_indices]
+                expert_output = expert(expert_input)
+                weighted_output = expert_output * expert_weights.unsqueeze(-1)
+                final_hidden_states.index_add_(0, token_indices, weighted_output)
+
+        # in original deepseek, the output of the experts are gathered once we leave this module
+        # thus the moe module is itelsf an IsolatedParallel module
+        # and all expert are "local" meaning we shard but we don't gather
+        return final_hidden_states.type(hidden_states.dtype)
+
+    def forward(self, hidden_states):
+        residuals = hidden_states
+        orig_shape = hidden_states.shape
+        topk_indices, topk_weights = self.gate(hidden_states)
+        hidden_states = hidden_states.view(-1, hidden_states.shape[-1])
+        hidden_states = self.moe(hidden_states, topk_indices, topk_weights).view(*orig_shape)
+        hidden_states = hidden_states + self.shared_experts(residuals)
+        return hidden_states
+
+
+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
+
+
+def apply_rotary_pos_emb_interleave(q, k, cos, sin, position_ids=None, unsqueeze_dim=1):
+    r"""
+    TODO let's just use the original freqcis computation to not have the view
+    transpose + reshape! This is not optimized!
+    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`):
+            The position indices of the tokens corresponding to the query and key tensors. For example, this can be
+            used to pass offsetted position ids when working with a KV-cache.
+        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)
+
+    b, h, s, d = q.shape
+    q = q.view(b, h, s, d // 2, 2).transpose(4, 3).reshape(b, h, s, d)
+
+    b, h, s, d = k.shape
+    k = k.view(b, h, s, d // 2, 2).transpose(4, 3).reshape(b, h, s, d)
+
+    q_embed = (q * cos) + (rotate_half(q) * sin)
+    k_embed = (k * cos) + (rotate_half(k) * sin)
+    return q_embed, k_embed
+
+
+def yarn_get_mscale(scale=1, mscale=1):
+    if scale <= 1:
+        return 1.0
+    return 0.1 * mscale * math.log(scale) + 1.0
+
+
+class DeepseekV3Attention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: DeepseekV3Config, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
+        self.attention_dropout = config.attention_dropout
+        self.num_heads = config.num_attention_heads
+        self.rope_theta = config.rope_theta
+        self.q_lora_rank = config.q_lora_rank
+        self.qk_rope_head_dim = config.qk_rope_head_dim
+        self.kv_lora_rank = config.kv_lora_rank
+        self.v_head_dim = config.v_head_dim
+        self.qk_nope_head_dim = config.qk_nope_head_dim
+        self.qk_head_dim = config.qk_head_dim
+
+        self.is_causal = True
+        self.q_a_proj = nn.Linear(config.hidden_size, config.q_lora_rank, bias=config.attention_bias)
+        self.q_a_layernorm = DeepseekV3RMSNorm(config.q_lora_rank)
+        self.q_b_proj = nn.Linear(config.q_lora_rank, self.num_heads * self.qk_head_dim, bias=False)
+
+        self.kv_a_proj_with_mqa = nn.Linear(
+            config.hidden_size,
+            self.kv_lora_rank + self.qk_rope_head_dim,
+            bias=config.attention_bias,
+        )
+        self.kv_a_layernorm = DeepseekV3RMSNorm(self.kv_lora_rank)
+        self.kv_b_proj = nn.Linear(
+            self.kv_lora_rank,
+            self.num_heads * (self.qk_nope_head_dim + self.v_head_dim),
+            bias=False,
+        )
+
+        self.o_proj = nn.Linear(
+            self.num_heads * self.v_head_dim,
+            config.hidden_size,
+            bias=config.attention_bias,
+        )
+
+        self.scaling = self.qk_head_dim ** (-0.5)
+        if self.config.rope_scaling is not None:
+            mscale_all_dim = self.config.rope_scaling.get("mscale_all_dim", 0)
+            scaling_factor = self.config.rope_scaling["factor"]
+            if mscale_all_dim:
+                mscale = yarn_get_mscale(scaling_factor, mscale_all_dim)
+                self.scaling = self.scaling * mscale * mscale
+
+    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]]]:
+        batch_size, seq_length = hidden_states.shape[:-1]
+        query_shape = (batch_size, seq_length, -1, self.qk_head_dim)
+        key_shape = (batch_size, seq_length, -1, self.qk_nope_head_dim + self.v_head_dim)
+
+        q_states = self.q_b_proj(self.q_a_layernorm(self.q_a_proj(hidden_states))).view(query_shape).transpose(1, 2)
+        q_pass, q_rot = torch.split(q_states, [self.qk_nope_head_dim, self.qk_rope_head_dim], dim=-1)
+
+        compressed_kv = self.kv_a_proj_with_mqa(hidden_states)
+        k_pass, k_rot = torch.split(compressed_kv, [self.kv_lora_rank, self.qk_rope_head_dim], dim=-1)
+
+        k_pass = self.kv_b_proj(self.kv_a_layernorm(k_pass)).view(key_shape).transpose(1, 2)
+        k_pass, value_states = torch.split(k_pass, [self.qk_nope_head_dim, self.v_head_dim], dim=-1)
+
+        k_rot = k_rot.view(batch_size, 1, seq_length, self.qk_rope_head_dim)
+
+        cos, sin = position_embeddings
+        if self.config.rope_interleave:  # support using interleaved weights for efficiency
+            q_rot, k_rot = apply_rotary_pos_emb_interleave(q_rot, k_rot, cos, sin)
+        else:
+            q_rot, k_rot = apply_rotary_pos_emb(q_rot, k_rot, cos, sin)
+        k_rot = k_rot.expand(*k_pass.shape[:-1], -1)
+
+        query_states = torch.cat((q_pass, q_rot), dim=-1)
+        key_states = torch.cat((k_pass, k_rot), dim=-1)
+
+        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)
+
+        if self.config._attn_implementation == "flash_attention_2" and self.qk_head_dim != self.v_head_dim:
+            value_states = F.pad(value_states, [0, self.qk_head_dim - self.v_head_dim])
+
+        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]
+
+        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,
+            **kwargs,
+        )
+
+        if self.config._attn_implementation == "flash_attention_2" and self.qk_head_dim != self.v_head_dim:
+            attn_output = attn_output[:, :, :, : self.v_head_dim]
+
+        attn_output = attn_output.reshape(batch_size, seq_length, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+        return attn_output, attn_weights
+
+
+class DeepseekV3DecoderLayer(nn.Module):
+    def __init__(self, config: DeepseekV3Config, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+
+        self.self_attn = DeepseekV3Attention(config=config, layer_idx=layer_idx)
+
+        if layer_idx >= config.first_k_dense_replace:
+            self.mlp = DeepseekV3MoE(config)
+        else:
+            self.mlp = DeepseekV3MLP(config)
+
+        self.input_layernorm = DeepseekV3RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = DeepseekV3RMSNorm(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[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,  # necessary, but kept here for BC
+        **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,
+            **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
+
+
+DEEPSEEK_V3_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 ([`DeepseekV3Config`]):
+            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 DeepseekV3 Model outputting raw hidden-states without any specific head on top.",
+    DEEPSEEK_V3_START_DOCSTRING,
+)
+class DeepseekV3PreTrainedModel(PreTrainedModel):
+    config_class = DeepseekV3Config
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["DeepseekV3DecoderLayer"]
+    _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_()
+        elif isinstance(module, DeepseekV3TopkRouter):
+            module.weight.data.normal_(mean=0.0, std=std)
+        elif isinstance(module, nn.Parameter):
+            module.weight.data.normal_(mean=0.0, std=std)
+
+
+DEEPSEEK_V3_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`, *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`.
+
+            It is a [`~cache_utils.Cache`] instance. For more details, see our [kv cache guide](https://huggingface.co/docs/transformers/en/kv_cache).
+
+            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 DeepseekV3 Model outputting raw hidden-states without any specific head on top.",
+    DEEPSEEK_V3_START_DOCSTRING,
+)
+class DeepseekV3Model(DeepseekV3PreTrainedModel):
+    """
+    Transformer decoder consisting of *config.num_hidden_layers* layers. Each layer is a [`DeepseekV3DecoderLayer`]
+
+    Args:
+        config: DeepseekV3Config
+    """
+
+    _keys_to_ignore_on_load_unexpected = [r"model\.layers\.61.*"]
+
+    def __init__(self, config: DeepseekV3Config):
+        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(
+            [DeepseekV3DecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.norm = DeepseekV3RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = DeepseekV3RotaryEmbedding(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
+
+    @can_return_tuple
+    @add_start_docstrings_to_model_forward(DEEPSEEK_V3_INPUTS_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[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **flash_attn_kwargs: Unpack[FlashAttentionKwargs],
+    ) -> 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
+
+        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
+
+        # TODO (joao): remove this exception in v4.56 -- it exists for users that try to pass a legacy cache
+        if not isinstance(past_key_values, (type(None), Cache)):
+            raise ValueError("The `past_key_values` should be either a `Cache` object or `None`.")
+
+        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(
+                    partial(decoder_layer.__call__, **flash_attn_kwargs),
+                    hidden_states,
+                    causal_mask,
+                    position_ids,
+                    past_key_values,
+                    output_attentions,
+                    use_cache,
+                    cache_position,
+                    position_embeddings,
+                )
+            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,
+                    **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,)
+
+        return 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,
+        )
+
+    def _update_causal_mask(
+        self,
+        attention_mask: torch.Tensor,
+        input_tensor: torch.Tensor,
+        cache_position: torch.Tensor,
+        past_key_values: Cache,
+        output_attentions: bool = False,
+    ):
+        if self.config._attn_implementation == "flash_attention_2":
+            if attention_mask is not None and (attention_mask == 0.0).any():
+                return attention_mask
+            return None
+        if self.config._attn_implementation == "flex_attention":
+            if isinstance(attention_mask, torch.Tensor):
+                attention_mask = make_flex_block_causal_mask(attention_mask)
+            if isinstance(attention_mask, BlockMask):
+                return attention_mask
+
+        # 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)
+
+        # 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 and not output_attentions:
+            if AttentionMaskConverter._ignore_causal_mask_sdpa(
+                attention_mask,
+                inputs_embeds=input_tensor,
+                past_key_values_length=past_seen_tokens,
+                is_training=self.training,
+            ):
+                return None
+
+        dtype, device = input_tensor.dtype, input_tensor.device
+        sequence_length = input_tensor.shape[1]
+        if using_static_cache:
+            target_length = past_key_values.get_max_cache_shape()
+        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],
+        )
+
+        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
+            min_dtype = torch.finfo(dtype).min
+            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,
+        **kwargs,
+    ):
+        """
+        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 place 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.
+        """
+        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
+            )
+            if sequence_length != 1:
+                causal_mask = torch.triu(causal_mask, diagonal=1)
+            causal_mask *= torch.arange(target_length, device=device) > cache_position.reshape(-1, 1)
+            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
+                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, LossKwargs): ...
+
+
+class DeepseekV3ForCausalLM(DeepseekV3PreTrainedModel, 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 = DeepseekV3Model(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
+
+    @can_return_tuple
+    @deprecate_kwarg("num_logits_to_keep", version="4.50", new_name="logits_to_keep")
+    @add_start_docstrings_to_model_forward(DEEPSEEK_V3_INPUTS_DOCSTRING)
+    @replace_return_docstrings(output_type=CausalLMOutputWithPast, config_class=_CONFIG_FOR_DOC)
+    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[Cache] = 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,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        **kwargs: Unpack[KwargsForCausalLM],
+    ) -> CausalLMOutputWithPast:
+        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]`.
+
+            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, DeepseekV3ForCausalLM
+
+        >>> model = DeepseekV3ForCausalLM.from_pretrained("meta-deepseek_v3/DeepseekV3-2-7b-hf")
+        >>> tokenizer = AutoTokenizer.from_pretrained("meta-deepseek_v3/DeepseekV3-2-7b-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
+        )
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs: BaseModelOutputWithPast = 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,
+            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=logits, labels=labels, vocab_size=self.config.vocab_size, **kwargs)
+
+        return CausalLMOutputWithPast(
+            loss=loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+        )
+
+
+__all__ = ["DeepseekV3PreTrainedModel", "DeepseekV3Model", "DeepseekV3ForCausalLM"]

mtp-1-of-1.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e708907b5c5a584e0d81ebd2858ecf9f0f22798616a61fc273f0d39eac9512c0
+size 687105960