training-free expert prune K=24/32 (PR=25%) via routing-mass calibration

LICENSE-MODEL

@@ -0,0 +1,56 @@
+MINIMAX MODEL LICENSE
+MiniMax-M2.5 Version Release Date: 2026-02-13
+
+1. Definitions
+
+"Agreement" means the terms and conditions for use, reproduction, distribution and modification of the Model Materials set forth herein.
+
+"Licensee" or "you" means you, or your employer or any other person or entity (if you are entering into this Agreement on such person or entity's behalf), of the age required under applicable laws, rules or regulations to provide legal consent and that has legal authority to bind your employer or such other person or entity if you are entering in this Agreement on their behalf.
+
+"Model" means the foundational large language models and software and algorithms, including machine-learning model code, trained model weights, inference-enabling code, training-enabling code, fine-tuning enabling code and other elements of the foregoing distributed by MiniMax.
+
+"Model Materials" means, collectively, the Model and any source code, scripts, specifications, manuals and documentation accompanying the Model (and any portion thereof) made available under this Agreement.
+
+"MiniMax" or "we" means MiniMax AI.
+
+2. License Rights and Redistribution
+
+a. Grant of Rights. You are granted a non-exclusive, worldwide and royalty-free limited license under MiniMax's intellectual property or other rights owned by MiniMax embodied in the Model Materials to use, reproduce, distribute, copy, create derivative works of, and make modifications to the Model Materials.
+
+b. Redistribution and Use.
+
+   i. If you distribute or make available the Model Materials (or any derivative works thereof), or a product or service that uses any of them, including another AI model, you shall provide a copy of this Agreement with any such the Model Materials or derivative works and cause any modified files to carry prominent notices stating that you changed the files. You may add your own copyright statement to your modifications and may provide additional or different license terms and conditions for use, reproduction, or distribution of your modifications, or for any such derivative works as a whole, provided your use, reproduction, and distribution of the work otherwise complies with the terms and conditions of this Agreement.
+
+   ii. You must retain in all copies of the Model Materials that you distribute the following attribution notice within a "Notice" text file distributed as a part of such copies: "MiniMax AI model is licensed under the MiniMax Model License, Copyright © MiniMax. All Rights Reserved."
+
+   iii. Your use of the Model Materials must comply with applicable laws and regulations (including trade compliance laws and regulations) and adhere to the Prohibited Uses Policy for the Model Materials, which is hereby incorporated by reference into this Agreement.
+
+3. Disclaimer of Warranty. UNLESS REQUIRED BY APPLICABLE LAW, THE MODEL MATERIALS AND ANY OUTPUT AND RESULTS THEREFROM ARE PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, AND MINIMAX DISCLAIMS ALL WARRANTIES OF ANY KIND, BOTH EXPRESS AND IMPLIED, INCLUDING, WITHOUT LIMITATION, ANY WARRANTIES OF TITLE, NON-INFRINGEMENT, MERCHANTABILITY, OR FITNESS FOR A PARTICULAR PURPOSE. YOU ARE SOLELY RESPONSIBLE FOR DETERMINING THE APPROPRIATENESS OF USING OR REDISTRIBUTING THE MODEL MATERIALS AND ASSUME ANY RISKS ASSOCIATED WITH YOUR USE OF THE MODEL MATERIALS AND ANY OUTPUT AND RESULTS.
+
+4. Limitation of Liability. IN NO EVENT WILL MINIMAX OR ITS AFFILIATES BE LIABLE UNDER ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, TORT, NEGLIGENCE, PRODUCTS LIABILITY, OR OTHERWISE, ARISING OUT OF THIS AGREEMENT, FOR ANY LOST PROFITS OR ANY INDIRECT, SPECIAL, CONSEQUENTIAL, INCIDENTAL, EXEMPLARY OR PUNITIVE DAMAGES, EVEN IF MINIMAX OR ITS AFFILIATES HAVE BEEN ADVISED OF THE POSSIBILITY OF ANY OF THE FOREGOING.
+
+5. Intellectual Property.
+
+a. No trademark license is granted to use the trade names, trademarks, service marks, or product names of MiniMax, except as required to fulfill notice requirements under this Agreement.
+
+b. Subject to MiniMax's ownership of the Model Materials and derivatives made by or for MiniMax, with respect to any derivative works and modifications of the Model Materials that are made by you, as between you and MiniMax, you are and will be the owner of such derivative works and modifications.
+
+c. If you institute litigation or other proceedings against MiniMax or any entity (including a cross-claim or counterclaim in a lawsuit) alleging that the Model Materials or outputs or results, or any portion of any of the foregoing, constitutes infringement of intellectual property or other rights owned or licensable by you, then any licenses granted to you under this Agreement shall terminate as of the date such litigation or claim is filed or instituted. You will indemnify and hold harmless MiniMax from and against any claim by any third party arising out of or related to your use or distribution of the Model Materials.
+
+6. Term and Termination. The term of this Agreement will commence upon your acceptance of this Agreement or access to the Model Materials and will continue in full force and effect until terminated in accordance with the terms and conditions herein. MiniMax may terminate this Agreement if you are in breach of any term or condition of this Agreement. Upon termination of this Agreement, you shall delete and cease use of the Model Materials. Sections 3, 4 and 7 shall survive the termination of this Agreement.
+
+7. Governing Law and Jurisdiction. This Agreement will be governed and construed under the laws of Singapore without regard to choice of law principles, and the UN Convention on Contracts for the International Sale of Goods does not apply to this Agreement. Any dispute arising out of or in connection with this Agreement, including any question regarding its existence, validity or termination, shall be referred to and finally resolved by arbitration administered by the Singapore International Arbitration Centre ("SIAC") in accordance with the Arbitration Rules of the Singapore International Arbitration Centre ("SIAC Rules") for the time being in force, which rules are deemed to be incorporated by reference in this clause.
+
+Prohibited Uses Policy
+
+You agree you will not use, or allow others to use, the Models or any derivatives of the Models to:
+
+1. Violate any applicable federal, state, local, or international law or regulation, or infringe upon the lawful rights or interests of any third party.
+2. Assist with, engage in or otherwise support any military purpose.
+3. Exploit, harm, or attempt to exploit or harm minors in any way.
+4. Generate or disseminate false or misleading information with the intent to cause harm.
+5. Generate or disseminate content prohibited by applicable laws or regulations.
+6. Generate or disseminate personally identifiable information without proper authorization or for unlawful or unreasonable purposes.
+7. Defame, disparage, harass, or cause harm to any individual or entity.
+8. Conduct fully automated decision-making that adversely affects an individual's legal rights or creates or modifies a binding, enforceable obligation.
+9. Promote discrimination, hate speech, or harmful behavior against individuals or groups based on race or ethnic origin, religion, disability, age, nationality and national origin, veteran status, sexual orientation, gender or gender identity, caste, immigration status, or any other characteristic that is associated with systemic discrimination or marginalization.

README.md

@@ -0,0 +1,87 @@
+---
+license: other
+license_name: model-license
+library_name: transformers
+pipeline_tag: text-generation
+tags:
+  - minimax
+  - mixture-of-experts
+  - moe
+  - pruning
+  - expert-pruning
+  - fp8
+base_model: morriszjm/MiniMax-M2.5-tiny
+---
+
+# MiniMax-M2.5-tiny-24e
+
+Training-free expert-pruned variant of [`morriszjm/MiniMax-M2.5-tiny`](https://huggingface.co/morriszjm/MiniMax-M2.5-tiny),
+produced by the [`minimax_expert_pruning`](https://github.com/-) pipeline.
+
+## What changed
+
+- `num_local_experts`: **32 → 24** (pruning rate: **25.0 %**)
+- All non-MoE tensors (attention, layernorms, embeddings, lm_head, MTP heads if any)
+  are **bit-identical** to the source model.
+- `gate.weight` and `e_score_correction_bias` per MoE layer are row-sliced to the
+  kept experts; per-expert tensors of dropped experts are absent; kept experts
+  are renumbered contiguously to `0..23`.
+- `top_k = num_experts_per_tok` is unchanged (8).
+
+## Method (one-paragraph)
+
+We run a small calibration set (64 prompts spanning Nokia AI4Code,
+general English Q&A, multilingual, and reasoning) through the unpruned source
+model and hook every MoE layer's router. Per layer, we accumulate each
+expert's **selected probability mass** — the post-sigmoid routing weight that
+the expert receives, summed over all calibration tokens that selected it
+in their top-8. We keep the top-K by this score per layer (uniform K)
+and atomically slice the on-disk per-expert tensors. No gradients, no
+fine-tuning.
+
+## Layer-level statistics
+
+- Layers covered: **8**
+- Tokens per layer (calibration): **1,851**
+- Calibration prompts by bucket: `{"ai4code": 1008, "general_en": 416, "reasoning": 257, "multilingual": 170}`
+- Median per-layer "kept-min vs drop-max" routing-mass gap: **+0.7197**
+  (positive = clean separation between the kept and dropped experts;
+   close to zero or negative = experts of similar utility, expect more
+   quality risk)
+
+## Intended use
+
+Production-style serving of the source model's domain (Nokia / Merlin AI4Code
+plus general English) at reduced HBM footprint. Expect graceful quality
+degradation versus the unpruned source on tasks well-covered by the
+calibration mix; quality on out-of-distribution domains may drop further.
+
+## Limitations
+
+- Training-free: no fine-tune recovery, no distillation, no merge.
+- Uniform K per layer: late layers may tolerate more pruning than early ones,
+  unexploited here.
+- Calibration mix is small (64 prompts). Domain coverage is biased
+  toward the included buckets.
+
+## Files
+
+`config.json`, `model-NNNNN-of-NNNNN.safetensors` (FP8), `model.safetensors.index.json`,
+tokenizer, custom `modeling_minimax_m2.py` + `configuration_minimax_m2.py`, and
+`expert_prune_plan.json` (full record of which experts were kept per layer).
+
+## Loading
+
+```python
+from transformers import AutoModelForCausalLM, AutoTokenizer
+import torch
+
+tok = AutoTokenizer.from_pretrained("morriszjm/MiniMax-M2.5-tiny-24e", trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(
+    "morriszjm/MiniMax-M2.5-tiny-24e", trust_remote_code=True,
+    torch_dtype=torch.bfloat16, device_map="auto",
+)
+```
+
+For vLLM serving, pass `--trust-remote-code` and (on multi-GPU) match
+`--data-parallel-size` to the EP topology you compiled the K against.

chat_template.jinja

@@ -0,0 +1,159 @@
+{# ----------‑‑‑ special token variables ‑‑‑---------- #}
+{%- set toolcall_begin_token   = '<minimax:tool_call>'         -%}
+{%- set toolcall_end_token     = '</minimax:tool_call>'        -%}
+{#- Tool Rendering Functions ============================================== -#}
+{%- macro render_tool_namespace(namespace_name, tool_list) -%}
+{%- for tool in tool_list -%}
+<tool>{{ tool.function | tojson(ensure_ascii=False) }}</tool>
+{% endfor -%}
+{%- endmacro -%}
+{%- macro visible_text(content) -%}
+    {%- if content is string -%}
+        {{ content }}
+    {%- elif content is iterable and content is not mapping -%}
+        {%- for item in content -%}
+            {%- if item is mapping and item.type == 'text' -%}
+                {{- item.text }}
+            {%- elif item is string -%}
+                {{- item }}
+            {%- endif -%}
+        {%- endfor -%}
+    {%- else -%}
+        {{- content }}
+    {%- endif -%}
+{%- endmacro -%}
+{#- System Message Construction ============================================ -#}
+{%- macro build_system_message(system_message) -%}
+    {%- if system_message and system_message.content -%}
+        {{- visible_text(system_message.content) }}
+    {%- else -%}
+        {%- if model_identity is not defined -%}
+            {%- set model_identity = "You are a helpful assistant. Your name is MiniMax-M2.5 and is built by MiniMax." -%}
+        {%- endif -%}
+        {{- model_identity }}
+    {%- endif -%}
+    
+    {#- Handle current_date -#}
+    {%- if system_message and system_message.current_date -%}
+        {{- '\n' ~ 'Current date: ' + system_message.current_date }}
+    {%- endif -%}
+    {#- Handle current_location -#}
+    {%- if system_message and system_message.current_location -%}
+        {{- '\n' ~ 'Current location: ' + system_message.current_location }}
+    {%- endif -%}
+{%- endmacro -%}
+{#- Main Template Logic ================================================= -#}
+{#- Extract system message (only first message if it's system) -#}
+{%- set system_message = none -%}
+{%- set conversation_messages = messages -%}
+{%- if messages and messages[0].role == "system" -%}
+    {%- set system_message = messages[0] -%}
+    {%- set conversation_messages = messages[1:] -%}
+{%- endif -%}
+{#- Get the last user message turn, for interleved thinking -#}
+{%- set ns = namespace(last_user_index=-1) %}
+{% for m in conversation_messages %}
+    {%- if m.role == 'user' %}
+        {% set ns.last_user_index = loop.index0 -%}
+    {%- endif %}
+{%- endfor %}
+{#- Render system message -#}
+{{- ']~!b[' ~ ']~b]system' ~ '\n' }}
+{{- build_system_message(system_message) }}
+{#- Render tools if available -#}
+{%- if tools -%}
+    {{- '\n\n' ~ '# Tools' ~ '\n' ~ 'You may call one or more tools to assist with the user query.\nHere are the tools available in JSONSchema format:' ~ '\n' }}
+    {{- '\n' ~ '<tools>' ~ '\n' }}
+    {{- render_tool_namespace("functions", tools) }}
+    {{- '</tools>' ~ '\n\n' }}
+{{- 'When making tool calls, use XML format to invoke tools and pass parameters:' ~ '\n' }}
+{{- '\n' ~ toolcall_begin_token }}
+<invoke name="tool-name-1">
+<parameter name="param-key-1">param-value-1</parameter>
+<parameter name="param-key-2">param-value-2</parameter>
+...
+</invoke>
+{{- '\n' ~ toolcall_end_token }}
+{%- endif -%}
+{{- '[e~[\n' }}
+
+{#- Render messages -#}
+{%- set last_tool_call = namespace(name=none) -%}
+{%- for message in conversation_messages -%}
+    {%- if message.role == 'assistant' -%}
+        {#- Only render reasoning_content if no user message follows -#}
+        {{- ']~b]ai' ~ '\n' }}
+
+        {%- set reasoning_content = '' %}
+        {%- set content = visible_text(message.content) %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].strip('\n').split('<think>')[-1].strip('\n') %}
+                {%- set content = content.split('</think>')[-1].strip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- if reasoning_content and loop.index0 > ns.last_user_index -%}
+            {{- '<think>' ~ '\n' ~ reasoning_content ~ '\n' ~ '</think>' ~ '\n\n' }}
+        {%- endif -%}
+        {%- if content -%}
+            {{- content }}
+        {%- endif -%}
+        {%- if message.tool_calls -%}
+            {{- '\n' ~ toolcall_begin_token ~ '\n' }}
+
+            {%- for tool_call in message.tool_calls -%}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<invoke name="' + tool_call.name + '">' }}
+                {% set _args = tool_call.arguments %}
+                {%- for k, v in _args.items() %}
+                {{- '<parameter name="' + k + '">' }}
+                {{- v | tojson(ensure_ascii=False) if v is not string else v }}
+                {{- '</parameter>' }}
+                {% endfor %}
+                {{- '</invoke>' ~ '\n' }}
+            {%- endfor -%}
+            
+            {{- toolcall_end_token}}
+            {%- set last_tool_call.name = message.tool_calls[-1].name -%}
+        {%- else -%}
+            {%- set last_tool_call.name = none -%}
+        {%- endif -%}
+        {{- '[e~[' ~ '\n' }}
+        
+    {%- elif message.role == 'tool' -%}
+    {%- if last_tool_call.name is none -%}
+        {{- raise_exception("Message has tool role, but there was no previous assistant message with a tool call!") }}
+    {%- endif -%}
+    {%- if loop.first or (conversation_messages[loop.index0 - 1].role != 'tool') -%}
+        {{- ']~b]tool' }}
+    {%- endif -%}
+    {%- if message.content is string -%}
+        {{- '\n<response>' }}
+        {{- message.content }}
+        {{- '</response>' }}
+    {%- else -%}
+        {%- for tr in message.content -%}
+            {{- '\n<response>' }}
+            {{- tr.output if tr.output is defined else (tr.text if tr.type == 'text' and tr.text is defined else tr) }}
+            {{- '\n</response>' }}
+        {%- endfor -%}
+    {%- endif -%}
+    {%- if loop.last or (conversation_messages[loop.index0 + 1].role != 'tool') -%}
+        {{- '[e~[\n' -}}
+    {%- endif -%}
+        
+    {%- elif message.role == 'user' -%}
+        {{- ']~b]user' ~ '\n' }}
+        {{- visible_text(message.content) }}
+        {{- '[e~[' ~ '\n' }}
+    {%- endif -%}
+{%- endfor -%}
+
+{#- Generation prompt -#}
+{%- if add_generation_prompt -%}
+{{- ']~b]ai' ~ '\n' ~ '<think>' ~ '\n' }}
+{%- endif -%}

config.json

@@ -0,0 +1,61 @@
+{
+  "_pruned_from": "morriszjm/MiniMax-M2.5-tiny",
+  "_pruned_strategy": "expert_prune.routing_calibration.v1",
+  "architectures": [
+    "MiniMaxM2ForCausalLM"
+  ],
+  "attn_type_list": [
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_minimax_m2.MiniMaxM2Config",
+    "AutoModelForCausalLM": "modeling_minimax_m2.MiniMaxM2ForCausalLM"
+  },
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 3072,
+  "intermediate_size": 1536,
+  "max_position_embeddings": 196608,
+  "model_type": "minimax_m2",
+  "mtp_transformer_layers": 1,
+  "num_attention_heads": 48,
+  "num_experts_per_tok": 8,
+  "num_hidden_layers": 8,
+  "num_key_value_heads": 8,
+  "num_local_experts": 24,
+  "num_mtp_modules": 3,
+  "qk_norm_type": "per_layer",
+  "quantization_config": {
+    "activation_scheme": "dynamic",
+    "fmt": "float8_e4m3fn",
+    "modules_to_not_convert": [
+      "gate",
+      "e_score_correction_bias",
+      "lm_head"
+    ],
+    "quant_method": "fp8",
+    "weight_block_size": [
+      128,
+      128
+    ]
+  },
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 5000000,
+  "rotary_dim": 64,
+  "scoring_func": "sigmoid",
+  "shared_intermediate_size": 0,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.46.1",
+  "use_cache": true,
+  "use_mtp": true,
+  "use_qk_norm": true,
+  "use_routing_bias": true,
+  "vocab_size": 200064
+}

configuration_minimax_m2.py

@@ -0,0 +1,200 @@
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           This file was automatically generated from src/transformers/models/minimax_m2/modular_minimax_m2.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_minimax_m2.py file directly. One of our CI enforces this.
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+# coding=utf-8
+# Copyright 2025 the HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class MiniMaxM2Config(PretrainedConfig):
+    r"""
+    This is the configuration class to store the configuration of a [`MiniMaxM2Model`]. It is used to instantiate an
+    MiniMaxM2 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 MiniMaxM2-7B-v0.1 or MiniMaxM2-7B-Instruct-v0.1.
+
+    [minimax_m2ai/MiniMaxM2-8x7B](https://huggingface.co/minimax_m2ai/MiniMaxM2-8x7B)
+    [minimax_m2ai/MiniMaxM2-7B-Instruct-v0.1](https://huggingface.co/minimax_m2ai/MiniMaxM2-7B-Instruct-v0.1)
+
+    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 32000):
+            Vocabulary size of the MiniMaxM2 model. Defines the number of different tokens that can be represented by the
+            `inputs_ids` passed when calling [`MiniMaxM2Model`]
+        hidden_size (`int`, *optional*, defaults to 4096):
+            Dimension of the hidden representations.
+        intermediate_size (`int`, *optional*, defaults to 14336):
+            Dimension of the MLP representations.
+        num_hidden_layers (`int`, *optional*, defaults to 32):
+            Number of hidden layers in the Transformer encoder.
+        num_attention_heads (`int`, *optional*, defaults to 32):
+            Number of attention heads for each attention layer in the Transformer encoder.
+        num_key_value_heads (`int`, *optional*, defaults to 8):
+            This is the number of key_value heads that should be used to implement Grouped Query Attention. If
+            `num_key_value_heads=num_attention_heads`, the model will use Multi Head Attention (MHA), if
+            `num_key_value_heads=1` the model will use Multi Query Attention (MQA) otherwise GQA is used. When
+            converting a multi-head checkpoint to a GQA checkpoint, each group key and value head should be constructed
+            by meanpooling all the original heads within that group. For more details, check out [this
+            paper](https://huggingface.co/papers/2305.13245). If it is not specified, will default to `8`.
+        head_dim (`int`, *optional*, defaults to `hidden_size // num_attention_heads`):
+            The attention head dimension.
+        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*32`):
+            The maximum sequence length that this model might ever be used with. MiniMaxM2's sliding window attention
+            allows sequence of up to 4096*32 tokens.
+        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-05):
+            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*):
+            The id of the padding token.
+        bos_token_id (`int`, *optional*, defaults to 1):
+            The id of the "beginning-of-sequence" token.
+        eos_token_id (`int`, *optional*, defaults to 2):
+            The id of the "end-of-sequence" token.
+        tie_word_embeddings (`bool`, *optional*, defaults to `False`):
+            Whether the model's input and output word embeddings should be tied.
+        rope_theta (`float`, *optional*, defaults to 1000000.0):
+            The base period of the RoPE embeddings.
+        sliding_window (`int`, *optional*):
+            Sliding window attention window size. If not specified, will default to `4096`.
+        attention_dropout (`float`, *optional*, defaults to 0.0):
+            The dropout ratio for the attention probabilities.
+        num_experts_per_tok (`int`, *optional*, defaults to 2):
+            The number of experts to route per-token, can be also interpreted as the `top-k` routing
+            parameter
+        num_local_experts (`int`, *optional*, defaults to 8):
+            Number of experts per Sparse MLP layer.
+        output_router_logits (`bool`, *optional*, defaults to `False`):
+            Whether or not the router logits should be returned by the model. Enabling this will also
+            allow the model to output the auxiliary loss. See [here]() for more details
+        router_aux_loss_coef (`float`, *optional*, defaults to 0.001):
+            The aux loss factor for the total loss.
+        router_jitter_noise (`float`, *optional*, defaults to 0.0):
+            Amount of noise to add to the router.
+
+    ```python
+    >>> from transformers import MiniMaxM2Model, MiniMaxM2Config
+
+    >>> # Initializing a MiniMaxM2 7B style configuration
+    >>> configuration = MiniMaxM2Config()
+
+    >>> # Initializing a model from the MiniMaxM2 7B style configuration
+    >>> model = MiniMaxM2Model(configuration)
+
+    >>> # Accessing the model configuration
+    >>> configuration = model.config
+    ```"""
+
+    model_type = "minimax_m2"
+    keys_to_ignore_at_inference = ["past_key_values"]
+    base_model_tp_plan = {
+        "layers.*.self_attn.q_proj": "colwise",
+        "layers.*.self_attn.k_proj": "colwise",
+        "layers.*.self_attn.v_proj": "colwise",
+        "layers.*.self_attn.o_proj": "rowwise",
+        "layers.*.block_sparse_moe.gate": "colwise_rep",  # we need to replicate here to correctly route experts
+        "layers.*.block_sparse_moe.experts.*.w1": "colwise",
+        "layers.*.block_sparse_moe.experts.*.w2": "rowwise",
+        "layers.*.block_sparse_moe.experts.*.w3": "colwise",
+    }
+    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=32000,
+        hidden_size=4096,
+        intermediate_size=14336,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=8,
+        head_dim=None,
+        hidden_act="silu",
+        max_position_embeddings=4096 * 32,
+        initializer_range=0.02,
+        rms_norm_eps=1e-5,
+        use_cache=True,
+        pad_token_id=None,
+        bos_token_id=1,
+        eos_token_id=2,
+        tie_word_embeddings=False,
+        rope_theta=1e6,
+        sliding_window=None,
+        attention_dropout=0.0,
+        num_experts_per_tok=2,
+        num_local_experts=8,
+        output_router_logits=False,
+        router_aux_loss_coef=0.001,
+        router_jitter_noise=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.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+        self.sliding_window = sliding_window
+
+        # 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.use_cache = use_cache
+        self.rope_theta = rope_theta
+        self.attention_dropout = attention_dropout
+        self.head_dim = head_dim
+
+        self.num_experts_per_tok = num_experts_per_tok
+        self.num_local_experts = num_local_experts
+        self.output_router_logits = output_router_logits
+        self.router_aux_loss_coef = router_aux_loss_coef
+        self.router_jitter_noise = router_jitter_noise
+
+        self.use_qk_norm = kwargs.pop("use_qk_norm", False)
+        self.rotary_dim = kwargs.pop("rotary_dim", self.head_dim)
+        self.partial_rotary_factor = kwargs.pop("partial_rotary_factor", 1)
+        if self.head_dim is not None:
+            self.partial_rotary_factor = self.rotary_dim / self.head_dim
+
+        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__ = ["MiniMaxM2Config"]

expert_prune_plan.json

@@ -0,0 +1,263 @@
+{
+  "src_model": "morriszjm/MiniMax-M2.5-tiny",
+  "src_inventory": "artifacts/tiny_inventory.json",
+  "src_importance": "artifacts/tiny_importance.json",
+  "method": "routing_softmax_topk_mass + per-layer uniform top-K, drop-the-rest",
+  "K": 24,
+  "L": 8,
+  "E_orig": 32,
+  "n_dropped": 8,
+  "pruning_rate": 0.25,
+  "ep_size_check": 8,
+  "top_k": 8,
+  "kept_per_layer_old_indices": [
+    [
+      0,
+      1,
+      2,
+      4,
+      5,
+      6,
+      7,
+      10,
+      12,
+      13,
+      14,
+      15,
+      17,
+      18,
+      19,
+      20,
+      21,
+      22,
+      23,
+      25,
+      26,
+      27,
+      30,
+      31
+    ],
+    [
+      0,
+      1,
+      2,
+      3,
+      4,
+      6,
+      7,
+      8,
+      9,
+      11,
+      12,
+      13,
+      14,
+      16,
+      17,
+      18,
+      19,
+      23,
+      24,
+      27,
+      28,
+      29,
+      30,
+      31
+    ],
+    [
+      0,
+      1,
+      2,
+      3,
+      4,
+      6,
+      7,
+      8,
+      9,
+      10,
+      12,
+      14,
+      16,
+      17,
+      19,
+      20,
+      22,
+      23,
+      24,
+      26,
+      27,
+      28,
+      29,
+      30
+    ],
+    [
+      0,
+      2,
+      3,
+      5,
+      6,
+      7,
+      8,
+      9,
+      11,
+      15,
+      17,
+      18,
+      19,
+      20,
+      21,
+      22,
+      23,
+      24,
+      25,
+      26,
+      27,
+      28,
+      29,
+      30
+    ],
+    [
+      0,
+      1,
+      2,
+      3,
+      4,
+      5,
+      7,
+      8,
+      9,
+      10,
+      12,
+      14,
+      16,
+      17,
+      18,
+      21,
+      22,
+      23,
+      24,
+      25,
+      26,
+      27,
+      28,
+      30
+    ],
+    [
+      1,
+      2,
+      3,
+      7,
+      8,
+      10,
+      11,
+      12,
+      14,
+      15,
+      16,
+      17,
+      18,
+      19,
+      20,
+      21,
+      22,
+      23,
+      24,
+      26,
+      28,
+      29,
+      30,
+      31
+    ],
+    [
+      0,
+      1,
+      2,
+      4,
+      5,
+      6,
+      7,
+      8,
+      9,
+      10,
+      12,
+      16,
+      17,
+      19,
+      21,
+      22,
+      23,
+      25,
+      26,
+      27,
+      28,
+      29,
+      30,
+      31
+    ],
+    [
+      0,
+      1,
+      3,
+      6,
+      7,
+      9,
+      10,
+      11,
+      12,
+      13,
+      14,
+      15,
+      16,
+      17,
+      19,
+      22,
+      24,
+      25,
+      26,
+      27,
+      28,
+      29,
+      30,
+      31
+    ]
+  ],
+  "kept_score_min_per_layer": [
+    5.091672860085964,
+    14.596360385417938,
+    33.47318064421415,
+    8.822094783186913,
+    12.742229215800762,
+    19.971908435225487,
+    0.0,
+    5.011361248791218
+  ],
+  "drop_score_max_per_layer": [
+    4.3719402849674225,
+    14.47503225505352,
+    30.457208454608917,
+    8.272616788744926,
+    11.775540083646774,
+    15.646341614425182,
+    0.0,
+    4.902323044836521
+  ],
+  "kept_dropped_score_gap_per_layer": [
+    0.7197325751185417,
+    0.12132813036441803,
+    3.015972189605236,
+    0.5494779944419861,
+    0.9666891321539879,
+    4.325566820800304,
+    0.0,
+    0.10903820395469666
+  ],
+  "calibration_meta": {
+    "n_prompts_used": 64,
+    "tokens_per_layer": 1851,
+    "tokens_per_bucket": {
+      "ai4code": 1008,
+      "general_en": 416,
+      "reasoning": 257,
+      "multilingual": 170
+    }
+  }
+}

generation_config.json

@@ -0,0 +1,9 @@
+{
+  "bos_token_id": 200019,
+  "do_sample": true,
+  "eos_token_id": 200020,
+  "temperature": 1.0,
+  "top_p": 0.95,
+  "top_k": 40,
+  "transformers_version": "4.46.1"
+}

model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0ee1cfa73ace1680fb80cb621a0b9b40c2f46655676c6924555e853831b9b86
+size 5365009616

model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8ceab04cabeab3612960f483fd18a07ea1e571f751aaed1143f709b7d9aca57
+size 167100136

modeling_minimax_m2.py

@@ -0,0 +1,706 @@
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           This file was automatically generated from src/transformers/models/minimax_m2/modular_minimax_m2.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_minimax_m2.py file directly. One of our CI enforces this.
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+# coding=utf-8
+# Copyright 2025 the HuggingFace Team. All rights reserved.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+
+from collections.abc import Callable
+from typing import Optional, Union, Unpack
+
+import torch
+from torch import nn
+
+from transformers.activations import ACT2FN
+from transformers.cache_utils import Cache, DynamicCache
+from transformers.generation import GenerationMixin
+from transformers.integrations import use_kernel_forward_from_hub
+from transformers.masking_utils import create_causal_mask, create_sliding_window_causal_mask
+from transformers.modeling_flash_attention_utils import FlashAttentionKwargs
+from transformers.modeling_layers import (
+    GenericForQuestionAnswering,
+    GenericForSequenceClassification,
+    GenericForTokenClassification,
+    GradientCheckpointingLayer,
+)
+from transformers.modeling_outputs import MoeCausalLMOutputWithPast, MoeModelOutputWithPast
+from transformers.modeling_rope_utils import ROPE_INIT_FUNCTIONS, dynamic_rope_update
+from transformers.modeling_utils import ALL_ATTENTION_FUNCTIONS, PreTrainedModel
+from transformers.utils import TransformersKwargs, auto_docstring, can_return_tuple
+from transformers.utils.deprecation import deprecate_kwarg
+from transformers.utils.generic import OutputRecorder, check_model_inputs
+from .configuration_minimax_m2 import MiniMaxM2Config
+
+
+class MiniMaxM2MLP(nn.Module):
+    def __init__(self, config: MiniMaxM2Config):
+        super().__init__()
+        self.ffn_dim = config.intermediate_size
+        self.hidden_dim = config.hidden_size
+
+        self.w1 = nn.Linear(self.hidden_dim, self.ffn_dim, bias=False)
+        self.w2 = nn.Linear(self.ffn_dim, self.hidden_dim, bias=False)
+        self.w3 = nn.Linear(self.hidden_dim, self.ffn_dim, bias=False)
+
+        self.act_fn = ACT2FN[config.hidden_act]
+
+    def forward(self, hidden_states):
+        current_hidden_states = self.act_fn(self.w1(hidden_states)) * self.w3(hidden_states)
+        current_hidden_states = self.w2(current_hidden_states)
+        return current_hidden_states
+
+
+class MiniMaxM2Experts(nn.ModuleList):
+    """
+    ModuleList of experts.
+    """
+
+    def __init__(self, config: MiniMaxM2Config):
+        super().__init__()
+        self.top_k = config.num_experts_per_tok
+        self.num_experts = config.num_local_experts
+        for _ in range(self.num_experts):
+            self.append(MiniMaxM2MLP(config))
+
+    def forward(
+        self, hidden_states: torch.Tensor, top_k_index: torch.Tensor, top_k_weights: torch.Tensor
+    ) -> torch.Tensor:
+        """
+        Args:
+            hidden_states: (batch_size * sequence_length, hidden_dim)
+            selected_experts: (batch_size * sequence_length, top_k)
+            routing_weights: (batch_size * sequence_length, top_k)
+        Returns:
+            (batch_size * sequence_length, hidden_dim)
+        """
+        final_hidden_states = torch.zeros_like(hidden_states)
+        expert_mask = torch.nn.functional.one_hot(top_k_index, num_classes=self.num_experts).permute(2, 1, 0)
+
+        expert_hit = torch.greater(expert_mask.sum(dim=(-1, -2)), 0).nonzero()
+        for expert_idx in expert_hit:
+            idx, top_x = torch.where(expert_mask[expert_idx].squeeze(0))
+            current_state = hidden_states[None, top_x].reshape(-1, hidden_states.shape[-1])
+            current_hidden_states = self[expert_idx](current_state) * top_k_weights[top_x, idx, None]
+            final_hidden_states.index_add_(0, top_x, current_hidden_states.to(hidden_states.dtype))
+        return final_hidden_states
+
+
+class MiniMaxM2SparseMoeBlock(nn.Module):
+    def __init__(self, config):
+        super().__init__()
+        self.top_k = config.num_experts_per_tok
+        self.jitter_noise = config.router_jitter_noise
+        self.gate = nn.Linear(config.hidden_size, config.num_local_experts, bias=False)
+        self.experts = MiniMaxM2Experts(config)
+        self.register_buffer("e_score_correction_bias", torch.zeros(config.num_local_experts))
+
+    def route_tokens_to_experts(self, router_logits):
+        routing_weights = torch.nn.functional.sigmoid(router_logits.float())
+        scores_for_choice = routing_weights + self.e_score_correction_bias
+        _, top_k_index = torch.topk(scores_for_choice, self.top_k, dim=-1, sorted=False)
+        top_k_weights = routing_weights.gather(1, top_k_index)
+        top_k_weights /= top_k_weights.sum(dim=-1, keepdim=True)
+        return top_k_index, top_k_weights.to(router_logits.dtype)
+
+    def forward(self, hidden_states: torch.Tensor) -> tuple[torch.Tensor, torch.Tensor]:
+        batch_size, sequence_length, hidden_dim = hidden_states.shape
+        if self.training and self.jitter_noise > 0:
+            hidden_states *= torch.empty_like(hidden_states).uniform_(1.0 - self.jitter_noise, 1.0 + self.jitter_noise)
+        hidden_states = hidden_states.view(-1, hidden_states.shape[-1])
+        router_logits = self.gate(hidden_states)
+        top_k_index, top_k_weights = self.route_tokens_to_experts(router_logits)
+        hidden_states = self.experts(hidden_states, top_k_index, top_k_weights.to(hidden_states.dtype))
+        hidden_states = hidden_states.reshape(batch_size, sequence_length, hidden_dim)
+        return hidden_states, router_logits
+
+
+@use_kernel_forward_from_hub("RMSNorm")
+class MiniMaxM2RMSNorm(nn.Module):
+    def __init__(self, hidden_size, eps=1e-6):
+        """
+        MiniMaxM2RMSNorm 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}"
+
+
+def repeat_kv(hidden_states: torch.Tensor, n_rep: int) -> torch.Tensor:
+    """
+    This is the equivalent of torch.repeat_interleave(x, dim=1, repeats=n_rep). The hidden states go from (batch,
+    num_key_value_heads, seqlen, head_dim) to (batch, num_attention_heads, seqlen, head_dim)
+    """
+    batch, num_key_value_heads, slen, head_dim = hidden_states.shape
+    if n_rep == 1:
+        return hidden_states
+    hidden_states = hidden_states[:, :, None, :, :].expand(batch, num_key_value_heads, n_rep, slen, head_dim)
+    return hidden_states.reshape(batch, num_key_value_heads * n_rep, slen, head_dim)
+
+
+def eager_attention_forward(
+    module: nn.Module,
+    query: torch.Tensor,
+    key: torch.Tensor,
+    value: torch.Tensor,
+    attention_mask: Optional[torch.Tensor],
+    scaling: float,
+    dropout: float = 0.0,
+    **kwargs: Unpack[TransformersKwargs],
+):
+    key_states = repeat_kv(key, module.num_key_value_groups)
+    value_states = repeat_kv(value, module.num_key_value_groups)
+
+    attn_weights = torch.matmul(query, key_states.transpose(2, 3)) * scaling
+    if attention_mask is not None:
+        causal_mask = attention_mask[:, :, :, : key_states.shape[-2]]
+        attn_weights = attn_weights + causal_mask
+
+    attn_weights = nn.functional.softmax(attn_weights, dim=-1, dtype=torch.float32).to(query.dtype)
+    attn_weights = nn.functional.dropout(attn_weights, p=dropout, training=module.training)
+    attn_output = torch.matmul(attn_weights, value_states)
+    attn_output = attn_output.transpose(1, 2).contiguous()
+
+    return attn_output, attn_weights
+
+
+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)
+
+    # Keep half or full tensor for later concatenation
+    rotary_dim = cos.shape[-1]
+    q_rot, q_pass = q[..., :rotary_dim], q[..., rotary_dim:]
+    k_rot, k_pass = k[..., :rotary_dim], k[..., rotary_dim:]
+
+    # Apply rotary embeddings on the first half or full tensor
+    q_embed = (q_rot * cos) + (rotate_half(q_rot) * sin)
+    k_embed = (k_rot * cos) + (rotate_half(k_rot) * sin)
+
+    # Concatenate back to full shape
+    q_embed = torch.cat([q_embed, q_pass], dim=-1)
+    k_embed = torch.cat([k_embed, k_pass], dim=-1)
+    return q_embed, k_embed
+
+
+class MiniMaxM2Attention(nn.Module):
+    """Multi-headed attention from 'Attention Is All You Need' paper"""
+
+    def __init__(self, config: MiniMaxM2Config, layer_idx: int):
+        super().__init__()
+        self.config = config
+        self.layer_idx = layer_idx
+        self.head_dim = getattr(config, "head_dim", None) or config.hidden_size // config.num_attention_heads
+        self.num_key_value_groups = config.num_attention_heads // config.num_key_value_heads
+        self.scaling = self.head_dim**-0.5
+        self.attention_dropout = config.attention_dropout
+        self.is_causal = True
+        self.q_proj = nn.Linear(config.hidden_size, config.num_attention_heads * self.head_dim, bias=False)
+        self.k_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=False)
+        self.v_proj = nn.Linear(config.hidden_size, config.num_key_value_heads * self.head_dim, bias=False)
+        self.o_proj = nn.Linear(config.num_attention_heads * self.head_dim, config.hidden_size, bias=False)
+
+        self.use_qk_norm = config.use_qk_norm
+        if self.use_qk_norm:
+            self.q_norm = MiniMaxM2RMSNorm(self.head_dim * config.num_attention_heads, eps=config.rms_norm_eps)
+            self.k_norm = MiniMaxM2RMSNorm(self.head_dim * config.num_key_value_heads, eps=config.rms_norm_eps)
+
+    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor],
+        attention_mask: Optional[torch.Tensor],
+        past_key_values: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[FlashAttentionKwargs],
+    ) -> tuple[torch.Tensor, Optional[torch.Tensor]]:
+        input_shape = hidden_states.shape[:-1]
+        hidden_shape = (*input_shape, -1, self.head_dim)
+
+        query_states = self.q_proj(hidden_states)
+        key_states = self.k_proj(hidden_states)
+        value_states = self.v_proj(hidden_states)
+
+        if self.use_qk_norm:  # main diff from Llama
+            query_states = self.q_norm(query_states)
+            key_states = self.k_norm(key_states)
+
+        key_states = key_states.view(hidden_shape)
+        query_states = query_states.view(hidden_shape)
+        value_states = value_states.view(hidden_shape)
+
+        query_states = query_states.transpose(1, 2)
+        key_states = key_states.transpose(1, 2)
+        value_states = value_states.transpose(1, 2)
+
+        cos, sin = position_embeddings
+        query_states, key_states = apply_rotary_pos_emb(query_states, key_states, cos, sin)
+
+        if past_key_values is not None:
+            # sin and cos are specific to RoPE models; position_ids needed for the static cache
+            cache_kwargs = {"sin": sin, "cos": cos, "cache_position": cache_position}
+            key_states, value_states = past_key_values.update(key_states, value_states, self.layer_idx, cache_kwargs)
+
+        attention_interface: Callable = eager_attention_forward
+        if self.config._attn_implementation != "eager":
+            attention_interface = ALL_ATTENTION_FUNCTIONS[self.config._attn_implementation]
+
+        attn_output, attn_weights = attention_interface(
+            self,
+            query_states,
+            key_states,
+            value_states,
+            attention_mask,
+            dropout=0.0 if not self.training else self.attention_dropout,
+            scaling=self.scaling,
+            **kwargs,
+        )
+
+        attn_output = attn_output.reshape(*input_shape, -1).contiguous()
+        attn_output = self.o_proj(attn_output)
+        return attn_output, attn_weights
+
+
+class MiniMaxM2DecoderLayer(GradientCheckpointingLayer):
+    def __init__(self, config: MiniMaxM2Config, layer_idx: int):
+        super().__init__()
+        self.hidden_size = config.hidden_size
+
+        self.self_attn = MiniMaxM2Attention(config, layer_idx)
+
+        self.block_sparse_moe = MiniMaxM2SparseMoeBlock(config)
+        self.input_layernorm = MiniMaxM2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.post_attention_layernorm = MiniMaxM2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+
+    @deprecate_kwarg("past_key_value", new_name="past_key_values", version="4.58")
+    def forward(
+        self,
+        hidden_states: torch.Tensor,
+        position_embeddings: tuple[torch.Tensor, torch.Tensor],
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> torch.FloatTensor:
+        residual = hidden_states
+
+        hidden_states = self.input_layernorm(hidden_states)
+
+        # Self Attention
+        hidden_states, _ = self.self_attn(
+            hidden_states=hidden_states,
+            position_embeddings=position_embeddings,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            cache_position=cache_position,
+            **kwargs,
+        )
+        hidden_states = residual + hidden_states
+
+        # Fully Connected
+        residual = hidden_states
+        hidden_states = self.post_attention_layernorm(hidden_states)
+        hidden_states, _ = self.block_sparse_moe(hidden_states)
+        hidden_states = residual + hidden_states
+
+        return hidden_states
+
+
+class MiniMaxM2RotaryEmbedding(nn.Module):
+    inv_freq: torch.Tensor  # fix linting for `register_buffer`
+
+    def __init__(self, config: MiniMaxM2Config, device=None):
+        super().__init__()
+        # BC: "rope_type" was originally "type"
+        if hasattr(config, "rope_scaling") and isinstance(config.rope_scaling, dict):
+            self.rope_type = config.rope_scaling.get("rope_type", config.rope_scaling.get("type"))
+        else:
+            self.rope_type = "default"
+        self.max_seq_len_cached = config.max_position_embeddings
+        self.original_max_seq_len = config.max_position_embeddings
+
+        self.config = config
+        self.rope_init_fn = ROPE_INIT_FUNCTIONS[self.rope_type]
+
+        inv_freq, self.attention_scaling = self.rope_init_fn(self.config, device)
+        self.register_buffer("inv_freq", inv_freq, persistent=False)
+        self.original_inv_freq = self.inv_freq
+
+    @torch.no_grad()
+    @dynamic_rope_update  # power user: used with advanced RoPE types (e.g. dynamic rope)
+    def forward(self, x, position_ids):
+        inv_freq_expanded = self.inv_freq[None, :, None].float().expand(position_ids.shape[0], -1, 1).to(x.device)
+        position_ids_expanded = position_ids[:, None, :].float()
+
+        device_type = x.device.type if isinstance(x.device.type, str) and x.device.type != "mps" else "cpu"
+        with torch.autocast(device_type=device_type, enabled=False):  # Force float32
+            freqs = (inv_freq_expanded.float() @ position_ids_expanded.float()).transpose(1, 2)
+            emb = torch.cat((freqs, freqs), dim=-1)
+            cos = emb.cos() * self.attention_scaling
+            sin = emb.sin() * self.attention_scaling
+
+        return cos.to(dtype=x.dtype), sin.to(dtype=x.dtype)
+
+
+@auto_docstring
+class MiniMaxM2PreTrainedModel(PreTrainedModel):
+    config: MiniMaxM2Config
+    base_model_prefix = "model"
+    supports_gradient_checkpointing = True
+    _no_split_modules = ["MiniMaxM2DecoderLayer"]
+    _skip_keys_device_placement = ["past_key_values"]
+    _supports_flash_attn = True
+    _supports_sdpa = True
+    _supports_flex_attn = True
+    _can_compile_fullgraph = False  # MoE models don't work with torch.compile (`torch.where(condition)` not supported)
+    _supports_attention_backend = True
+    _can_record_outputs = {
+        "router_logits": OutputRecorder(MiniMaxM2SparseMoeBlock, index=1),
+        "hidden_states": MiniMaxM2DecoderLayer,
+        "attentions": MiniMaxM2Attention,
+    }
+
+
+@auto_docstring
+class MiniMaxM2Model(MiniMaxM2PreTrainedModel):
+    def __init__(self, config: MiniMaxM2Config):
+        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(
+            [MiniMaxM2DecoderLayer(config, layer_idx) for layer_idx in range(config.num_hidden_layers)]
+        )
+        self.norm = MiniMaxM2RMSNorm(config.hidden_size, eps=config.rms_norm_eps)
+        self.rotary_emb = MiniMaxM2RotaryEmbedding(config=config)
+        self.gradient_checkpointing = False
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @check_model_inputs
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        use_cache: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> MoeModelOutputWithPast:
+        if (input_ids is None) ^ (inputs_embeds is not None):
+            raise ValueError("You must specify exactly one of input_ids or inputs_embeds")
+
+        if use_cache and past_key_values is None:
+            past_key_values = DynamicCache(config=self.config)
+
+        if inputs_embeds is None:
+            inputs_embeds = self.embed_tokens(input_ids)
+
+        if cache_position is None:
+            past_seen_tokens = past_key_values.get_seq_length() if past_key_values is not None else 0
+            cache_position = torch.arange(
+                past_seen_tokens, past_seen_tokens + inputs_embeds.shape[1], device=inputs_embeds.device
+            )
+        if position_ids is None:
+            position_ids = cache_position.unsqueeze(0)
+
+        mask_function = create_causal_mask if self.config.sliding_window is None else create_sliding_window_causal_mask
+        causal_mask = mask_function(
+            config=self.config,
+            input_embeds=inputs_embeds,
+            attention_mask=attention_mask,
+            cache_position=cache_position,
+            past_key_values=past_key_values,
+            position_ids=position_ids,
+        )
+
+        hidden_states = inputs_embeds
+
+        # create position embeddings to be shared across the decoder layers
+        position_embeddings = self.rotary_emb(hidden_states, position_ids)
+
+        for decoder_layer in self.layers[: self.config.num_hidden_layers]:
+            hidden_states = decoder_layer(
+                hidden_states,
+                position_embeddings=position_embeddings,
+                attention_mask=causal_mask,
+                position_ids=position_ids,
+                past_key_values=past_key_values,
+                use_cache=use_cache,
+                cache_position=cache_position,
+                **kwargs,
+            )
+
+        hidden_states = self.norm(hidden_states)
+
+        return MoeModelOutputWithPast(  # only diff with Mistral is the output type, we need MoE
+            last_hidden_state=hidden_states,
+            past_key_values=past_key_values,
+        )
+
+
+def load_balancing_loss_func(
+    gate_logits: Union[torch.Tensor, tuple[torch.Tensor], None],
+    num_experts: Optional[int] = None,
+    top_k=2,
+    attention_mask: Optional[torch.Tensor] = None,
+) -> Union[torch.Tensor, int]:
+    r"""
+    Computes auxiliary load balancing loss as in Switch Transformer - implemented in Pytorch.
+
+    See Switch Transformer (https://huggingface.co/papers/2101.03961) for more details. This function implements the loss
+    function presented in equations (4) - (6) of the paper. It aims at penalizing cases where the routing between
+    experts is too unbalanced.
+
+    Args:
+        gate_logits:
+            Logits from the `gate`, should be a tuple of model.config.num_hidden_layers tensors of
+            shape [batch_size X sequence_length, num_experts].
+        num_experts:
+            Number of experts
+        top_k:
+            The number of experts to route per-token, can be also interpreted as the `top-k` routing
+            parameter.
+        attention_mask (`torch.Tensor`, *optional*):
+            The attention_mask used in forward function
+            shape [batch_size X sequence_length] if not None.
+
+    Returns:
+        The auxiliary loss.
+    """
+    if gate_logits is None or not isinstance(gate_logits, tuple):
+        return 0
+
+    if isinstance(gate_logits, tuple):
+        compute_device = gate_logits[0].device
+        concatenated_gate_logits = torch.cat([layer_gate.to(compute_device) for layer_gate in gate_logits], dim=0)
+
+    routing_weights = torch.nn.functional.softmax(concatenated_gate_logits, dim=-1)
+
+    _, selected_experts = torch.topk(routing_weights, top_k, dim=-1)
+
+    expert_mask = torch.nn.functional.one_hot(selected_experts, num_experts)
+
+    if attention_mask is None:
+        # Compute the percentage of tokens routed to each experts
+        tokens_per_expert = torch.mean(expert_mask.float(), dim=0)
+
+        # Compute the average probability of routing to these experts
+        router_prob_per_expert = torch.mean(routing_weights, dim=0)
+    else:
+        batch_size, sequence_length = attention_mask.shape
+        num_hidden_layers = concatenated_gate_logits.shape[0] // (batch_size * sequence_length)
+
+        # Compute the mask that masks all padding tokens as 0 with the same shape of expert_mask
+        expert_attention_mask = (
+            attention_mask[None, :, :, None, None]
+            .expand((num_hidden_layers, batch_size, sequence_length, top_k, num_experts))
+            .reshape(-1, top_k, num_experts)
+            .to(compute_device)
+        )
+
+        # Compute the percentage of tokens routed to each experts
+        tokens_per_expert = torch.sum(expert_mask.float() * expert_attention_mask, dim=0) / torch.sum(
+            expert_attention_mask, dim=0
+        )
+
+        # Compute the mask that masks all padding tokens as 0 with the same shape of tokens_per_expert
+        router_per_expert_attention_mask = (
+            attention_mask[None, :, :, None]
+            .expand((num_hidden_layers, batch_size, sequence_length, num_experts))
+            .reshape(-1, num_experts)
+            .to(compute_device)
+        )
+
+        # Compute the average probability of routing to these experts
+        router_prob_per_expert = torch.sum(routing_weights * router_per_expert_attention_mask, dim=0) / torch.sum(
+            router_per_expert_attention_mask, dim=0
+        )
+
+    overall_loss = torch.sum(tokens_per_expert * router_prob_per_expert.unsqueeze(0))
+    return overall_loss * num_experts
+
+
+@auto_docstring
+class MiniMaxM2ForCausalLM(MiniMaxM2PreTrainedModel, 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 = MiniMaxM2Model(config)
+        self.vocab_size = config.vocab_size
+        self.lm_head = nn.Linear(config.hidden_size, config.vocab_size, bias=False)
+        self.router_aux_loss_coef = config.router_aux_loss_coef
+        self.num_experts = config.num_local_experts
+        self.num_experts_per_tok = config.num_experts_per_tok
+
+        # Initialize weights and apply final processing
+        self.post_init()
+
+    @can_return_tuple
+    @auto_docstring
+    def forward(
+        self,
+        input_ids: Optional[torch.LongTensor] = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Cache] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_router_logits: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        logits_to_keep: Union[int, torch.Tensor] = 0,
+        **kwargs: Unpack[TransformersKwargs],
+    ) -> MoeCausalLMOutputWithPast:
+        r"""
+        labels (`torch.LongTensor` of shape `(batch_size, sequence_length)`, *optional*):
+            Labels for computing the masked language modeling loss. Indices should either be in `[0, ...,
+            config.vocab_size]` or -100 (see `input_ids` docstring). Tokens with indices set to `-100` are ignored
+            (masked), the loss is only computed for the tokens with labels in `[0, ..., config.vocab_size]`.
+
+        Example:
+
+        ```python
+        >>> from transformers import AutoTokenizer, MiniMaxM2ForCausalLM
+
+        >>> model = MiniMaxM2ForCausalLM.from_pretrained("mistralai/MiniMaxM2-8x7B-v0.1")
+        >>> tokenizer = AutoTokenizer.from_pretrained("mistralai/MiniMaxM2-8x7B-v0.1")
+
+        >>> 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_router_logits = (
+            output_router_logits if output_router_logits is not None else self.config.output_router_logits
+        )
+
+        # decoder outputs consists of (dec_features, layer_state, dec_hidden, dec_attn)
+        outputs: MoeModelOutputWithPast = self.model(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            use_cache=use_cache,
+            output_router_logits=output_router_logits,
+            cache_position=cache_position,
+            **kwargs,
+        )
+
+        hidden_states = outputs.last_hidden_state
+        # Only compute necessary logits, and do not upcast them to float if we are not computing the loss
+        slice_indices = slice(-logits_to_keep, None) if isinstance(logits_to_keep, int) else logits_to_keep
+        logits = self.lm_head(hidden_states[:, slice_indices, :])
+
+        loss = None
+        if labels is not None:
+            loss = self.loss_function(logits, labels, self.vocab_size, **kwargs)
+
+        aux_loss = None
+        if output_router_logits:
+            aux_loss = load_balancing_loss_func(
+                outputs.router_logits,
+                self.num_experts,
+                self.num_experts_per_tok,
+                attention_mask,
+            )
+            if labels is not None:
+                loss += self.router_aux_loss_coef * aux_loss.to(loss.device)  # make sure to reside in the same device
+
+        return MoeCausalLMOutputWithPast(
+            loss=loss,
+            aux_loss=aux_loss,
+            logits=logits,
+            past_key_values=outputs.past_key_values,
+            hidden_states=outputs.hidden_states,
+            attentions=outputs.attentions,
+            router_logits=outputs.router_logits,
+        )
+
+
+class MiniMaxM2ForSequenceClassification(GenericForSequenceClassification, MiniMaxM2PreTrainedModel):
+    pass
+
+
+class MiniMaxM2ForTokenClassification(GenericForTokenClassification, MiniMaxM2PreTrainedModel):
+    pass
+
+
+class MiniMaxM2ForQuestionAnswering(GenericForQuestionAnswering, MiniMaxM2PreTrainedModel):
+    pass
+
+
+__all__ = [
+    "MiniMaxM2ForCausalLM",
+    "MiniMaxM2ForQuestionAnswering",
+    "MiniMaxM2Model",
+    "MiniMaxM2PreTrainedModel",
+    "MiniMaxM2ForSequenceClassification",
+    "MiniMaxM2ForTokenClassification",
+]

tokenizer_config.json

@@ -0,0 +1,495 @@
+{
+  "added_tokens_decoder": {
+  "200000": {
+      "content": "]!p~[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200001": {
+      "content": "<fim_prefix>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200002": {
+      "content": "<fim_middle>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200003": {
+      "content": "<fim_suffix>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200004": {
+      "content": "<fim_pad>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200005": {
+      "content": "<reponame>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200006": {
+      "content": "<filename>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200007": {
+      "content": "<gh_stars>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200008": {
+      "content": "<issue_start>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200009": {
+      "content": "<issue_comment>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200010": {
+      "content": "<issue_closed>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200011": {
+    "content": "<jupyter_start>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false,
+    "special": true
+  },
+  "200012": {
+      "content": "<jupyter_text>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200013": {
+      "content": "<jupyter_code>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200014": {
+      "content": "<jupyter_output>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200015": {
+    "content": "<empty_output>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false,
+    "special": true
+  },
+  "200016": {
+      "content": "<commit_before>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200017": {
+      "content": "<commit_msg>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200018": {
+      "content": "<commit_after>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200019": {
+      "content": "]~b]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200020": {
+      "content": "[e~[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200021": {
+      "content": "]!d~[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200022": {
+      "content": "<function_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200023": {
+      "content": "<code_interpreter>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200024": {
+      "content": "]<]speech[>[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200025": {
+      "content": "]<]image[>[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200026": {
+      "content": "]<]video[>[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200027": {
+      "content": "]<]start of speech[>[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200028": {
+      "content": "]<]end of speech[>[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200029": {
+      "content": "]<]start of image[>[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200030": {
+      "content": "]<]end of image[>[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200031": {
+      "content": "]<]start of video[>[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200032": {
+      "content": "]<]end of video[>[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200033": {
+      "content": "]<]vision pad[>[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200034": {
+      "content": "]~!b[",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200035": {
+      "content": "<jupyter_error>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200036": {
+      "content": "<add_file>",
+      "single_word": false,
+      "lstrip": false,
+      "rstrip": false,
+      "normalized": false,
+      "special": true
+  },
+  "200037": {
+      "content": "<delete_file>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200038": {
+      "content": "<rename_file>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200039": {
+      "content": "<edit_file>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200040": {
+      "content": "<commit_message>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200041": {
+      "content": "<empty_source_file>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200042": {
+      "content": "<repo_struct>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+  },
+  "200043": {
+    "content": "<code_context>",
+    "single_word": false,
+    "lstrip": false,
+    "rstrip": false,
+    "normalized": false,
+    "special": true
+  },
+  "200044": {
+    "content": "<file_content>",
+    "single_word": false,
+    "lstrip": false,
+    "rstrip": false,
+    "normalized": false,
+    "special": true
+  },
+  "200045": {
+    "content": "<source_files>",
+    "single_word": false,
+    "lstrip": false,
+    "rstrip": false,
+    "normalized": false,
+    "special": true
+  },
+  "200046": {
+    "content": "<pr_start>",
+    "single_word": false,
+    "lstrip": false,
+    "rstrip": false,
+    "normalized": false,
+    "special": true
+  },
+  "200047": {
+    "content": "<review_comment>",
+    "single_word": false,
+    "lstrip": false,
+    "rstrip": false,
+    "normalized": false,
+    "special": true
+  },
+  "200048": {
+    "content": "<filepath>",
+    "single_word": false,
+    "lstrip": false,
+    "rstrip": false,
+    "normalized": false,
+    "special": true
+  },
+  "200049": {
+    "content": "<file_sep>",
+    "single_word": false,
+    "lstrip": false,
+    "rstrip": false,
+    "normalized": false,
+    "special": true
+  },
+  "200050": {
+    "content": "<think>",
+    "single_word": false,
+    "lstrip": false,
+    "rstrip": false,
+    "normalized": false,
+    "special": false
+  },
+  "200051": {
+    "content": "</think>",
+    "single_word": false,
+    "lstrip": false,
+    "rstrip": false,
+    "normalized": false,
+    "special": false
+  },
+  "200052": {
+    "content": "<minimax:tool_call>",
+    "single_word": false,
+    "lstrip": false,
+    "rstrip": false,
+    "normalized": false,
+    "special": false
+  },  
+  "200053": {
+    "content": "</minimax:tool_call>",
+    "single_word": false,
+    "lstrip": false,
+    "rstrip": false,
+    "normalized": false,
+    "special": false
+  }
+  },
+  "additional_special_tokens": [
+        "<code_interpreter>",
+        "<commit_after>",
+        "<commit_before>",
+        "<commit_msg>",
+        "<empty_output>",
+        "<filename>",
+        "<fim_middle>",
+        "<fim_pad>",
+        "<fim_prefix>",
+        "<fim_suffix>",
+        "<function_call>",
+        "<gh_stars>",
+        "]<]speech[>[",
+        "]<]image[>[",
+        "]<]video[>[",
+        "]<]start of speech[>[",
+        "]<]end of speech[>[",
+        "]<]start of image[>[",
+        "]<]end of image[>[",
+        "]<]start of video[>[",
+        "]<]end of video[>[",
+        "]<]vision pad[>[",
+        "]~!b[",
+        "<issue_closed>",
+        "<issue_comment>",
+        "<issue_start>",
+        "<jupyter_code>",
+        "<jupyter_output>",
+        "<jupyter_start>",
+        "<jupyter_text>",
+        "<reponame>",
+        "[e~[",
+        "]!d~[",
+        "]!p~[",
+        "]~b]",
+        "<jupyter_error>",
+        "<add_file>",
+        "<delete_file>",
+        "<rename_file>",
+        "<edit_file>",
+        "<commit_message>",
+        "<empty_source_file>",
+        "<repo_struct>",
+        "<code_context>",
+        "<file_content>",
+        "<source_files>",
+        "<pr_start>",
+        "<review_comment>",
+        "<filepath>",
+        "<file_sep>"
+    ],
+  "add_prefix_space": false,
+  "bos_token": "]~!b[",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "[e~[",
+  "model_max_length": 40960000,
+  "tokenizer_class": "GPT2Tokenizer",
+  "unk_token": "]!d~["
+}