Upload folder using huggingface_hub

.gitattributes

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

README.md

@@ -0,0 +1,318 @@
+---
+pipeline_tag: text-generation
+license: other
+license_name: modified-mit
+license_link: https://github.com/MiniMax-AI/MiniMax-M2.1/blob/main/LICENSE
+library_name: llm-compressor
+tags:
+- fp8
+- awq
+- conversational
+- vllm
+- code
+- devops
+- software engineering
+- engineer
+- developer
+- architect
+- stem
+- agent
+datasets:
+- HuggingFaceH4/ultrachat_200k
+- databricks/databricks-dolly-15k
+- neuralmagic/calibration
+- HuggingFaceH4/no_robots
+- nvidia/HelpSteer
+- garage-bAInd/Open-Platypus
+- PJMixers/grimulkan_physical-reasoning-ShareGPT
+- PJMixers/grimulkan_theory-of-mind-ShareGPT
+- HuggingFaceH4/Multilingual-Thinking
+- ServiceNow-AI/M2Lingual
+- interstellarninja/hermes_reasoning_tool_use
+- deepmind/code_contests
+- dh02391735/stackoverflow-kubernetes-questions
+- diversoailab/humaneval-rust
+- ammarnasr/the-stack-rust-clean
+- CSJianYang/CodeArena
+- nvidia/OpenCodeInstruct
+- nvidia/Llama-Nemotron-Post-Training-Dataset
+- nvidia/Nemotron-Competitive-Programming-v1
+- rombodawg/code_bagel_hermes-2.5
+- MathArena/project_euler
+- nvidia/Nemotron-Math-Proofs-v1
+- nvidia/OpenMathInstruct-2
+- nvidia/OpenScienceReasoning-2
+- MegaScience/MegaScience
+- OpenMed/Medical-Reasoning-SFT-GPT-OSS-120B
+- ccdv/pubmed-summarization
+- gbharti/finance-alpaca
+- vladlen32230/summarization-yahoo-stock-finance-article-text
+- fka/awesome-chatgpt-prompts
+- theoldmandthesea/17k_business_book
+- ruggsea/stanford-encyclopedia-of-philosophy_instruct
+- mlfoundations-dev/stackexchange_philosophy
+- FreedomIntelligence/SocraticChat
+- Gryphe/Opus-WritingPrompts
+- anthracite-org/nopm_claude_writing_fixed
+- zerofata/Roleplay-Anime-Characters
+- zerofata/Instruct-Anime
+- zerofata/Instruct-Anime-CreativeWriting
+- sam-paech/gutenberg3-generalfiction-scifi-fantasy-romance-adventure-dpo
+- PocketDoc/Dans-Prosemaxx-Adventure
+- anthracite-org/stheno-filtered-v1.1
+- KaraKaraWitch/TvTroper-2025
+- AquaV/US-Army-Survival-Sharegpt
+- AquaV/Interrogation-Sharegpt
+- AquaV/Multi-Environment-Operations-Sharegpt
+- AquaV/Resistance-Sharegpt
+- PocketDoc/Dans-Kinomaxx-VanillaBackrooms
+base_model:
+- MiniMaxAI/MiniMax-M2.1
+---
+
+# MiniMax M2.1 (Mixed-Precision BF16 + INT4 AWQ)
+
+This strives to be the highest quality quant that can run on 192GiB VRAM
+
+> [!TIP]  
+> 💡This is a sister model to [mratsim/MiniMax-M2.1-FP8-INT4-AWQ](https://huggingface.co/mratsim/MiniMax-M2.1-FP8-INT4-AWQ)
+> with the original model FP8 weights pre-dequantized to BF16.
+>
+> This makes it compatible with 8x3090 systems (which don't have hardware FP8)
+> and also compatible with SGLang.
+
+It features:
+- That model has ensured that all experts are calibrated, not doing so is extremely detrimental, PR: https://github.com/vllm-project/llm-compressor/pull/2171
+  <details>
+  <summary>Visual showcase of why ensuring quantization of all MoE experts is important</summary>
+  
+  - Source: https://avtc.github.io/aquarium-side-by-side/
+  - Context: https://github.com/ModelCloud/GPTQModel/pull/2235
+  
+  ![image](https://cdn-uploads.huggingface.co/production/uploads/67f26fd2c7b14380431d1f5a/BDc3-0m3_WLl3ZmbBMhmd.png)
+  
+  </details>
+- Mixed precision with:
+  - self-attention weights copied directly from the official version (default FP8 with 2D-blocks)
+  - experts weights quantized using AWQ W4A16G32 scheme (4-bit weights, 16-bit activations, scaling factor per group of 32 weights)
+- High-quality large and diverse dataset with programming and devops focus
+  as well as domain-specific knowledge (math, sciences, medical, finance, business, humanities, philosophy, creative writing), general knowledge, pop culture and behavioral situations because we never code in a vacuum. And we want to make sure all experts are calibrated to the full range of their activations.
+- Calibration explicitly tests multilingual capabilities:
+  - Asia: Chinese, Hindi, Korean, Japanese
+  - Europe: French, German, Portuguese, Russian, Spanish
+  - Middle-East: Arabic, Hebrew, Turkish
+- Calibration explicitly tests 60 programming languages and not just Python:
+  - Imperative programming: C, C++, Go, Zig, ...
+  - Functional programming: Haskell, F#, OCaml, Erlang, Lisp, Clojure ...
+  - Web-focused: HTML/CSS, Typescript, PHP, ...
+  - Mixed paradigm: D, Kotlin, Nim, Rust, Swift, ...
+  - Theorem provers: Coq, Lean
+  - Low-level: ARM64 assembly, x86-64 assembly, LLVM IR
+  - GPU Programming: Cuda, Vulkan, Apple Metal
+  - Game Programming: GDScript, GLSL
+  - Domain-specific: MATLAB, Julia, Solidity, R
+- Calibration tries to ensure coverage for a wide variety of experience (from explaining concepts to your grandmother to debugging Kubernetes logs)
+- Built by a dev, for devs (and it looks very good for STEM as well)
+
+It uses my new declarative quantization framework https://github.com/mratsim/quantizers which facilitates highly-tuned calibration sets: [calibrate_software_engineer.yaml](./calibrate_software_engineer.yaml)
+
+<details>
+<summary>This has taken several days and contribution and bug reports to the ecosystem, I hope you find it useful.</summary>
+  
+- https://github.com/vllm-project/llm-compressor/pull/2171
+- https://github.com/vllm-project/llm-compressor/issues/2172
+- https://github.com/vllm-project/vllm/issues/31623
+- https://github.com/sgl-project/sglang/issues/16276
+- https://github.com/sgl-project/sglang/issues/16295
+
+</details>
+
+## 📥 Usage & Running Instructions
+
+The model was tested with SGLang + 2x RTX Pro 6000, here is a script suitable for such configuration with the maximum 196,608 context length. This uses 92.5GiB of VRAM with the flashinfer backend.
+
+Please refer to [mratsim/MiniMax-M2.1-FP8-INT4-AWQ#running-script](https://huggingface.co/mratsim/MiniMax-M2.1-FP8-INT4-AWQ#running-script)
+for running it in vLLM
+
+### Running script
+
+`--trust-remote-code` is necessary until the transformers team merges github.com/huggingface/transformers/pull/42028
+
+You have 2 reasoning parsers;
+- `minimax`, puts the reasoning content in a special field like DeepSeek models that is usually rendered in a specific manner in frontends.
+- `minimax_append_think`, puts the reasoning into `<think>reasoning_content</think>` and that is sent as normal text. Few frontends properly render that, I'm aware of [Cherry Studio](https://github.com/CherryHQ/cherry-studio) on Desktop and [ChatterUI](https://github.com/Vali-98/ChatterUI) on Android.
+
+The reason why `minimax_append_think` was introduced was Interleaved Thinking and having the model build upon it's previous thinking (usually frontends discard the thinking trace)
+
+> [!TIP]  
+> 💡In the sister model, I mentioned that with the recommended parameters the model tends to get stuck in repetition loops.\
+> This does not seem to happen with SGLang hence "repetition_penalty: 1.10, frequency_penalty: 0.40" are not used. \
+> There is no way to override such settings without editing generation_config.json anyway: https://github.com/sgl-project/sglang/issues/15487
+
+> [!NOTE]
+> I have not yet found a way to enable speculative decoding and the max 196608 context size and over 10 concurrent requests within 192 GIB of VRAM
+
+```bash
+# Model configuration (Mandatory)
+MODEL="mratsim/MiniMax-M2.1-BF16-INT4-AWQ"
+MODELNAME="MiniMax-M2.1"
+GPU_UTIL=0.97
+SGLANG_PORT=8000
+
+# Prevent memory fragmentation
+export PYTORCH_ALLOC_CONF=expandable_segments:True,max_split_size_mb:512
+
+python3 -m sglang.launch_server \
+    --host 0.0.0.0 \
+    --port "${SGLANG_PORT}" \
+    --sleep-on-idle \
+    --disable-custom-all-reduce \
+    --max-running-requests 12 \
+    --cuda-graph-max-bs 12 \
+    --attention-backend flashinfer \
+    --served-model-name "${MODELNAME}" \
+    --model-path "${MODEL}" \
+    --tool-call-parser minimax-m2 \
+    --reasoning-parser minimax \
+    --trust-remote-code \
+    --tp 2 \
+    --mem-fraction-static ${GPU_UTIL} \
+    "$@"
+    # --reasoning-parser minimax-append-think \
+```
+
+## 🔬 Quantization method
+
+Quantization was quite complex for this model and was done in 3 steps:
+1. Original weights are in FP8, they were dequantized to FP16 due to llm-compressor not being able to process FP8.
+2. llm-compressor was used to quantize the MLP experts projection using AWQ, with [PR #2171](https://github.com/vllm-project/llm-compressor/pull/2171) to ensure they were all activated.
+3. Stitching the FrankenQuant: I combined the original weights, including the 2D-block FP8, with the experts-only AWQ weights.
+
+The llmcompressor library was used with the following recipe:
+
+```yaml
+default_stage:
+  default_modifiers:
+    AWQModifier:
+      config_groups:
+        mlp_experts_projections:
+          # Include only MLP expert weights for 4-bit quantization
+          targets: ["re:.*block_sparse_moe\\.experts\\.\\d+\\.(w1|w2|w3)$"]
+          weights:
+            num_bits: 4
+            type: int
+            symmetric: true
+            group_size: 32
+            strategy: group
+            dynamic: false
+            # actorder: group
+            observer: memoryless_minmax
+
+      mappings:
+        - smooth_layer: re:.*post_attention_layernorm$
+          balance_layers: ["re:.*w1$", "re:.*w3$"]
+        - smooth_layer: re:.*w3$
+          balance_layers: ["re:.*w2$"]
+      duo_scaling: true
+```
+
+The calibration set had 590 examples, 8192 sequence length, 60 programming languages, 12 spoken languages and is detailed at [calibrate_software_engineer.yaml](./calibrate_software_engineer.yaml)
+
+## Quantization theory and heuristics for manual tuning
+
+<details>
+<summary>In-depth overview of quantization theory and heuristics for manual tuning</summary>
+
+### Layers to quantize
+
+Quantization should be focused on Linear layers (also called Dense or Fully-Connected layers i.e. MatMul+Bias)
+In particular quantizing LayerNorm/RMSnorm layer is strongly discouraged, see [1]
+> LayerNorm in Quantization. Kovaleva et al. (2021); Wei et al. (2022) find that outliers in the
+> LayerNorm parameters of BERT (Devlin et al., 2019) cause difficulties in model compression.
+> Given the importance of LayerNorm, all the quantization methods we discuss above leave LayerNorm unquantized.
+
+This is also reported in Intel and Nvidia repo:
+- https://github.com/intel/neural-compressor/issues/1963#issuecomment-2274873441
+- https://github.com/NVIDIA/TensorRT/issues/4084#issuecomment-2294513950
+
+### Tensors to up-quantize
+
+If there is enough bits, down projections should be prioritized.
+
+According to [4]
+> Fig. 3: Maximum absolute value over layers for a LLaMA3-8B.
+> Each color represent a different projection and we clearly see that down_proj has the biggest
+> spikes in input and output. We also observe that RMSNorm propagate spikes through the entire model
+
+According to [5]
+> Figure 5(a) illustrates the extremal ratio across layers and modules in LLaMA2-7B, highlighting
+> that weight outliers are concentrated in the down-projection matrices Wdown
+> ℓ of the second layer and
+> the last two layers. Figures 5(b) and 5(c) provide detailed visualizations of these outliers in the last
+> two layers.
+
+### Mixture-of-Experts quantization (MoE)
+
+Mixture-of-Experts require specific quantization techniques.
+
+#### Mixed-precision quantization
+
+Some layers have a higher impact on LLM performance.
+According to [2], spending more bits in attention layers results in large gain compared to spending them in FFN layers.
+According to [3] on 2-bit quantization:
+- quantizing expert FFN layers do not seriously impact model quality
+- quantizing cross-attention has some impact
+- quantizing self-attention has a large impact
+- quantizing dense FFN has a very significant impact
+
+Hence to preserve model quality we should choose not to quantize dense FFN layers and self-attention layers.
+
+We notice that:
+- official MXFP4 weights of gpt-oss-120b from OpenAI keep self-attention in BF16:
+  - https://huggingface.co/openai/gpt-oss-120b/blob/main/model.safetensors.index.json
+- NVFP4 weights of DeepSeek-R1 quantized by Nvidia also keep self-attention in BF16:
+  - https://huggingface.co/nvidia/DeepSeek-R1-0528-FP4/blob/main/model.safetensors.index.json
+
+#### Layers with high-impact
+
+According to [2], giving more bits to the first `k` blocks have a significantly higher impact on model quality than for the same last `k` blocks.
+
+#### Expert quantization
+
+When quantizing MoE, quantizing activations is tricky as only a subset of experts are activated per request. You have to make sure all experts are calibrated.
+
+<details>
+<summary>Visual showcase of why ensuring quantization of all MoE experts is important</summary>
+
+- Source: https://avtc.github.io/aquarium-side-by-side/
+- Context: https://github.com/ModelCloud/GPTQModel/pull/2235
+
+![image](https://cdn-uploads.huggingface.co/production/uploads/67f26fd2c7b14380431d1f5a/BDc3-0m3_WLl3ZmbBMhmd.png)
+
+</details>
+
+## References
+
+1. Why Do Some Inputs Break Low-Bit LLM Quantization? (2025)\
+  Ting-Yun Chang, Muru Zhang, Jesse Thomason, Robin Jia\
+  https://arxiv.org/pdf/2506.12044
+
+2. Examining Post-Training Quantization for Mixture-of-Experts: A Benchmark (2024)\
+  Pingzhi Li, Xiaolong Jin, Yu Cheng, Tianlong Chen\
+  https://arxiv.org/pdf/2406.08155v1
+
+3. Mixture of Quantized Experts (MoQE): Complementary Effect of Low-bit Quantization and Robustness (2023)\
+  Young Jin Kim, Raffy Fahim, Hany Hassan Awadalla\
+  https://arxiv.org/pdf/2310.02410
+
+4. Precision Where It Matters: A Novel Spike\
+   Aware Mixed-Precision Quantization Strategy for\
+   LLaMA-based Language Models (2025)\
+   Lucas Maisonnave, Cyril Moineau, Olivier Bichler, and Fabrice Rastello\
+   https://arxiv.org/pdf/2504.21553
+
+5. Systematic Outliers in Large Language Models (2025)\
+   Yongqi An, Xu Zhao, Tao Yu, Ming Tang, Jinqiao Wang\
+   https://arxiv.org/pdf/2502.06415v2
+
+</details>

added_tokens.json

@@ -0,0 +1,56 @@
+{
+  "</minimax:tool_call>": 200053,
+  "</think>": 200051,
+  "<add_file>": 200036,
+  "<code_context>": 200043,
+  "<code_interpreter>": 200023,
+  "<commit_after>": 200018,
+  "<commit_before>": 200016,
+  "<commit_message>": 200040,
+  "<commit_msg>": 200017,
+  "<delete_file>": 200037,
+  "<edit_file>": 200039,
+  "<empty_output>": 200015,
+  "<empty_source_file>": 200041,
+  "<file_content>": 200044,
+  "<file_sep>": 200049,
+  "<filename>": 200006,
+  "<filepath>": 200048,
+  "<fim_middle>": 200002,
+  "<fim_pad>": 200004,
+  "<fim_prefix>": 200001,
+  "<fim_suffix>": 200003,
+  "<function_call>": 200022,
+  "<gh_stars>": 200007,
+  "<issue_closed>": 200010,
+  "<issue_comment>": 200009,
+  "<issue_start>": 200008,
+  "<jupyter_code>": 200013,
+  "<jupyter_error>": 200035,
+  "<jupyter_output>": 200014,
+  "<jupyter_start>": 200011,
+  "<jupyter_text>": 200012,
+  "<minimax:tool_call>": 200052,
+  "<pr_start>": 200046,
+  "<rename_file>": 200038,
+  "<repo_struct>": 200042,
+  "<reponame>": 200005,
+  "<review_comment>": 200047,
+  "<source_files>": 200045,
+  "<think>": 200050,
+  "[e~[": 200020,
+  "]!d~[": 200021,
+  "]!p~[": 200000,
+  "]<]end of image[>[": 200030,
+  "]<]end of speech[>[": 200028,
+  "]<]end of video[>[": 200032,
+  "]<]image[>[": 200025,
+  "]<]speech[>[": 200024,
+  "]<]start of image[>[": 200029,
+  "]<]start of speech[>[": 200027,
+  "]<]start of video[>[": 200031,
+  "]<]video[>[": 200026,
+  "]<]vision pad[>[": 200033,
+  "]~!b[": 200034,
+  "]~b]": 200019
+}

chat_template.jinja

@@ -0,0 +1,165 @@
+{# ----------‑‑‑ 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 -%}
+    {%- elif content is none -%}
+        {{- '' }}
+    {%- 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-M2.1 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}}
+            {%- if message.tool_calls[-1].function -%}
+                {%- set last_tool_call.name = message.tool_calls[-1].function.name -%}
+            {%- else -%}
+                {%- set last_tool_call.name = message.tool_calls[-1].name -%}
+            {%- endif -%}
+        {%- 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 +1,518 @@
+{
+  "architectures": [
+    "MiniMaxM2ForCausalLM"
+  ],
+  "attention_dropout": 0.0,
+  "attn_type_list": [
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1,
+    1
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_minimax_m2.MiniMaxM2Config",
+    "AutoModelForCausalLM": "modeling_minimax_m2.MiniMaxM2ForCausalLM"
+  },
+  "bos_token_id": 1,
+  "dtype": "bfloat16",
+  "eos_token_id": 2,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 3072,
+  "initializer_range": 0.02,
+  "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": 62,
+  "num_key_value_heads": 8,
+  "num_local_experts": 256,
+  "num_mtp_modules": 3,
+  "output_router_logits": false,
+  "partial_rotary_factor": 0.5,
+  "qk_norm_type": "per_layer",
+  "quantization_config": {
+    "config_groups": {
+      "group_0": {
+        "format": "pack-quantized",
+        "input_activations": null,
+        "output_activations": null,
+        "targets": [
+          "Linear",
+          "re:.*block_sparse_moe\\.experts\\.\\d+\\.(w1|w2|w3)$",
+          "re:.*self_attn\\.qkv_proj$"
+        ],
+        "weights": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": false,
+          "group_size": 32,
+          "num_bits": 4,
+          "observer": "memoryless_minmax",
+          "observer_kwargs": {},
+          "strategy": "group",
+          "symmetric": true,
+          "type": "int"
+        }
+      }
+    },
+    "format": "pack-quantized",
+    "global_compression_ratio": null,
+    "ignore": [
+      "model.layers.0.self_attn.q_proj",
+      "model.layers.0.self_attn.k_proj",
+      "model.layers.0.self_attn.v_proj",
+      "model.layers.0.self_attn.o_proj",
+      "model.layers.0.self_attn.qkv_proj",
+      "model.layers.0.block_sparse_moe.gate",
+      "model.layers.1.self_attn.q_proj",
+      "model.layers.1.self_attn.k_proj",
+      "model.layers.1.self_attn.v_proj",
+      "model.layers.1.self_attn.o_proj",
+      "model.layers.1.self_attn.qkv_proj",
+      "model.layers.1.block_sparse_moe.gate",
+      "model.layers.2.self_attn.q_proj",
+      "model.layers.2.self_attn.k_proj",
+      "model.layers.2.self_attn.v_proj",
+      "model.layers.2.self_attn.o_proj",
+      "model.layers.2.self_attn.qkv_proj",
+      "model.layers.2.block_sparse_moe.gate",
+      "model.layers.3.self_attn.q_proj",
+      "model.layers.3.self_attn.k_proj",
+      "model.layers.3.self_attn.v_proj",
+      "model.layers.3.self_attn.o_proj",
+      "model.layers.3.self_attn.qkv_proj",
+      "model.layers.3.block_sparse_moe.gate",
+      "model.layers.4.self_attn.q_proj",
+      "model.layers.4.self_attn.k_proj",
+      "model.layers.4.self_attn.v_proj",
+      "model.layers.4.self_attn.o_proj",
+      "model.layers.4.self_attn.qkv_proj",
+      "model.layers.4.block_sparse_moe.gate",
+      "model.layers.5.self_attn.q_proj",
+      "model.layers.5.self_attn.k_proj",
+      "model.layers.5.self_attn.v_proj",
+      "model.layers.5.self_attn.o_proj",
+      "model.layers.5.self_attn.qkv_proj",
+      "model.layers.5.block_sparse_moe.gate",
+      "model.layers.6.self_attn.q_proj",
+      "model.layers.6.self_attn.k_proj",
+      "model.layers.6.self_attn.v_proj",
+      "model.layers.6.self_attn.o_proj",
+      "model.layers.6.self_attn.qkv_proj",
+      "model.layers.6.block_sparse_moe.gate",
+      "model.layers.7.self_attn.q_proj",
+      "model.layers.7.self_attn.k_proj",
+      "model.layers.7.self_attn.v_proj",
+      "model.layers.7.self_attn.o_proj",
+      "model.layers.7.self_attn.qkv_proj",
+      "model.layers.7.block_sparse_moe.gate",
+      "model.layers.8.self_attn.q_proj",
+      "model.layers.8.self_attn.k_proj",
+      "model.layers.8.self_attn.v_proj",
+      "model.layers.8.self_attn.o_proj",
+      "model.layers.8.self_attn.qkv_proj",
+      "model.layers.8.block_sparse_moe.gate",
+      "model.layers.9.self_attn.q_proj",
+      "model.layers.9.self_attn.k_proj",
+      "model.layers.9.self_attn.v_proj",
+      "model.layers.9.self_attn.o_proj",
+      "model.layers.9.self_attn.qkv_proj",
+      "model.layers.9.block_sparse_moe.gate",
+      "model.layers.10.self_attn.q_proj",
+      "model.layers.10.self_attn.k_proj",
+      "model.layers.10.self_attn.v_proj",
+      "model.layers.10.self_attn.o_proj",
+      "model.layers.10.self_attn.qkv_proj",
+      "model.layers.10.block_sparse_moe.gate",
+      "model.layers.11.self_attn.q_proj",
+      "model.layers.11.self_attn.k_proj",
+      "model.layers.11.self_attn.v_proj",
+      "model.layers.11.self_attn.o_proj",
+      "model.layers.11.self_attn.qkv_proj",
+      "model.layers.11.block_sparse_moe.gate",
+      "model.layers.12.self_attn.q_proj",
+      "model.layers.12.self_attn.k_proj",
+      "model.layers.12.self_attn.v_proj",
+      "model.layers.12.self_attn.o_proj",
+      "model.layers.12.self_attn.qkv_proj",
+      "model.layers.12.block_sparse_moe.gate",
+      "model.layers.13.self_attn.q_proj",
+      "model.layers.13.self_attn.k_proj",
+      "model.layers.13.self_attn.v_proj",
+      "model.layers.13.self_attn.o_proj",
+      "model.layers.13.self_attn.qkv_proj",
+      "model.layers.13.block_sparse_moe.gate",
+      "model.layers.14.self_attn.q_proj",
+      "model.layers.14.self_attn.k_proj",
+      "model.layers.14.self_attn.v_proj",
+      "model.layers.14.self_attn.o_proj",
+      "model.layers.14.self_attn.qkv_proj",
+      "model.layers.14.block_sparse_moe.gate",
+      "model.layers.15.self_attn.q_proj",
+      "model.layers.15.self_attn.k_proj",
+      "model.layers.15.self_attn.v_proj",
+      "model.layers.15.self_attn.o_proj",
+      "model.layers.15.self_attn.qkv_proj",
+      "model.layers.15.block_sparse_moe.gate",
+      "model.layers.16.self_attn.q_proj",
+      "model.layers.16.self_attn.k_proj",
+      "model.layers.16.self_attn.v_proj",
+      "model.layers.16.self_attn.o_proj",
+      "model.layers.16.self_attn.qkv_proj",
+      "model.layers.16.block_sparse_moe.gate",
+      "model.layers.17.self_attn.q_proj",
+      "model.layers.17.self_attn.k_proj",
+      "model.layers.17.self_attn.v_proj",
+      "model.layers.17.self_attn.o_proj",
+      "model.layers.17.self_attn.qkv_proj",
+      "model.layers.17.block_sparse_moe.gate",
+      "model.layers.18.self_attn.q_proj",
+      "model.layers.18.self_attn.k_proj",
+      "model.layers.18.self_attn.v_proj",
+      "model.layers.18.self_attn.o_proj",
+      "model.layers.18.self_attn.qkv_proj",
+      "model.layers.18.block_sparse_moe.gate",
+      "model.layers.19.self_attn.q_proj",
+      "model.layers.19.self_attn.k_proj",
+      "model.layers.19.self_attn.v_proj",
+      "model.layers.19.self_attn.o_proj",
+      "model.layers.19.self_attn.qkv_proj",
+      "model.layers.19.block_sparse_moe.gate",
+      "model.layers.20.self_attn.q_proj",
+      "model.layers.20.self_attn.k_proj",
+      "model.layers.20.self_attn.v_proj",
+      "model.layers.20.self_attn.o_proj",
+      "model.layers.20.self_attn.qkv_proj",
+      "model.layers.20.block_sparse_moe.gate",
+      "model.layers.21.self_attn.q_proj",
+      "model.layers.21.self_attn.k_proj",
+      "model.layers.21.self_attn.v_proj",
+      "model.layers.21.self_attn.o_proj",
+      "model.layers.21.self_attn.qkv_proj",
+      "model.layers.21.block_sparse_moe.gate",
+      "model.layers.22.self_attn.q_proj",
+      "model.layers.22.self_attn.k_proj",
+      "model.layers.22.self_attn.v_proj",
+      "model.layers.22.self_attn.o_proj",
+      "model.layers.22.self_attn.qkv_proj",
+      "model.layers.22.block_sparse_moe.gate",
+      "model.layers.23.self_attn.q_proj",
+      "model.layers.23.self_attn.k_proj",
+      "model.layers.23.self_attn.v_proj",
+      "model.layers.23.self_attn.o_proj",
+      "model.layers.23.self_attn.qkv_proj",
+      "model.layers.23.block_sparse_moe.gate",
+      "model.layers.24.self_attn.q_proj",
+      "model.layers.24.self_attn.k_proj",
+      "model.layers.24.self_attn.v_proj",
+      "model.layers.24.self_attn.o_proj",
+      "model.layers.24.self_attn.qkv_proj",
+      "model.layers.24.block_sparse_moe.gate",
+      "model.layers.25.self_attn.q_proj",
+      "model.layers.25.self_attn.k_proj",
+      "model.layers.25.self_attn.v_proj",
+      "model.layers.25.self_attn.o_proj",
+      "model.layers.25.self_attn.qkv_proj",
+      "model.layers.25.block_sparse_moe.gate",
+      "model.layers.26.self_attn.q_proj",
+      "model.layers.26.self_attn.k_proj",
+      "model.layers.26.self_attn.v_proj",
+      "model.layers.26.self_attn.o_proj",
+      "model.layers.26.self_attn.qkv_proj",
+      "model.layers.26.block_sparse_moe.gate",
+      "model.layers.27.self_attn.q_proj",
+      "model.layers.27.self_attn.k_proj",
+      "model.layers.27.self_attn.v_proj",
+      "model.layers.27.self_attn.o_proj",
+      "model.layers.27.self_attn.qkv_proj",
+      "model.layers.27.block_sparse_moe.gate",
+      "model.layers.28.self_attn.q_proj",
+      "model.layers.28.self_attn.k_proj",
+      "model.layers.28.self_attn.v_proj",
+      "model.layers.28.self_attn.o_proj",
+      "model.layers.28.self_attn.qkv_proj",
+      "model.layers.28.block_sparse_moe.gate",
+      "model.layers.29.self_attn.q_proj",
+      "model.layers.29.self_attn.k_proj",
+      "model.layers.29.self_attn.v_proj",
+      "model.layers.29.self_attn.o_proj",
+      "model.layers.29.self_attn.qkv_proj",
+      "model.layers.29.block_sparse_moe.gate",
+      "model.layers.30.self_attn.q_proj",
+      "model.layers.30.self_attn.k_proj",
+      "model.layers.30.self_attn.v_proj",
+      "model.layers.30.self_attn.o_proj",
+      "model.layers.30.self_attn.qkv_proj",
+      "model.layers.30.block_sparse_moe.gate",
+      "model.layers.31.self_attn.q_proj",
+      "model.layers.31.self_attn.k_proj",
+      "model.layers.31.self_attn.v_proj",
+      "model.layers.31.self_attn.o_proj",
+      "model.layers.31.self_attn.qkv_proj",
+      "model.layers.31.block_sparse_moe.gate",
+      "model.layers.32.self_attn.q_proj",
+      "model.layers.32.self_attn.k_proj",
+      "model.layers.32.self_attn.v_proj",
+      "model.layers.32.self_attn.o_proj",
+      "model.layers.32.self_attn.qkv_proj",
+      "model.layers.32.block_sparse_moe.gate",
+      "model.layers.33.self_attn.q_proj",
+      "model.layers.33.self_attn.k_proj",
+      "model.layers.33.self_attn.v_proj",
+      "model.layers.33.self_attn.o_proj",
+      "model.layers.33.self_attn.qkv_proj",
+      "model.layers.33.block_sparse_moe.gate",
+      "model.layers.34.self_attn.q_proj",
+      "model.layers.34.self_attn.k_proj",
+      "model.layers.34.self_attn.v_proj",
+      "model.layers.34.self_attn.o_proj",
+      "model.layers.34.self_attn.qkv_proj",
+      "model.layers.34.block_sparse_moe.gate",
+      "model.layers.35.self_attn.q_proj",
+      "model.layers.35.self_attn.k_proj",
+      "model.layers.35.self_attn.v_proj",
+      "model.layers.35.self_attn.o_proj",
+      "model.layers.35.self_attn.qkv_proj",
+      "model.layers.35.block_sparse_moe.gate",
+      "model.layers.36.self_attn.q_proj",
+      "model.layers.36.self_attn.k_proj",
+      "model.layers.36.self_attn.v_proj",
+      "model.layers.36.self_attn.o_proj",
+      "model.layers.36.self_attn.qkv_proj",
+      "model.layers.36.block_sparse_moe.gate",
+      "model.layers.37.self_attn.q_proj",
+      "model.layers.37.self_attn.k_proj",
+      "model.layers.37.self_attn.v_proj",
+      "model.layers.37.self_attn.o_proj",
+      "model.layers.37.self_attn.qkv_proj",
+      "model.layers.37.block_sparse_moe.gate",
+      "model.layers.38.self_attn.q_proj",
+      "model.layers.38.self_attn.k_proj",
+      "model.layers.38.self_attn.v_proj",
+      "model.layers.38.self_attn.o_proj",
+      "model.layers.38.self_attn.qkv_proj",
+      "model.layers.38.block_sparse_moe.gate",
+      "model.layers.39.self_attn.q_proj",
+      "model.layers.39.self_attn.k_proj",
+      "model.layers.39.self_attn.v_proj",
+      "model.layers.39.self_attn.o_proj",
+      "model.layers.39.self_attn.qkv_proj",
+      "model.layers.39.block_sparse_moe.gate",
+      "model.layers.40.self_attn.q_proj",
+      "model.layers.40.self_attn.k_proj",
+      "model.layers.40.self_attn.v_proj",
+      "model.layers.40.self_attn.o_proj",
+      "model.layers.40.self_attn.qkv_proj",
+      "model.layers.40.block_sparse_moe.gate",
+      "model.layers.41.self_attn.q_proj",
+      "model.layers.41.self_attn.k_proj",
+      "model.layers.41.self_attn.v_proj",
+      "model.layers.41.self_attn.o_proj",
+      "model.layers.41.self_attn.qkv_proj",
+      "model.layers.41.block_sparse_moe.gate",
+      "model.layers.42.self_attn.q_proj",
+      "model.layers.42.self_attn.k_proj",
+      "model.layers.42.self_attn.v_proj",
+      "model.layers.42.self_attn.o_proj",
+      "model.layers.42.self_attn.qkv_proj",
+      "model.layers.42.block_sparse_moe.gate",
+      "model.layers.43.self_attn.q_proj",
+      "model.layers.43.self_attn.k_proj",
+      "model.layers.43.self_attn.v_proj",
+      "model.layers.43.self_attn.o_proj",
+      "model.layers.43.self_attn.qkv_proj",
+      "model.layers.43.block_sparse_moe.gate",
+      "model.layers.44.self_attn.q_proj",
+      "model.layers.44.self_attn.k_proj",
+      "model.layers.44.self_attn.v_proj",
+      "model.layers.44.self_attn.o_proj",
+      "model.layers.44.self_attn.qkv_proj",
+      "model.layers.44.block_sparse_moe.gate",
+      "model.layers.45.self_attn.q_proj",
+      "model.layers.45.self_attn.k_proj",
+      "model.layers.45.self_attn.v_proj",
+      "model.layers.45.self_attn.o_proj",
+      "model.layers.45.self_attn.qkv_proj",
+      "model.layers.45.block_sparse_moe.gate",
+      "model.layers.46.self_attn.q_proj",
+      "model.layers.46.self_attn.k_proj",
+      "model.layers.46.self_attn.v_proj",
+      "model.layers.46.self_attn.o_proj",
+      "model.layers.46.self_attn.qkv_proj",
+      "model.layers.46.block_sparse_moe.gate",
+      "model.layers.47.self_attn.q_proj",
+      "model.layers.47.self_attn.k_proj",
+      "model.layers.47.self_attn.v_proj",
+      "model.layers.47.self_attn.o_proj",
+      "model.layers.47.self_attn.qkv_proj",
+      "model.layers.47.block_sparse_moe.gate",
+      "model.layers.48.self_attn.q_proj",
+      "model.layers.48.self_attn.k_proj",
+      "model.layers.48.self_attn.v_proj",
+      "model.layers.48.self_attn.o_proj",
+      "model.layers.48.self_attn.qkv_proj",
+      "model.layers.48.block_sparse_moe.gate",
+      "model.layers.49.self_attn.q_proj",
+      "model.layers.49.self_attn.k_proj",
+      "model.layers.49.self_attn.v_proj",
+      "model.layers.49.self_attn.o_proj",
+      "model.layers.49.self_attn.qkv_proj",
+      "model.layers.49.block_sparse_moe.gate",
+      "model.layers.50.self_attn.q_proj",
+      "model.layers.50.self_attn.k_proj",
+      "model.layers.50.self_attn.v_proj",
+      "model.layers.50.self_attn.o_proj",
+      "model.layers.50.self_attn.qkv_proj",
+      "model.layers.50.block_sparse_moe.gate",
+      "model.layers.51.self_attn.q_proj",
+      "model.layers.51.self_attn.k_proj",
+      "model.layers.51.self_attn.v_proj",
+      "model.layers.51.self_attn.o_proj",
+      "model.layers.51.self_attn.qkv_proj",
+      "model.layers.51.block_sparse_moe.gate",
+      "model.layers.52.self_attn.q_proj",
+      "model.layers.52.self_attn.k_proj",
+      "model.layers.52.self_attn.v_proj",
+      "model.layers.52.self_attn.o_proj",
+      "model.layers.52.self_attn.qkv_proj",
+      "model.layers.52.block_sparse_moe.gate",
+      "model.layers.53.self_attn.q_proj",
+      "model.layers.53.self_attn.k_proj",
+      "model.layers.53.self_attn.v_proj",
+      "model.layers.53.self_attn.o_proj",
+      "model.layers.53.self_attn.qkv_proj",
+      "model.layers.53.block_sparse_moe.gate",
+      "model.layers.54.self_attn.q_proj",
+      "model.layers.54.self_attn.k_proj",
+      "model.layers.54.self_attn.v_proj",
+      "model.layers.54.self_attn.o_proj",
+      "model.layers.54.self_attn.qkv_proj",
+      "model.layers.54.block_sparse_moe.gate",
+      "model.layers.55.self_attn.q_proj",
+      "model.layers.55.self_attn.k_proj",
+      "model.layers.55.self_attn.v_proj",
+      "model.layers.55.self_attn.o_proj",
+      "model.layers.55.self_attn.qkv_proj",
+      "model.layers.55.block_sparse_moe.gate",
+      "model.layers.56.self_attn.q_proj",
+      "model.layers.56.self_attn.k_proj",
+      "model.layers.56.self_attn.v_proj",
+      "model.layers.56.self_attn.o_proj",
+      "model.layers.56.self_attn.qkv_proj",
+      "model.layers.56.block_sparse_moe.gate",
+      "model.layers.57.self_attn.q_proj",
+      "model.layers.57.self_attn.k_proj",
+      "model.layers.57.self_attn.v_proj",
+      "model.layers.57.self_attn.o_proj",
+      "model.layers.57.self_attn.qkv_proj",
+      "model.layers.57.block_sparse_moe.gate",
+      "model.layers.58.self_attn.q_proj",
+      "model.layers.58.self_attn.k_proj",
+      "model.layers.58.self_attn.v_proj",
+      "model.layers.58.self_attn.o_proj",
+      "model.layers.58.self_attn.qkv_proj",
+      "model.layers.58.block_sparse_moe.gate",
+      "model.layers.59.self_attn.q_proj",
+      "model.layers.59.self_attn.k_proj",
+      "model.layers.59.self_attn.v_proj",
+      "model.layers.59.self_attn.o_proj",
+      "model.layers.59.self_attn.qkv_proj",
+      "model.layers.59.block_sparse_moe.gate",
+      "model.layers.60.self_attn.q_proj",
+      "model.layers.60.self_attn.k_proj",
+      "model.layers.60.self_attn.v_proj",
+      "model.layers.60.self_attn.o_proj",
+      "model.layers.60.self_attn.qkv_proj",
+      "model.layers.60.block_sparse_moe.gate",
+      "model.layers.61.self_attn.q_proj",
+      "model.layers.61.self_attn.k_proj",
+      "model.layers.61.self_attn.v_proj",
+      "model.layers.61.self_attn.o_proj",
+      "model.layers.61.self_attn.qkv_proj",
+      "model.layers.61.block_sparse_moe.gate",
+      "lm_head"
+    ],
+    "kv_cache_scheme": null,
+    "quant_method": "compressed-tensors",
+    "quantization_status": "compressed",
+    "sparsity_config": {},
+    "transform_config": {},
+    "version": "0.13.1.dev0+g797d301.d20251228"
+  },
+  "rms_norm_eps": 1e-06,
+  "rope_theta": 5000000,
+  "rotary_dim": 64,
+  "router_aux_loss_coef": 0.001,
+  "router_jitter_noise": 0.0,
+  "scoring_func": "sigmoid",
+  "shared_intermediate_size": 0,
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "transformers_version": "4.57.3",
+  "use_cache": true,
+  "use_mtp": true,
+  "use_qk_norm": true,
+  "use_routing_bias": true,
+  "vocab_size": 200064
+}

configuration_minimax_m2.py

@@ -0,0 +1,214 @@
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           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,
+        )
+
+    def __repr__(self):
+        """
+        Overriding __repr__ to ensure that quantization_config does not contain Enum objects.
+        This prevents SyntaxError in torch.fx generated code which relies on repr() of the config.
+        """
+        print("Config sanitized in configuration_minimax_m2.py")
+        q_config = getattr(self, "quantization_config", None)
+        if isinstance(q_config, dict):
+            for k, v in list(q_config.items()):
+                if hasattr(v, "value"):
+                    # Replace Enum with its value
+                    q_config[k] = v.value
+                    print(f"  Enum: q_config[{k}] = {v.value}")
+        return super().__repr__()
+
+__all__ = ["MiniMaxM2Config"]

generation_config.json

@@ -0,0 +1,8 @@
+{
+  "bos_token_id": 200019,
+  "do_sample": true,
+  "eos_token_id": 200020,
+  "top_k": 40,
+  "top_p": 0.95,
+  "transformers_version": "4.57.3"
+}

minimax_to_bf16.py

@@ -0,0 +1,142 @@
+
+import os
+import json
+from argparse import ArgumentParser
+from glob import glob
+from tqdm import tqdm
+
+import torch
+from safetensors.torch import load_file, save_file
+
+
+def weight_dequant_fp8(weight_fp8, scale_inv):
+    """
+    Dequantize FP8 weights to BF16 using scale_inv.
+
+    Args:
+        weight_fp8: FP8 tensor
+        scale_inv: Inverse scale tensor (F32)
+
+    Returns:
+        BF16 tensor
+    """
+    # Convert FP8 to float32 first
+    weight_f32 = weight_fp8.to(torch.float32)
+
+    # Apply inverse scaling
+    # scale_inv shape is typically [out_features_blocks, in_features_blocks]
+    # We need to broadcast it properly to match weight dimensions
+    if scale_inv.dim() == 2:
+        # Expand scale_inv to match weight dimensions
+        out_blocks, in_blocks = scale_inv.shape
+        weight_blocks_out = weight_fp8.shape[0] // out_blocks
+        weight_blocks_in = weight_fp8.shape[1] // in_blocks
+
+        # Repeat scale_inv to match weight shape
+        scale_inv_expanded = scale_inv.repeat_interleave(weight_blocks_out, dim=0)
+        scale_inv_expanded = scale_inv_expanded.repeat_interleave(weight_blocks_in, dim=1)
+
+        weight_f32 = weight_f32 * scale_inv_expanded
+    else:
+        weight_f32 = weight_f32 * scale_inv
+
+    # Convert to BF16
+    return weight_f32.to(torch.bfloat16)
+
+
+def main(fp8_path, bf16_path):
+    torch.set_default_dtype(torch.bfloat16)
+    os.makedirs(bf16_path, exist_ok=True)
+
+    model_index_file = os.path.join(fp8_path, "model.safetensors.index.json")
+    with open(model_index_file, "r") as f:
+        model_index = json.load(f)
+
+    weight_map = model_index["weight_map"]
+
+    # Cache for loaded safetensor files
+    loaded_files = {}
+    fp8_weight_names = []
+
+    # Helper function to get tensor from the correct file
+    def get_tensor(tensor_name):
+        if tensor_name not in weight_map:
+            return None
+        file_name = weight_map[tensor_name]
+        if file_name not in loaded_files:
+            file_path = os.path.join(fp8_path, file_name)
+            loaded_files[file_name] = load_file(file_path, device="cuda")
+        return loaded_files[file_name][tensor_name]
+
+    safetensor_files = list(glob(os.path.join(fp8_path, "*.safetensors")))
+    safetensor_files = [f for f in safetensor_files if not f.endswith(".index.json")]
+    safetensor_files.sort()
+
+    print(f"Found {len(safetensor_files)} safetensor files to convert")
+
+    for safetensor_file in tqdm(safetensor_files, desc="Converting files"):
+        file_name = os.path.basename(safetensor_file)
+        current_state_dict = load_file(safetensor_file, device="cuda")
+        loaded_files[file_name] = current_state_dict
+
+        new_state_dict = {}
+
+        for weight_name, weight in current_state_dict.items():
+            # Skip scale_inv tensors
+            if weight_name.endswith("_scale_inv"):
+                continue
+
+            # Check if this is an FP8 weight (F8_E4M3 has element_size of 1)
+            if weight.dtype == torch.float8_e4m3fn or weight.element_size() == 1:
+                scale_inv_name = f"{weight_name}_scale_inv"
+                scale_inv = get_tensor(scale_inv_name)
+
+                if scale_inv is not None:
+                    fp8_weight_names.append(weight_name)
+                    new_state_dict[weight_name] = weight_dequant_fp8(weight, scale_inv)
+                else:
+                    print(f"Warning: Missing scale_inv tensor for {weight_name}, keeping as-is")
+                    new_state_dict[weight_name] = weight
+            else:
+                # Already BF16 or F32, keep as-is
+                new_state_dict[weight_name] = weight
+
+        # Save converted weights
+        new_safetensor_file = os.path.join(bf16_path, file_name)
+        save_file(new_state_dict, new_safetensor_file)
+
+        # Memory management: keep only the 2 most recently used files
+        if len(loaded_files) > 2:
+            oldest_file = next(iter(loaded_files))
+            del loaded_files[oldest_file]
+            torch.cuda.empty_cache()
+
+    # Update model index - remove all _scale_inv entries
+    print("Updating model index...")
+    new_weight_map = {}
+    for weight_name, file_name in weight_map.items():
+        if not weight_name.endswith("_scale_inv"):
+            new_weight_map[weight_name] = file_name
+
+    new_model_index = {
+        "metadata": model_index.get("metadata", {}),
+        "weight_map": new_weight_map
+    }
+
+    new_model_index_file = os.path.join(bf16_path, "model.safetensors.index.json")
+    with open(new_model_index_file, "w") as f:
+        json.dump(new_model_index, f, indent=2)
+
+    print(f"Conversion complete! Converted {len(fp8_weight_names)} FP8 weights to BF16")
+    print(f"Output saved to: {bf16_path}")
+
+
+if __name__ == "__main__":
+    parser = ArgumentParser(description="Convert MiniMax-M2 from FP8 to BF16")
+    parser.add_argument("--input-fp8-hf-path", type=str, required=True,
+                        help="Path to the FP8 model directory")
+    parser.add_argument("--output-bf16-hf-path", type=str, required=True,
+                        help="Path to save the BF16 model")
+    args = parser.parse_args()
+
+    main(args.input_fp8_hf_path, args.output_bf16_hf_path)

model-00001-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6fd947e3a34495c361fa273f04cd1ddcdcd652dfd97cb1c830777ee0f7192e17
+size 5000235232

model-00002-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cf2068a3e58418d0c28a49ae0277bcf80bbe48fe7b58c2bbbe288b8578f198f3
+size 4999565832

model-00003-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5192599924c00ca7cc0b7eeba473001ec87662297f082245de350fa56f863dc2
+size 5000140792

model-00004-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7ad9284db7d17eeaf648b1b24e8f76574b12ba0c9b634d64345d969cc692c14
+size 5000141504

model-00005-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03b08a5bd89174904b6b264f35289cd78eba8c2e447481188023b51cb55a6847
+size 4999568344

model-00006-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1b09414abf5089823fd17c7bdac530b2b953398e3ff8893f98f05e60cdebc16
+size 5000146376

model-00007-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ca955c56a35c0474cf135607f5e1ccc986dc5fc9d922ec24a5c2a80734bf225
+size 5000146968

model-00008-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3eea2dd55548cc9e9a54c513c9a7a14a72d973a8e73dc75472d43176a2432292
+size 4999570984

model-00009-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d3849ce508301c78a4b9df9b553a25c5421901f0be35d03b6520405a76aacc12
+size 5000146488

model-00010-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5461b48f80774f438967b0fbe614aaaa757e8c1312386af5f8b5b510ea9e17ec
+size 5000146968

model-00011-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc655de5faaa6c6ca3b58fa72cc156911ede65e2abd3e18b6d7b51b47f579f33
+size 4999570864

model-00012-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:58decbdba998a995cdfe4c8aea8f4d8b2936f49e50b9964a14042eee250d5b73
+size 5000146600

model-00013-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cecce84d9d57d2e269832b4d9e80a9e855e28061dd009748a331fe60b18f5373
+size 5000146968

model-00014-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:686ba28eebe91f8e0e15d8a3940bc47dce4a9008cad4893d4a79f7da94404d7f
+size 4999570752

model-00015-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a751935c3135867ca8b7d1a2ccc10f788ff3e23c36c3e264fa98c22f9f1bf290
+size 5000146712

model-00016-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5c7b10931032f3fd098e0278f1e0ce9ee6e064e1ba183ba0d1dd56f5509b52e7
+size 4981577024

model-00017-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:19b22fddf120bd654b496aed01fedacbbea07d55a877fcd830fe9011b5875f6f
+size 4999558272

model-00018-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6c24b4f46ca4910889fc10809e81202581036e68090891e723830d4526b3dc15
+size 5000146808

model-00019-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:deed0a529777b09ff6b2a04ec53d9afea5ad0b54fa67eccd829dd28e87258127
+size 4999571464

model-00020-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6b0ab9cefc9b4671f78713d9966e8ab509037a5a0df4c13f8a8f4eefae5d35e
+size 5000146048

model-00021-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:03dd85c6ad514ad8b65a844f67d54275031c444984879593ef71f268a5c3703c
+size 5000146920

model-00022-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a19326e82ca0b741e37e75e41a022f4628fd003d5acc7a3840001ec84f2dc453
+size 4999571272

model-00023-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d5dc2b75fd3a288a43a6bafd712d6eaad0c54f00feb669e902739ac752312fa5
+size 5000146192

model-00024-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:11e49d2957c2784937c166bd55e6baf60002527d1a5236e48102f7ce8185922b
+size 5000146968

model-00025-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:763a7e002a0a8ed2316d73ba3ea744e83784ce7bdf3fb25b0983845770f914f0
+size 4999571160

model-00026-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:28a16be45fdcb11e636b5741d995b8804980c392079537bbfedd7cd34b65ddcc
+size 5000146312

model-00027-of-00027.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0c12ec8e012ef4d311a0fdd81d51896432a01752c51346a945fd50edd39c2593
+size 4440722160

model.safetensors.index.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf9222d462bd87f371ed549bbb43ead4f6d4cbd8ab53f95d5b089818e4b91c9e
+size 14872848

modeling_minimax_m2.py

@@ -0,0 +1,725 @@
+#                🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨🚨
+#           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
+
+def _sanitize_config(config):
+    """
+    Helper function to sanitize the configuration object.
+    Specifically, it converts Enum values in `quantization_config` to their string values.
+    This prevents SyntaxErrors during torch.fx tracing where repr(Enum) (<Enum.Val: 'val'>)
+    is emitted into the generated code.
+    """
+    print("Config sanitized in modeling_minimax_m2.py")
+    q_config = getattr(config, "quantization_config", None)
+    if isinstance(q_config, dict):
+        for k, v in list(q_config.items()):
+            # Check for Enum by looking for 'value' attr
+            if hasattr(v, "value"):
+                q_config[k] = v.value
+                print(f"  Enum: q_config[{k}] = {v.value}")
+
+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)
+        _sanitize_config(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)
+        _sanitize_config(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",
+]

recipe.yaml

@@ -0,0 +1,31 @@
+default_stage:
+  default_modifiers:
+    AWQModifier:
+      config_groups:
+        mlp_experts_projections:
+          targets: ['re:.*block_sparse_moe\.experts\.\d+\.(w1|w2|w3)$']
+          weights:
+            num_bits: 4
+            type: int
+            symmetric: true
+            group_size: 32
+            strategy: group
+            block_structure: null
+            dynamic: false
+            actorder: null
+            scale_dtype: null
+            zp_dtype: null
+            observer: memoryless_minmax
+            observer_kwargs: {}
+          input_activations: null
+          output_activations: null
+          format: null
+      targets: [Linear]
+      ignore: []
+      mappings:
+      - smooth_layer: re:.*post_attention_layernorm$
+        balance_layers: ['re:.*w1$', 're:.*w3$']
+      - smooth_layer: re:.*w3$
+        balance_layers: ['re:.*w2$']
+      duo_scaling: true
+      n_grid: 20

special_tokens_map.json

@@ -0,0 +1,75 @@
+{
+  "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>"
+  ],
+  "bos_token": {
+    "content": "]~!b[",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "[e~[",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "]!d~[",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e7b90ed7f55d905175bc26771d6d7d33b40b46742f073675bc816fedaf482ea1
+size 15522763

tokenizer_config.json

@@ -0,0 +1,496 @@
+{
+  "add_prefix_space": false,
+  "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>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": 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>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "200044": {
+      "content": "<file_content>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "200045": {
+      "content": "<source_files>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "200046": {
+      "content": "<pr_start>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "200047": {
+      "content": "<review_comment>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "200048": {
+      "content": "<filepath>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "200049": {
+      "content": "<file_sep>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "200050": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "200051": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "200052": {
+      "content": "<minimax:tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "200053": {
+      "content": "</minimax:tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": 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>"
+  ],
+  "bos_token": "]~!b[",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "[e~[",
+  "extra_special_tokens": {},
+  "model_max_length": 40960000,
+  "tokenizer_class": "GPT2Tokenizer",
+  "unk_token": "]!d~["
+}