Duplicate from unsloth/Kimi-K3

Co-authored-by: Daniel (Unsloth) <danielhanchen@users.noreply.huggingface.co>

.eval_results/apex-agents.yaml

@@ -0,0 +1,7 @@
+- dataset:
+    id: mercor/apex-agents
+    task_id: apex-agents
+  value: 41
+  source:
+    url: https://huggingface.co/moonshotai/Kimi-K3
+    name: Model Card

.eval_results/deep-swe.yaml

@@ -0,0 +1,7 @@
+- dataset:
+    id: datacurve/deep-swe
+    task_id: deep_swe
+  value: 67.5
+  source:
+    url: https://huggingface.co/moonshotai/Kimi-K3
+    name: Model Card

.eval_results/gpqa.yaml

@@ -0,0 +1,7 @@
+- dataset:
+    id: Idavidrein/gpqa
+    task_id: diamond
+  value: 93.5
+  source:
+    url: https://huggingface.co/moonshotai/Kimi-K3
+    name: Model Card

.eval_results/hle.yaml

@@ -0,0 +1,7 @@
+- dataset:
+    id: cais/hle
+    task_id: hle
+  value: 56
+  source:
+    url: https://huggingface.co/moonshotai/Kimi-K3
+    name: Model Card

.gitattributes

@@ -0,0 +1,37 @@
+*.7z filter=lfs diff=lfs merge=lfs -text
+*.arrow filter=lfs diff=lfs merge=lfs -text
+*.bin filter=lfs diff=lfs merge=lfs -text
+*.bz2 filter=lfs diff=lfs merge=lfs -text
+*.ckpt filter=lfs diff=lfs merge=lfs -text
+*.ftz filter=lfs diff=lfs merge=lfs -text
+*.gz filter=lfs diff=lfs merge=lfs -text
+*.h5 filter=lfs diff=lfs merge=lfs -text
+*.joblib filter=lfs diff=lfs merge=lfs -text
+*.lfs.* filter=lfs diff=lfs merge=lfs -text
+*.mlmodel filter=lfs diff=lfs merge=lfs -text
+*.model filter=lfs diff=lfs merge=lfs -text
+*.msgpack filter=lfs diff=lfs merge=lfs -text
+*.npy filter=lfs diff=lfs merge=lfs -text
+*.npz filter=lfs diff=lfs merge=lfs -text
+*.onnx filter=lfs diff=lfs merge=lfs -text
+*.ot filter=lfs diff=lfs merge=lfs -text
+*.parquet filter=lfs diff=lfs merge=lfs -text
+*.pb filter=lfs diff=lfs merge=lfs -text
+*.pickle filter=lfs diff=lfs merge=lfs -text
+*.pkl filter=lfs diff=lfs merge=lfs -text
+*.pt filter=lfs diff=lfs merge=lfs -text
+*.pth filter=lfs diff=lfs merge=lfs -text
+*.rar filter=lfs diff=lfs merge=lfs -text
+*.safetensors filter=lfs diff=lfs merge=lfs -text
+saved_model/**/* filter=lfs diff=lfs merge=lfs -text
+*.tar.* filter=lfs diff=lfs merge=lfs -text
+*.tar filter=lfs diff=lfs merge=lfs -text
+*.tflite filter=lfs diff=lfs merge=lfs -text
+*.tgz filter=lfs diff=lfs merge=lfs -text
+*.wasm filter=lfs diff=lfs merge=lfs -text
+*.xz 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
+assets/k3-hero.jpg filter=lfs diff=lfs merge=lfs -text

LICENSE

@@ -0,0 +1,52 @@
+Kimi K3 License
+
+Copyright (c) 2026 Moonshot AI
+
+Permission is hereby granted, free of charge, to any person (the "Licensee")
+obtaining a copy of this software — including the model weights, parameters,
+configuration files, inference and training code, and associated documentation
+(collectively, the "Software") — to deal in the Software without restriction.
+This includes, without limitation, the rights to use, copy, modify, merge,
+publish, distribute, sublicense, and/or sell copies of the Software; to run,
+deploy, fine-tune, or otherwise modify the Software and create derivative works
+from it; and to permit persons to whom the Software is furnished to do so, in
+each case subject to the following conditions:
+
+1. The above copyright notice and this permission notice shall be included in
+all copies or substantial portions of the Software. Licensee's use of the
+Software must comply with applicable laws and regulations.
+
+2. "Model as a Service" means giving a third party access to language model
+inference or fine-tuning (e.g., via API) in a manner that allows such third
+party to exercise meaningful control over the inputs, parameters, or training
+data. This does not include (a) end-user products with model capabilities solely
+embedded within specific features or harnesses, or (b) mere relaying of requests
+to models hosted by others.
+
+If the Licensee or any of its affiliates operates a Model as a Service business,
+and the aggregate revenue of the Licensee and its affiliates exceeds 20 million
+US dollars (or the equivalent in other currencies) in total over any consecutive
+12 months, the Licensee must enter into a separate agreement with Moonshot AI
+before using the Software or its derivative works for any commercial purpose. 
+
+3. If the Software (or any derivative works thereof) is used for any of the
+Licensee's commercial products or services that have more than 100 million
+monthly active users, or more than 20 million US dollars (or equivalent in other
+currencies) in monthly revenue, "Kimi K3" must be prominently displayed on the
+user interface of such product or service.
+
+4. The requirements set forth in Sections 2 and 3 do not apply to: (a) internal
+use of the Software, defined as any use that does not make the Software, its
+outputs, or its underlying capabilities available to third parties; or (b) any
+use of the Software accessed through Moonshot AI's official products or
+certified inference partners.
+
+5. THE SOFTWARE AND ANY OUTPUT AND RESULTS THEREFROM ARE PROVIDED ON AN “AS IS”
+BASIS, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT
+LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE
+AND NONINFRINGEMENT. IN NO EVENT SHALL MOONSHOT AI OR ITS AFFILIATES OR
+COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER
+IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+For any questions regarding this license, please contact <license@moonshot.ai>.

README.md

@@ -0,0 +1,700 @@
+---
+base_model:
+- moonshotai/Kimi-K3
+tags:
+- compressed-tensors
+- unsloth
+- conversational
+license: other
+license_name: "kimi-k3"
+library_name: transformers
+pipeline_tag: image-text-to-text
+---
+<div>
+<p style="margin-top: 0;margin-bottom: 0;">
+    <em><a href="https://docs.unsloth.ai/basics/unsloth-dynamic-v2.0-gguf">Unsloth Dynamic 2.0</a> achieves superior accuracy & outperforms other leading quants.</em>
+  </p>
+  <div style="display: flex; gap: 5px; align-items: center; ">
+    <a href="https://github.com/unslothai/unsloth/">
+      <img src="https://github.com/unslothai/unsloth/raw/main/images/unsloth%20new%20logo.png" width="133">
+    </a>
+    <a href="https://discord.gg/unsloth">
+      <img src="https://github.com/unslothai/unsloth/raw/main/images/Discord%20button.png" width="173">
+    </a>
+    <a href="https://docs.unsloth.ai/">
+      <img src="https://raw.githubusercontent.com/unslothai/unsloth/refs/heads/main/images/documentation%20green%20button.png" width="143">
+    </a>
+  </div>
+</div>
+
+<div align="center">
+  <picture>
+      <img src="assets/kimi-logo.png" width="30%" alt="Kimi K3">
+  </picture>
+</div>
+<hr>
+<div align="center" style="line-height:1">
+  <a href="https://www.kimi.com" target="_blank"><img alt="Chat" src="https://img.shields.io/badge/🤖%20Chat-Kimi%20K3-ff6b6b?color=1783ff&logoColor=white"/></a>
+  <a href="https://www.moonshot.ai" target="_blank"><img alt="Homepage" src="https://img.shields.io/badge/Homepage-Moonshot%20AI-white?logo=Kimi&logoColor=white"/></a>
+</div>
+
+<div align="center" style="line-height: 1;">
+  <a href="https://huggingface.co/moonshotai" target="_blank"><img alt="Hugging Face" src="https://img.shields.io/badge/%F0%9F%A4%97%20Hugging%20Face-Moonshot%20AI-ffc107?color=ffc107&logoColor=white"/></a>
+  <a href="https://twitter.com/kimi_moonshot" target="_blank"><img alt="Twitter Follow" src="https://img.shields.io/badge/Twitter-Kimi.ai-white?logo=x&logoColor=white"/></a>
+  <a href="https://discord.gg/TYU2fdJykW" target="_blank"><img alt="Discord" src="https://img.shields.io/badge/Discord-Kimi.ai-white?logo=discord&logoColor=white"/></a>
+  <a href="https://modelscope.cn/organization/moonshotai" target="_blank"><img alt="ModelScope" src="https://img.shields.io/badge/ModelScope-Moonshot%20AI-white?labelColor=rgb(99%2C%2074%2C%255)"/></a>
+</div>
+<div align="center" style="line-height: 1;">
+  <a href="https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/badge/License-Kimi_K3-f5de53?&color=f5de53"/></a>
+</div>
+
+
+<p align="center">
+📰&nbsp;&nbsp;<a href="https://www.kimi.com/blog/kimi-k3">Tech Blog</a> | &nbsp;&nbsp;&nbsp; <b>📄&nbsp;&nbsp;<a href="https://github.com/MoonshotAI/Kimi-K3/blob/main/k3_tech_report.pdf">Full Report</a></b>
+</p>
+
+
+
+## 1. Model Introduction
+
+Kimi K3 is an open-weight, native multimodal agentic model and our most capable model to date. It is a 2.8T-parameter model built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), with native vision capabilities and a 1-million-token context window. It is the world's first open 3T-class model, designed for frontier intelligence across long-horizon coding, knowledge work, and reasoning.
+
+### Key Features
+- **New Architecture**: Kimi K3 is built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), and scales up MoE sparsity with a Stable LatentMoE framework that activates 16 out of 896 experts — yielding an approximate 2.5× improvement in overall scaling efficiency over Kimi K2.
+- **Long-Horizon Coding**: Operating with minimal human oversight, Kimi K3 sustains long engineering sessions, navigates massive repositories, and orchestrates terminal tools — from GPU kernel optimization and compiler development to vision-in-the-loop game dev, CAD, and even chip design.
+- **Agentic Knowledge Work**: Kimi K3 advances end-to-end knowledge work, producing deep research with interactive visualizations, widgets and dashboards, and motion design and video editing, powered by its native multimodal architecture.
+- **Native Multimodality & Long Context**: Kimi K3 understands text, images, and video within the same model, and supports a 1-million-token context window.
+- **Open Frontier Weights**: We release the full Kimi K3 model weights under the Kimi K3 License, making frontier intelligence openly available for research, deployment, and further innovation.
+## 2. Model Summary
+
+<div align="center">
+<table>
+<tbody>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Architecture</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">Mixture-of-Experts (MoE)</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Total Parameters</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">2.8T</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Activated Parameters</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">104B</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Number of Layers</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">93</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Number of Dense Layers</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">1</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Attention-Layer Composition</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">69 KDA + 24 Gated MLA</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Attention Hidden Dimension</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">7168</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Number of Attention Heads</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">96</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Latent MoE Dimension</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">3584</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>MoE Hidden Dimension</strong> (per Expert)</td>
+<td align="center" style="vertical-align: middle; text-align: center">3072</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Number of Experts</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">896</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Selected Experts per Token</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">16</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Number of Shared Experts</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">2</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Vocabulary Size</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">160K</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Context Length</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">1048576</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Attention Mechanism</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">KDA &amp; Gated MLA</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Activation Function</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">SiTU-GLU</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Vision Encoder</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">MoonViT-V2</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Parameters of Vision Encoder</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">401M</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Quantization</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">MXFP4 weights / MXFP8 activations<br>(quantization-aware training)</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center"><strong>Modality</strong></td>
+<td align="center" style="vertical-align: middle; text-align: center">Text, Image</td>
+</tr>
+</tbody>
+</table>
+</div>
+
+
+## 3. Evaluation Results
+
+<div align="center">
+<table>
+<thead>
+<tr>
+<th align="center" style="text-align: center">Benchmark</th>
+<th align="center" style="text-align: center"><sup>Kimi K3<br><sup>(max)</sup></sup></th>
+<th align="center" style="text-align: center"><sup>Claude Fable 5<br><sup>(max, w/ fallback)</sup></sup></th>
+<th align="center" style="text-align: center"><sup>GPT-5.6 Sol<br><sup>(max)</sup></sup></th>
+<th align="center" style="text-align: center"><sup>Claude Opus 4.8<br><sup>(max)</sup></sup></th>
+<th align="center" style="text-align: center"><sup>GPT-5.5<br><sup>(xhigh)</sup></sup></th>
+<th align="center" style="text-align: center"><sup>GLM-5.2<br><sup>(max)</sup></sup></th>
+</tr>
+</thead>
+<tbody>
+<tr>
+<td align="center" colspan=7 style="text-align: center"><strong>Reasoning &amp; Knowledge</strong></td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">GPQA Diamond</td>
+<td align="center" style="vertical-align: middle; text-align: center">93.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">92.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">94.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">91.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">93.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">91.2</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">CritPt</td>
+<td align="center" style="vertical-align: middle; text-align: center">23.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">28.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">32.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">20.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">27.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">20.9</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">AA-LCR</td>
+<td align="center" style="vertical-align: middle; text-align: center">74.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">70.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">73.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">67.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">74.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">71.3</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">HLE-Full</td>
+<td align="center" style="vertical-align: middle; text-align: center">43.5 / 56.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">53.3 / 63.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">44.5 / 58.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">49.8 / 57.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">41.4 / 52.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" colspan=7 style="text-align: center"><strong>Coding</strong></td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">DeepSWE</td>
+<td align="center" style="vertical-align: middle; text-align: center">67.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">70.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">73.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">59.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">67.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">46.2</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">ProgramBench</td>
+<td align="center" style="vertical-align: middle; text-align: center">77.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">76.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">77.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">71.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">70.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">63.7</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">Terminal-Bench 2.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">88.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">88.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">88.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">84.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">83.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">82.7</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">FrontierSWE</td>
+<td align="center" style="vertical-align: middle; text-align: center">81.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">86.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">71.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">66.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">64.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">67.3</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">SWE-Marathon</td>
+<td align="center" style="vertical-align: middle; text-align: center">42.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">35.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">39.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">40.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">14.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">13.0</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">PostTrainBench</td>
+<td align="center" style="vertical-align: middle; text-align: center">36.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">41.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">34.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">34.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">28.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">34.3</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">MLS-Bench-Lite</td>
+<td align="center" style="vertical-align: middle; text-align: center">48.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">49.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">46.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">42.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">35.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">40.4</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">SciCode</td>
+<td align="center" style="vertical-align: middle; text-align: center">58.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">60.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">56.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">53.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">56.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">50.5</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">Kimi Code Bench 2.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">72.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">76.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">64.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">71.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">69.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">64.2</td>
+</tr>
+<tr>
+<td align="center" colspan=7 style="text-align: center"><strong>Agentic</strong></td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">BrowseComp</td>
+<td align="center" style="vertical-align: middle; text-align: center">91.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">88.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">90.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">84.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">84.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">DeepSearchQA (F1)</td>
+<td align="center" style="vertical-align: middle; text-align: center">95.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">94.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+<td align="center" style="vertical-align: middle; text-align: center">93.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">ResearchRubrics</td>
+<td align="center" style="vertical-align: middle; text-align: center">76.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+<td align="center" style="vertical-align: middle; text-align: center">73.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">73.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">64.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">71.1</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">GDPval-AA v2 (Elo)</td>
+<td align="center" style="vertical-align: middle; text-align: center">1686</td>
+<td align="center" style="vertical-align: middle; text-align: center">1747</td>
+<td align="center" style="vertical-align: middle; text-align: center">1736</td>
+<td align="center" style="vertical-align: middle; text-align: center">1593</td>
+<td align="center" style="vertical-align: middle; text-align: center">1491</td>
+<td align="center" style="vertical-align: middle; text-align: center">1510</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">Toolathlon-Verified</td>
+<td align="center" style="vertical-align: middle; text-align: center">76.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">77.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">74.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">76.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">73.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">59.9</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">MCPMark-Verified</td>
+<td align="center" style="vertical-align: middle; text-align: center">94.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">87.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">92.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">76.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">92.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">MCP-Atlas</td>
+<td align="center" style="vertical-align: middle; text-align: center">84.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">84.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">83.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">83.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">82.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">82.6</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">AutomationBench</td>
+<td align="center" style="vertical-align: middle; text-align: center">30.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">29.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">29.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">27.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">22.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">12.9</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">JobBench</td>
+<td align="center" style="vertical-align: middle; text-align: center">54.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">57.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">45.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">48.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">38.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">43.4</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">AA-Briefcase (Elo)</td>
+<td align="center" style="vertical-align: middle; text-align: center">1548</td>
+<td align="center" style="vertical-align: middle; text-align: center">1583</td>
+<td align="center" style="vertical-align: middle; text-align: center">1495</td>
+<td align="center" style="vertical-align: middle; text-align: center">1354</td>
+<td align="center" style="vertical-align: middle; text-align: center">1158</td>
+<td align="center" style="vertical-align: middle; text-align: center">1260</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">Agents' Last Exam</td>
+<td align="center" style="vertical-align: middle; text-align: center">28.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">25.7<sup>†</sup></td>
+<td align="center" style="vertical-align: middle; text-align: center">29.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">27.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">26.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">20.4</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">APEX-Agents</td>
+<td align="center" style="vertical-align: middle; text-align: center">41.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">43.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">39.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">39.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">38.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">35.6</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">OfficeQA Pro</td>
+<td align="center" style="vertical-align: middle; text-align: center">63.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">69.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">63.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">63.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">60.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">41.4</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">SpreadsheetBench 2</td>
+<td align="center" style="vertical-align: middle; text-align: center">34.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">34.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">32.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">31.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">29.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">28.1</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">OSWorld-Verified</td>
+<td align="center" style="vertical-align: middle; text-align: center">84.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">85.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">83.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">83.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">79.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">OSWorld 2.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">58.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">66.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">62.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">55.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">49.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">SaaS-Bench</td>
+<td align="center" style="vertical-align: middle; text-align: center">60.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+<td align="center" style="vertical-align: middle; text-align: center">61.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">56.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">43.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">τ³-Banking</td>
+<td align="center" style="vertical-align: middle; text-align: center">33.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">26.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">33.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">27.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">31.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">26.8</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">Harvey Lab-AA</td>
+<td align="center" style="vertical-align: middle; text-align: center">94.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">93.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">87.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">91.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">86.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">91.0</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">CorpFin v2</td>
+<td align="center" style="vertical-align: middle; text-align: center">71.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">71.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">64.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">66.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">68.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">66.1</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">Finance Agent v2</td>
+<td align="center" style="vertical-align: middle; text-align: center">54.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">56.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">53.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">53.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">51.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">49.7</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">Legal Research Bench</td>
+<td align="center" style="vertical-align: middle; text-align: center">44.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">49.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">48.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">43.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">40.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">31.3</td>
+</tr>
+<tr>
+<td align="center" colspan=7 style="text-align: center"><strong>Vision</strong></td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">WorldVQA ForceAnswer</td>
+<td align="center" style="vertical-align: middle; text-align: center">51.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">56.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">41.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">39.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">38.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">OmniDocBench</td>
+<td align="center" style="vertical-align: middle; text-align: center">91.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">89.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">85.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">87.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">89.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">PerceptionBench</td>
+<td align="center" style="vertical-align: middle; text-align: center">58.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">57.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">59.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">47.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">55.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">Video-MME (w. sub)</td>
+<td align="center" style="vertical-align: middle; text-align: center">90.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+<td align="center" style="vertical-align: middle; text-align: center">89.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">86.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">89.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">MMVU</td>
+<td align="center" style="vertical-align: middle; text-align: center">82.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+<td align="center" style="vertical-align: middle; text-align: center">81.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">79.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">81.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">BabyVision w/ python</td>
+<td align="center" style="vertical-align: middle; text-align: center">85.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">90.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">88.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">81.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">83.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">MMMU-Pro</td>
+<td align="center" style="vertical-align: middle; text-align: center">81.6 / 83.4</td>
+<td align="center" style="vertical-align: middle; text-align: center">81.2 / 86.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">83.0 / 84.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">78.9 / 82.7</td>
+<td align="center" style="vertical-align: middle; text-align: center">81.2 / 83.2</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">CharXiv (RQ)</td>
+<td align="center" style="vertical-align: middle; text-align: center">84.8 / 91.3</td>
+<td align="center" style="vertical-align: middle; text-align: center">88.9 / 93.5</td>
+<td align="center" style="vertical-align: middle; text-align: center">84.6 / 89.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">80.5 / 89.9</td>
+<td align="center" style="vertical-align: middle; text-align: center">84.1 / 89.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">MathVision</td>
+<td align="center" style="vertical-align: middle; text-align: center">94.3 / 97.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">94.8 / 98.6</td>
+<td align="center" style="vertical-align: middle; text-align: center">95.8 / 97.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">86.7 / 97.1</td>
+<td align="center" style="vertical-align: middle; text-align: center">92.2 / 96.8</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+<tr>
+<td align="center" style="vertical-align: middle; text-align: center">ZeroBench (pass@5)</td>
+<td align="center" style="vertical-align: middle; text-align: center">23.0 / 41.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">23.0 / 46.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">17.0 / 35.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">17.0 / 34.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">22.0 / 41.0</td>
+<td align="center" style="vertical-align: middle; text-align: center">—</td>
+</tr>
+</tbody>
+</table>
+</div>
+
+<details>
+<summary><b>Footnotes</b></summary>
+
+All Kimi K3 results are obtained with reasoning effort set to 'max' and temperature = 1.0. For single-step tasks, such as GPQA Diamond, HLE-Full, and vision benchmarks without tools, we set top-p = 0.95; for agentic tasks, we set top-p = 1.0. For HLE-Full, MMMU-Pro, CharXiv (RQ), MathVision, and ZeroBench, each cell reports the scores without and with tool augmentation (general tools for HLE-Full, Python for the vision benchmarks), in that order.
+
+1. **Reasoning & knowledge benchmarks**
+   - **CritPt and AA-LCR.** Scores are cited from [Artificial Analysis](https://artificialanalysis.ai/) as of July 23, 2026.
+2. **Coding benchmarks**
+   - **DeepSWE.** Kimi K3 is evaluated with the Kimi Code harness. The GLM-5.2 score is taken from the [GLM-5.2 release blog](https://z.ai/blog/glm-5.2); all remaining scores are from the official [DeepSWE leaderboard](https://deepswe.datacurve.ai/), under which Kimi K3 attains 67.3 with the mini-SWE-agent harness. We report the DeepSWE v1.1 tasks.
+   - **Terminal-Bench 2.1.** Kimi K3 is evaluated with the Kimi Code harness. For all other models, we report the best score across harnesses: GLM-5.2 with Claude Code ([GLM-5.2 release blog](https://z.ai/blog/glm-5.2)); Claude Opus 4.8 and Claude Fable 5 with Terminus 2 ([Artificial Analysis](https://artificialanalysis.ai/evaluations/terminalbench-v2-1)); GPT-5.5 and GPT-5.6 Sol with Codex ([OpenAI](https://openai.com/index/previewing-gpt-5-6-sol/)).
+   - **ProgramBench.** Kimi K3 is evaluated with the Kimi Code harness. The GLM-5.2 score is from the [GLM-5.2 release blog](https://z.ai/blog/glm-5.2); all other scores are from [Vals AI](https://www.vals.ai/benchmarks/programbench).
+   - **SWE-Marathon.** Kimi K3, Claude Opus 4.8, and Claude Fable 5 are evaluated with the Claude Code harness; GPT-5.6 Sol is evaluated with the Codex harness. The GLM-5.2 score is from the [GLM-5.2 release blog](https://z.ai/blog/glm-5.2). Our evaluation is based on an H20-calibrated branch of the [official tasks](https://www.swe-marathon.org/) as of July 9, 2026, prior to the final v1.1 release: the Docker images, performance gates, and reference oracles for the GPU tasks have been recalibrated for H20, while the correctness and anti-cheat validators remain unchanged. Additionally, Claude Fable 5 hit fallbacks on 35% of the tasks in our evaluation, which may have negatively impacted its measured performance.
+   - **FrontierSWE.** Kimi K3 is evaluated with the Kimi Code harness and GPT-5.6 Sol with the Codex harness; all other results are from [FrontierSWE](https://www.frontierswe.com/). Dominance scores are recomputed from the raw scores using the official evaluation script and are current as of July 16, 2026.
+   - **PostTrainBench.** Scores for GLM-5.2, GPT-5.5, and Claude Opus 4.8 are adopted from the official [PostTrainBench](https://posttrainbench.com/) results. Kimi K3, Claude Fable 5, and GPT-5.6 Sol are evaluated with the official Harbor implementation at maximum reasoning effort, averaged over three runs on H20 GPUs (instead of H100 in the official setting) — Kimi K3 and Claude Fable 5 with the Claude Code harness, and GPT-5.6 Sol with the Codex harness.
+   - **MLS-Bench-Lite.** Kimi K3 is evaluated with the Kimi Code harness; GLM-5.2 and the Claude models with the Claude Code harness; GPT-5.5 and GPT-5.6 Sol with the Codex harness.
+   - **SciCode.** Scores are cited from [Artificial Analysis](https://artificialanalysis.ai/) as of July 23, 2026.
+   - **Kimi Code Bench 2.0 (in-house).** Kimi K3 is evaluated with the Kimi Code harness (it attains 73.7 with the Claude Code harness); GLM-5.2, Claude Opus 4.8, and Claude Fable 5 with the Claude Code harness; GPT-5.5 and GPT-5.6 Sol with the Codex harness. All models are evaluated at maximum reasoning effort, except GPT-5.5, which uses the "xhigh" setting. As the benchmark includes cybersecurity and safety-related tasks, we also disclose the fraction of refused or fallback tasks: Claude Fable 5 hit 13 fallbacks and 1 refusal out of 80 tasks; 10 refusals out of 80 tasks entered GPT-5.6 Sol's cyber guard; GPT-5.5 had 3 refusals out of 80 tasks.
+3. **Agentic benchmarks**
+   - **OfficeQA Pro.** Each test case provides the agent with the entire PDF corpus, with all PDFs rendered as images and no machine-readable text available.
+   - **OfficeQA Pro and SpreadsheetBench 2.** Kimi K3, GLM-5.2, Claude Opus 4.8, and Claude Fable 5 are evaluated with the Claude Code harness; GPT-5.5 and GPT-5.6 Sol are evaluated with the Codex harness.
+   - **MCP-Atlas.** All models are evaluated on the 500-task public subset with a 100-turn limit, using Gemini 3.1 Pro as the judge.
+   - **AutomationBench.** All models are evaluated on the 600-task public subset, following the official GitHub setup in all other respects.
+   - **BrowseComp.** We adopt a context-compaction strategy triggered at 300K tokens. When evaluated with the full 1M-token context window and no context management, Kimi K3 achieves a score of 90.4. The results of Claude Fable 5, Claude Opus 4.8, GPT-5.6 Sol, and GPT-5.5 are cited from [Anthropic](https://www.anthropic.com/news/claude-fable-5-mythos-5) and [OpenAI](https://openai.com/index/gpt-5-6/).
+   - **GDPval-AA v2, AA-Briefcase, τ³-Banking, Harvey Lab-AA, and APEX-Agents.** Scores are cited from [Artificial Analysis](https://artificialanalysis.ai/) and the [APEX-Agents leaderboard](https://www.mercor.com/apex/apex-agents-leaderboard/) as of July 23, 2026. For Harvey Lab-AA, we report the criterion pass rate.
+   - **CorpFin v2, Finance Agent v2, and Legal Research Bench.** Scores are cited from [Vals AI](https://www.vals.ai/).
+   - **Agents' Last Exam.** Scores are cited from the [official leaderboard](https://agents-last-exam.org/leaderboard) as of July 23, 2026; we report the leaderboard's primary pass-rate metric. On the leaderboard, each model is paired with a specific harness: Kimi K3 with Kimi Code; GPT-5.6 Sol and GPT-5.5 with Codex; Claude Fable 5, Claude Opus 4.8, and GLM-5.2 with Claude Code. <sup>†</sup> The Claude Fable 5 entry runs at xhigh effort with 40% of tasks annotated as downgraded.
+4. **Multimodal benchmarks**
+   - Except for ZeroBench, which follows the official setting and is run five times, all multimodal scores are averaged over three runs. MMMU-Pro is evaluated following the official protocol, preserving the original input order and prepending images to the text input.
+   - **PerceptionBench** is an in-house benchmark that focuses on atomic visual perception capabilities.
+
+</details>
+
+## 4. Native MXFP4 Quantization
+
+Kimi K3 applies quantization-aware training from the SFT stage onward, using MXFP4 weights with MXFP8 activations for broad hardware compatibility.
+
+## 5. Deployment
+
+> [!Note]
+> You can access Kimi K3's API on https://platform.kimi.ai by selecting `kimi-k3`, and we provide OpenAI/Anthropic-compatible API for you. Currently, Kimi K3 is recommended to run on the following inference engines:
+
+- [vLLM](https://github.com/vllm-project/vllm) — see [recipes](https://recipes.vllm.ai/moonshotai/Kimi-K3)
+- [SGLang](https://github.com/sgl-project/sglang) — see [cookbook](https://docs.sglang.io/cookbook/autoregressive/Moonshotai/Kimi-K3)
+- [TokenSpeed](https://github.com/lightseekorg/tokenspeed) — see [recipes](https://lightseek.org/tokenspeed/recipes/models#kimi-k3)
+
+---
+## 6. Model Usage
+
+Kimi K3 always has thinking enabled, and will return `reasoning_content`. Thinking effort is configured with the top-level `reasoning_effort` request field, which supports `"low"`, `"high"`, and `"max"` (default `"max"`).
+
+Kimi K3 was trained in the preserved thinking history mode. For multi-turn conversations and tool calls, Kimi K3 requires the complete assistant message returned by the API to be passed back to `messages` as-is — including `reasoning_content` and `tool_calls`, not just `content`:
+
+```python
+import openai
+
+def chat_with_preserved_thinking(client: openai.OpenAI, model_name: str):
+    messages = [
+        {
+            "role": "user",
+            "content": "Tell me three random numbers."
+        },
+        {
+            "role": "assistant",
+            "reasoning_content": "I'll start by listing five numbers: 473, 921, 235, 215, 222, and I'll tell you the first three.",
+            "content": "473, 921, 235"
+        },
+        {
+            "role": "user",
+            "content": "What are the other two numbers you have in mind?"
+        }
+    ]
+
+    response = client.chat.completions.create(
+        model=model_name,
+        messages=messages,
+        stream=False,
+        max_tokens=4096,
+        reasoning_effort="max",
+    )
+    # the assistant should mention 215 and 222 that appear in the prior reasoning content
+    print(f"response: {response.choices[0].message.reasoning}")
+    return response.choices[0].message.content
+```
+
+For full guides and examples (vision input, structured output, partial mode, tool choice, dynamic tool loading, context caching), see the [Kimi K3 Quickstart](https://platform.kimi.ai/docs/guide/kimi-k3-quickstart) and [Thinking Effort](https://platform.kimi.ai/docs/guide/use-thinking-effort).
+
+### Coding Agent Framework
+
+Kimi K3 works best with [Kimi Code CLI](https://www.kimi.com/code) as its agent framework. We warmly invite you to give it a try — run Kimi Code in your terminal and select Kimi K3 using the `/model` command. We hope you enjoy building with Kimi K3, and we would love to hear your feedback!
+
+
+---
+
+## 7. License
+
+Both the code repository and the model weights are released under the [Kimi K3 License](https://huggingface.co/moonshotai/Kimi-K3/blob/main/LICENSE).
+
+---
+
+## 8. Contact Us
+
+If you have any questions, please reach out at [support@moonshot.ai](mailto:support@moonshot.ai).

added_tokens.json

@@ -0,0 +1,9 @@
+{
+  "<|end_header_id|>": 163844,
+  "<|im_assistant|>": 163842,
+  "<|im_end|>": 163840,
+  "<|im_middle|>": 163846,
+  "<|im_system|>": 163845,
+  "<|im_user|>": 163841,
+  "<|start_header_id|>": 163843
+}

chat_template.jinja

@@ -0,0 +1,679 @@
+{#- Kimi-K3 template transcribed from the official parser -#}
+{%- set imgns = namespace(idx = 0, on = (image_prompts is defined and image_prompts is not none)) -%}
+
+{%- macro img_next() -%}
+{%- if not imgns.on -%}
+{{- '<|kimi_image_placeholder|>' -}}
+{%- elif imgns.idx >= (image_prompts | length) -%}
+{{- raise_exception('More image placeholders than image prompts.') -}}
+{%- else -%}
+{{- image_prompts[imgns.idx] -}}
+{%- set imgns.idx = imgns.idx + 1 -%}
+{%- endif -%}
+{%- endmacro -%}{%- macro xtext(t) -%}{%- if not imgns.on or '<|kimi_image_placeholder|>' not in (t | string) -%}
+{{- t -}}
+{%- else -%}
+{%- set parts = (t | string).split('<|kimi_image_placeholder|>') -%}
+{%- for part in parts -%}
+{{- part -}}
+{%- if not loop.last -%}{{- img_next() -}}{%- endif -%}
+{%- endfor -%}
+{%- endif -%}
+{%- endmacro -%}{%- macro attr(k, v) -%}
+{{- ' ' + k + '="' + (v | string | replace('&', '&amp;') | replace('"', '&quot;')) + '"' -}}
+{%- endmacro -%}
+
+{%- macro otag(tag, attrs=[]) -%}
+{{- '<|open|>' + tag -}}
+{%- for a in attrs -%}{{- attr(a[0], a[1]) -}}{%- endfor -%}
+{{- '<|sep|>' -}}
+{%- endmacro -%}
+
+{%- macro ctag(tag) -%}
+{{- '<|close|>' + tag + '<|sep|>' -}}
+{%- endmacro -%}{%- macro numstr(v) -%}
+{%- set s = v | tojson -%}
+{%- if v is float and '.' not in s and 'e' not in s and 'E' not in s
+       and 'n' not in s and 'N' not in s -%}
+{{- s ~ '.0' -}}
+{%- else -%}
+{{- s -}}
+{%- endif -%}
+{%- endmacro -%}{%- macro jkey(k) -%}
+{%- if k is string -%}
+{{- k | tojson -}}
+{%- elif k is boolean -%}
+{{- ('true' if k else 'false') | tojson -}}
+{%- elif k is none -%}
+{{- 'null' | tojson -}}
+{%- elif k is number -%}
+{{- numstr(k) | tojson -}}
+{%- else -%}
+{{- raise_exception('keys must be str, int, float, bool or None') -}}
+{%- endif -%}
+{%- endmacro -%}{%- macro jdump(v) -%}
+{%- if v is mapping -%}
+{{- '{' -}}
+{%- for k, val in v | dictsort(case_sensitive=true) -%}
+{%- if not loop.first -%}{{- ',' -}}{%- endif -%}
+{{- jkey(k) -}}{{- ':' -}}{{- jdump(val) -}}
+{%- endfor -%}
+{{- '}' -}}
+{%- elif v is string or v is boolean or v is none or v is number -%}
+{{- numstr(v) -}}
+{%- elif v is undefined -%}
+{{- 'null' -}}
+{%- else -%}
+{{- '[' -}}
+{%- for item in v -%}
+{%- if not loop.first -%}{{- ',' -}}{%- endif -%}
+{{- jdump(item) -}}
+{%- endfor -%}
+{{- ']' -}}
+{%- endif -%}
+{%- endmacro -%}{%- macro jdumpsp(v) -%}
+{%- if v is mapping -%}
+{{- '{' -}}
+{%- for k, val in v | items -%}
+{%- if not loop.first -%}{{- ', ' -}}{%- endif -%}
+{{- jkey(k) -}}{{- ': ' -}}{{- jdumpsp(val) -}}
+{%- endfor -%}
+{{- '}' -}}
+{%- elif v is string or v is boolean or v is none or v is number -%}
+{{- numstr(v) -}}
+{%- elif v is undefined -%}
+{{- 'null' -}}
+{%- else -%}
+{{- '[' -}}
+{%- for item in v -%}
+{%- if not loop.first -%}{{- ', ' -}}{%- endif -%}
+{{- jdumpsp(item) -}}
+{%- endfor -%}
+{{- ']' -}}
+{%- endif -%}
+{%- endmacro -%}{%- macro jp_ws(s, st) -%}
+{%- set w = namespace(done=false) -%}
+{%- for unused in range(s | length + 1) -%}
+{%- if not w.done -%}
+{%- if st.pos < (s | length) and s[st.pos] in ' \t\r\n' -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- else -%}
+{%- set w.done = true -%}
+{%- endif -%}
+{%- endif -%}
+{%- endfor -%}
+{%- endmacro -%}
+
+{%- macro jp_string(s, st, out) -%}
+{%- if st.pos >= (s | length) or s[st.pos] != '"' -%}
+{%- set st.ok = false -%}
+{%- else -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- set q = namespace(text='', done=false) -%}
+{%- for unused in range(s | length + 1) -%}
+{%- if st.ok and not q.done -%}
+{%- if st.pos >= (s | length) -%}
+{%- set st.ok = false -%}
+{%- else -%}
+{%- set ch = s[st.pos] -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- if ch == '"' -%}
+{%- set q.done = true -%}
+{%- elif ch == '\\' -%}
+{%- if st.pos >= (s | length) -%}
+{%- set st.ok = false -%}
+{%- else -%}
+{%- set esc = s[st.pos] -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- if esc == '"' -%}{%- set q.text = q.text + '"' -%}
+{%- elif esc == '\\' -%}{%- set q.text = q.text + '\\' -%}
+{%- elif esc == '/' -%}{%- set q.text = q.text + '/' -%}
+{%- elif esc == 'b' -%}{%- set q.text = q.text + '\b' -%}
+{%- elif esc == 'f' -%}{%- set q.text = q.text + '\f' -%}
+{%- elif esc == 'n' -%}{%- set q.text = q.text + '\n' -%}
+{%- elif esc == 'r' -%}{%- set q.text = q.text + '\r' -%}
+{%- elif esc == 't' -%}{%- set q.text = q.text + '\t' -%}
+{%- elif esc == 'u' -%}
+{%- if st.pos + 4 > (s | length) -%}
+{%- set st.ok = false -%}
+{%- else -%}
+{%- set hx = s[st.pos:st.pos + 4] -%}
+{%- set hv = namespace(ok=true) -%}
+{%- for hc in hx -%}
+{%- if hc not in '0123456789abcdefABCDEF' -%}{%- set hv.ok = false -%}{%- endif -%}
+{%- endfor -%}
+{%- if not hv.ok -%}
+{%- set st.ok = false -%}
+{%- else -%}
+{%- set cp = hx | int(base=16) -%}
+{%- set st.pos = st.pos + 4 -%}{%- if cp >= 55296 and cp <= 56319
+       and st.pos + 6 <= (s | length)
+       and s[st.pos:st.pos + 2] == '\\u' -%}
+{%- set lh = s[st.pos + 2:st.pos + 6] -%}
+{%- set lv = namespace(ok=true) -%}
+{%- for lc in lh -%}
+{%- if lc not in '0123456789abcdefABCDEF' -%}{%- set lv.ok = false -%}{%- endif -%}
+{%- endfor -%}
+{%- set low = lh | int(base=16) if lv.ok else 0 -%}
+{%- if lv.ok and low >= 56320 and low <= 57343 -%}
+{%- set cp = 65536 + (cp - 55296) * 1024 + (low - 56320) -%}
+{%- set st.pos = st.pos + 6 -%}
+{%- endif -%}
+{%- endif -%}
+{%- set q.text = q.text + ('%c' | format(cp)) -%}
+{%- endif -%}
+{%- endif -%}
+{%- else -%}
+{%- set st.ok = false -%}
+{%- endif -%}
+{%- endif -%}
+{%- else -%}{%- set ce = ch | tojson -%}
+{%- if ce in ['"\\b"', '"\\f"', '"\\n"', '"\\r"', '"\\t"']
+       or (ce[1:5] == '\\u00' and ce[5:7] in [
+           '00','01','02','03','04','05','06','07','08','09','0a','0b','0c','0d','0e','0f',
+           '10','11','12','13','14','15','16','17','18','19','1a','1b','1c','1d','1e','1f'
+       ]) -%}
+{%- set st.ok = false -%}
+{%- else -%}
+{%- set q.text = q.text + ch -%}
+{%- endif -%}
+{%- endif -%}
+{%- endif -%}
+{%- endif -%}
+{%- endfor -%}
+{%- if st.ok and not q.done -%}{%- set st.ok = false -%}{%- endif -%}
+{%- if st.ok -%}{%- set out.node = ['string', q.text] -%}{%- endif -%}
+{%- endif -%}
+{%- endmacro -%}
+
+{%- macro jp_number(s, st, out) -%}
+{%- set start = st.pos -%}
+{%- set n = namespace(done=false, float=false) -%}
+{%- if s[st.pos:st.pos + 3] == 'NaN' -%}
+{%- set token = s[st.pos:st.pos + 3] -%}
+{%- set st.pos = st.pos + 3 -%}
+{%- set out.node = ['number', token | float] -%}
+{%- elif s[st.pos:st.pos + 8] == 'Infinity' -%}
+{%- set token = s[st.pos:st.pos + 8] -%}
+{%- set st.pos = st.pos + 8 -%}
+{%- set out.node = ['number', token | float] -%}
+{%- elif s[st.pos:st.pos + 9] == '-Infinity' -%}
+{%- set token = s[st.pos:st.pos + 9] -%}
+{%- set st.pos = st.pos + 9 -%}
+{%- set out.node = ['number', token | float] -%}
+{%- else -%}
+{%- if st.pos < (s | length) and s[st.pos] == '-' -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- endif -%}
+{%- if st.pos >= (s | length) or s[st.pos] not in '0123456789' -%}
+{%- set st.ok = false -%}
+{%- elif s[st.pos] == '0' -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- if st.pos < (s | length) and s[st.pos] in '0123456789' -%}{%- set st.ok = false -%}{%- endif -%}
+{%- else -%}
+{%- for unused in range(s | length + 1) -%}
+{%- if not n.done -%}
+{%- if st.pos < (s | length) and s[st.pos] in '0123456789' -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- else -%}{%- set n.done = true -%}
+{%- endif -%}
+{%- endif -%}
+{%- endfor -%}
+{%- endif -%}
+{%- if st.ok and st.pos < (s | length) and s[st.pos] == '.' -%}
+{%- set n.float = true -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- if st.pos >= (s | length) or s[st.pos] not in '0123456789' -%}
+{%- set st.ok = false -%}
+{%- else -%}
+{%- set n.done = false -%}
+{%- for unused in range(s | length + 1) -%}
+{%- if not n.done -%}
+{%- if st.pos < (s | length) and s[st.pos] in '0123456789' -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- else -%}{%- set n.done = true -%}
+{%- endif -%}
+{%- endif -%}
+{%- endfor -%}
+{%- endif -%}
+{%- endif -%}
+{%- if st.ok and st.pos < (s | length) and s[st.pos] in 'eE' -%}
+{%- set n.float = true -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- if st.pos < (s | length) and s[st.pos] in '+-' -%}{%- set st.pos = st.pos + 1 -%}{%- endif -%}
+{%- if st.pos >= (s | length) or s[st.pos] not in '0123456789' -%}
+{%- set st.ok = false -%}
+{%- else -%}
+{%- set n.done = false -%}
+{%- for unused in range(s | length + 1) -%}
+{%- if not n.done -%}
+{%- if st.pos < (s | length) and s[st.pos] in '0123456789' -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- else -%}{%- set n.done = true -%}
+{%- endif -%}
+{%- endif -%}
+{%- endfor -%}
+{%- endif -%}
+{%- endif -%}
+{%- if st.ok -%}
+{%- set token = s[start:st.pos] -%}
+{%- if n.float -%}{%- set out.node = ['number', token | float] -%}
+{%- else -%}{%- set out.node = ['number', token | int] -%}
+{%- endif -%}
+{%- endif -%}
+{%- endif -%}
+{%- endmacro -%}
+
+{%- macro jp_array(s, st, out) -%}
+{%- set st.pos = st.pos + 1 -%}
+{{- jp_ws(s, st) -}}
+{%- set a = namespace(elems=[], done=false) -%}
+{%- if st.pos < (s | length) and s[st.pos] == ']' -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- set a.done = true -%}
+{%- endif -%}
+{%- for unused in range(s | length + 1) -%}
+{%- if st.ok and not a.done -%}
+{%- set child = namespace(node=none) -%}
+{{- jp_value(s, st, child) -}}
+{%- if st.ok -%}
+{%- set a.elems = a.elems + [child.node] -%}
+{{- jp_ws(s, st) -}}
+{%- if st.pos < (s | length) and s[st.pos] == ',' -%}
+{%- set st.pos = st.pos + 1 -%}
+{{- jp_ws(s, st) -}}
+{%- elif st.pos < (s | length) and s[st.pos] == ']' -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- set a.done = true -%}
+{%- else -%}
+{%- set st.ok = false -%}
+{%- endif -%}
+{%- endif -%}
+{%- endif -%}
+{%- endfor -%}
+{%- if st.ok and not a.done -%}{%- set st.ok = false -%}{%- endif -%}
+{%- if st.ok -%}{%- set out.node = ["array", a.elems] -%}{%- endif -%}
+{%- endmacro -%}
+
+{%- macro jp_object(s, st, out) -%}
+{%- set st.pos = st.pos + 1 -%}
+{{- jp_ws(s, st) -}}
+{%- set o = namespace(pairs=[], done=false) -%}
+{%- if st.pos < (s | length) and s[st.pos] == '}' -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- set o.done = true -%}
+{%- endif -%}
+{%- for unused in range(s | length + 1) -%}
+{%- if st.ok and not o.done -%}
+{%- set key = namespace(node=none) -%}
+{{- jp_string(s, st, key) -}}
+{%- if st.ok -%}
+{{- jp_ws(s, st) -}}
+{%- if st.pos >= (s | length) or s[st.pos] != ':' -%}
+{%- set st.ok = false -%}
+{%- else -%}
+{%- set st.pos = st.pos + 1 -%}
+{{- jp_ws(s, st) -}}
+{%- set child = namespace(node=none) -%}
+{{- jp_value(s, st, child) -}}
+{%- if st.ok -%}{%- set hit = namespace(index=none) -%}
+{%- for pair in o.pairs -%}
+{%- if hit.index is none and pair[0] == key.node[1] -%}{%- set hit.index = loop.index0 -%}{%- endif -%}
+{%- endfor -%}
+{%- if hit.index is none -%}
+{%- set o.pairs = o.pairs + [[key.node[1], child.node]] -%}
+{%- else -%}
+{%- set repl = namespace(pairs=[]) -%}
+{%- for pair in o.pairs -%}
+{%- set repl.pairs = repl.pairs + [[pair[0], child.node] if loop.index0 == hit.index else pair] -%}
+{%- endfor -%}
+{%- set o.pairs = repl.pairs -%}
+{%- endif -%}
+{{- jp_ws(s, st) -}}
+{%- if st.pos < (s | length) and s[st.pos] == ',' -%}
+{%- set st.pos = st.pos + 1 -%}
+{{- jp_ws(s, st) -}}
+{%- elif st.pos < (s | length) and s[st.pos] == '}' -%}
+{%- set st.pos = st.pos + 1 -%}
+{%- set o.done = true -%}
+{%- else -%}
+{%- set st.ok = false -%}
+{%- endif -%}
+{%- endif -%}
+{%- endif -%}
+{%- endif -%}
+{%- endif -%}
+{%- endfor -%}
+{%- if st.ok and not o.done -%}{%- set st.ok = false -%}{%- endif -%}
+{%- if st.ok -%}{%- set out.node = ['object', o.pairs] -%}{%- endif -%}
+{%- endmacro -%}
+
+{%- macro jp_value(s, st, out) -%}
+{{- jp_ws(s, st) -}}
+{%- if st.pos >= (s | length) -%}
+{%- set st.ok = false -%}
+{%- elif s[st.pos] == '"' -%}
+{{- jp_string(s, st, out) -}}
+{%- elif s[st.pos] == '{' -%}
+{{- jp_object(s, st, out) -}}
+{%- elif s[st.pos] == '[' -%}
+{{- jp_array(s, st, out) -}}
+{%- elif s[st.pos:st.pos + 4] == 'true' -%}
+{%- set st.pos = st.pos + 4 -%}{%- set out.node = ['boolean', true] -%}
+{%- elif s[st.pos:st.pos + 5] == 'false' -%}
+{%- set st.pos = st.pos + 5 -%}{%- set out.node = ['boolean', false] -%}
+{%- elif s[st.pos:st.pos + 4] == 'null' -%}
+{%- set st.pos = st.pos + 4 -%}{%- set out.node = ['null', none] -%}
+{%- elif s[st.pos] in '-0123456789NI' -%}
+{{- jp_number(s, st, out) -}}
+{%- else -%}
+{%- set st.ok = false -%}
+{%- endif -%}
+{%- endmacro -%}
+
+{%- macro jp_type(node) -%}
+{%- if node[0] == 'boolean' -%}boolean
+{%- elif node[0] == 'null' -%}null
+{%- elif node[0] == 'number' -%}number
+{%- elif node[0] == 'string' -%}string
+{%- elif node[0] == 'object' -%}object
+{%- else -%}array
+{%- endif -%}
+{%- endmacro -%}
+
+{%- macro jp_dump(node) -%}
+{%- if node[0] == 'object' -%}
+{{- '{' -}}
+{%- for pair in node[1] -%}
+{%- if not loop.first -%}{{- ', ' -}}{%- endif -%}
+{{- pair[0] | tojson -}}{{- ': ' -}}{{- jp_dump(pair[1]) -}}
+{%- endfor -%}
+{{- '}' -}}
+{%- elif node[0] == 'array' -%}
+{{- '[' -}}
+{%- for child in node[1] -%}
+{%- if not loop.first -%}{{- ', ' -}}{%- endif -%}
+{{- jp_dump(child) -}}
+{%- endfor -%}
+{{- ']' -}}
+{%- elif node[0] == 'string' -%}
+{{- node[1] | tojson -}}
+{%- else -%}
+{{- numstr(node[1]) -}}
+{%- endif -%}
+{%- endmacro -%}{%- macro xtype(v) -%}
+{%- if v is boolean -%}boolean
+{%- elif v is none -%}null
+{%- elif v is number -%}number
+{%- elif v is string -%}string
+{%- elif v is mapping -%}object
+{%- else -%}array
+{%- endif -%}
+{%- endmacro -%}{%- macro xvalue(v) -%}
+{%- if v is string -%}{{- v -}}{%- else -%}{{- jdumpsp(v) -}}{%- endif -%}
+{%- endmacro -%}{%- macro render_content(c) -%}
+{%- if c is string -%}
+{{- xtext(c) -}}
+{%- elif c is mapping and (c | length) == 0 -%}
+{%- elif c is mapping -%}
+{{- raise_exception('Kimi K3 message content must be a string or a list of content parts.') -}}
+{%- elif c is defined and c is not none -%}
+{%- for part in c -%}
+{%- if part is not mapping or 'type' not in part -%}
+{{- raise_exception('Kimi K3 content parts must be objects carrying a `type`.') -}}
+{%- elif part['type'] in ['image', 'image_url'] -%}
+{{- img_next() -}}
+{%- elif 'text' not in part -%}
+{{- raise_exception('Kimi K3 non-image content parts require a `text`.') -}}
+{%- else -%}
+{{- xtext(part['text']) -}}
+{%- endif -%}
+{%- endfor -%}
+{%- endif -%}
+{%- endmacro -%}{%- macro internal_msg(kind, body) -%}
+{{- otag('message', [['role', 'system'], ['type', kind]]) -}}
+{{- body | trim -}}
+{{- ctag('message') -}}
+{{- '<|end_of_msg|>' -}}
+{%- endmacro -%}{%- macro render_tool_declare(tool_list, dynamic=false) -%}
+{{- otag('message', [['role', 'system'], ['type', 'tool-declare']]) -}}
+{%- if dynamic -%}
+{{- '## New Tools Available\nThe system dynamically extends the toolset via lazy-loading.\nYou have access to all existing and extended tools.\nHere are the specs for the extended tools.\n\n```json\n' -}}
+{%- else -%}
+{{- '# Tools\nHere are the available tools, described in JSONSchema.\n\n```json\n' -}}
+{%- endif -%}
+{{- jdump(tool_list) -}}
+{{- '\n```' -}}
+{{- ctag('message') -}}
+{{- '<|end_of_msg|>' -}}
+{%- endmacro -%}{%- macro render_assistant(m, thinking) -%}
+{%- if thinking -%}
+{%- set rc = m.get('reasoning_content') or m.get('reasoning') -%}
+{{- otag('think') -}}
+{%- if rc is not none and (rc | string | trim) -%}{{- xtext(rc) -}}{%- endif -%}
+{{- ctag('think') -}}
+{%- endif -%}
+{{- otag('response') -}}
+{{- render_content(m.get('content')) -}}
+{{- ctag('response') -}}
+{%- set tcs = m.get('tool_calls') -%}
+{%- if tcs -%}
+{{- otag('tools') -}}
+{%- for tc in tcs -%}
+{%- set fn = tc['function'] if ('function' in tc) else tc -%}
+{%- if 'name' not in fn -%}
+{{- raise_exception('Kimi K3 tool calls require a function name.') -}}
+{%- endif -%}
+{{- otag('call', [['tool', fn['name']], ['index', loop.index]]) -}}
+{%- set args = fn['arguments'] if ('arguments' in fn) else none -%}{%- if args is mapping -%}
+{%- for key, val in args | items -%}
+{{- otag('argument', [['key', key], ['type', xtype(val)]]) -}}
+{{- xtext(xvalue(val)) -}}
+{{- ctag('argument') -}}
+{%- endfor -%}
+{%- elif args is string -%}
+{%- if args | trim -%}
+{%- set ps = namespace(pos=0, ok=true) -%}
+{%- set parsed = namespace(node=none) -%}
+{{- jp_value(args, ps, parsed) -}}
+{{- jp_ws(args, ps) -}}
+{%- if not ps.ok or ps.pos != (args | length) -%}{{- otag('json', [['type', 'object']]) -}}{{- xtext(args) -}}{{- ctag('json') -}}
+{%- elif parsed.node[0] != 'object' -%}
+{{- raise_exception('Kimi K3 tool call arguments must be a JSON object.') -}}
+{%- else -%}
+{%- for pair in parsed.node[1] -%}
+{{- otag('argument', [['key', pair[0]], ['type', jp_type(pair[1])]]) -}}
+{%- if pair[1][0] == 'string' -%}
+{{- xtext(pair[1][1]) -}}
+{%- else -%}
+{{- xtext(jp_dump(pair[1])) -}}
+{%- endif -%}
+{{- ctag('argument') -}}
+{%- endfor -%}
+{%- endif -%}
+{%- endif -%}
+{%- elif args is not none and args is defined -%}
+{{- raise_exception('Kimi K3 tool call arguments must be a dict or a JSON object string.') -}}
+{%- endif -%}
+{{- ctag('call') -}}
+{%- endfor -%}
+{{- ctag('tools') -}}
+{%- endif -%}
+{%- endmacro -%}{%- macro render_tool_msg(m, idx, forced, tcalls) -%}
+{%- set r = namespace(name = none) -%}
+{%- if forced is not none -%}
+{%- set r.name = forced -%}
+{%- elif 'tool' in m -%}
+{%- set r.name = m['tool'] -%}
+{%- else -%}
+{%- set r.name = m.get('name') -%}
+{%- endif -%}
+{%- if r.name is none and tcalls and idx <= (tcalls | length) -%}
+{%- set tc = tcalls[idx - 1] -%}
+{%- set fn = tc['function'] if ('function' in tc) else tc -%}
+{%- set r.name = fn['name'] -%}
+{%- endif -%}
+{%- if r.name is not defined or r.name is none -%}
+{{- raise_exception('Kimi K3 tool messages need a resolvable tool name: carry `tool`/`name`, or match a preceding assistant tool_call by order.') -}}
+{%- endif -%}
+{{- otag('message', [['role', 'tool'], ['tool', r.name], ['index', idx]]) -}}
+{{- render_content(m.get('content')) -}}
+{{- ctag('message') -}}
+{{- '<|end_of_msg|>' -}}
+{%- endmacro -%}{%- set rens = namespace(off = false, effort = none) -%}
+{%- if reasoning_effort is defined and reasoning_effort is not none -%}
+{%- if reasoning_effort == 'none' -%}
+{%- set rens.off = true -%}
+{%- elif reasoning_effort in ['low', 'high', 'max'] -%}
+{%- set rens.effort = reasoning_effort -%}
+{%- endif -%}
+{%- endif -%}
+{%- if thinking is not defined -%}
+{%- if enable_thinking is defined -%}
+{%- set thinking = enable_thinking -%}
+{%- elif rens.off -%}
+{%- set thinking = false -%}
+{%- else -%}
+{%- set thinking = true -%}
+{%- endif -%}
+{%- endif -%}{%- if thinking_effort is not defined -%}{%- set thinking_effort = rens.effort or 'max' -%}
+{%- endif -%}{%- if thinking and thinking_effort is not none and thinking_effort not in ['low', 'high', 'max'] -%}
+{{- raise_exception("Unsupported thinking_effort=" + (thinking_effort | string) + "; supported values are ['high', 'low', 'max'].") -}}
+{%- endif -%}{%- if tools is defined and tools -%}
+{{- render_tool_declare(tools) -}}
+{%- endif -%}
+{%- if thinking and thinking_effort in ['low', 'high', 'max'] -%}
+{{- internal_msg('thinking-effort', '`thinking_effort` guides on how much to think in your thinking channel (not including the response channel), supported values include `low`, `medium`, `high`, and `max`.\nNow the system is invoked with `thinking_effort=' + (thinking_effort | string) + '`.') -}}
+{%- endif -%}{%- set ns = namespace(tool_calls = none, tool_index = 0) -%}
+{%- for message in messages -%}
+{%- if message is mapping -%}
+{%- if 'role' not in message -%}
+{{- raise_exception('Kimi K3 messages require a `role`.') -}}
+{%- endif -%}
+{%- set role = message['role'] -%}
+
+{%- if role == 'user' -%}
+{%- set attrs = [['role', 'user']] -%}
+{%- if message.get('name') -%}{%- set attrs = attrs + [['name', message['name']]] -%}{%- endif -%}
+{{- otag('message', attrs) -}}
+{{- render_content(message.get('content')) -}}
+{{- ctag('message') -}}
+{{- '<|end_of_msg|>' -}}
+
+{%- elif role == 'system' and message.get('tools') -%}{{- render_tool_declare(message['tools'], true) -}}
+
+{%- elif role == 'system' -%}
+{%- set attrs = [['role', 'system']] -%}
+{%- if message.get('name') -%}{%- set attrs = attrs + [['name', message['name']]] -%}{%- endif -%}
+{{- otag('message', attrs) -}}
+{{- render_content(message.get('content')) -}}
+{{- ctag('message') -}}
+{{- '<|end_of_msg|>' -}}
+
+{%- elif role == 'assistant' -%}
+{%- set ns.tool_calls = message.get('tool_calls') -%}
+{%- set ns.tool_index = 0 -%}
+{%- set attrs = [['role', 'assistant']] -%}
+{%- if message.get('name') -%}{%- set attrs = attrs + [['name', message['name']]] -%}{%- endif -%}
+{{- otag('message', attrs) -}}
+{{- render_assistant(message, thinking) -}}
+{{- ctag('message') -}}
+{{- '<|end_of_msg|>' -}}
+
+{%- elif role == 'tool' -%}{%- set here = loop.index0 -%}
+{%- set start = namespace(is_first = true) -%}
+{%- if not loop.first -%}
+{%- if messages[here - 1] is mapping -%}
+{%- if messages[here - 1].get('role') == 'tool' -%}{%- set start.is_first = false -%}{%- endif -%}
+{%- endif -%}
+{%- endif -%}
+{%- if start.is_first -%}{%- set rl = namespace(n = 0, stop = false) -%}
+{%- for cand in messages[here:] -%}
+{%- if not rl.stop -%}
+{%- if cand is mapping and cand.get('role') == 'tool' -%}
+{%- set rl.n = rl.n + 1 -%}
+{%- else -%}
+{%- set rl.stop = true -%}
+{%- endif -%}
+{%- endif -%}
+{%- endfor -%}{%- set rr = namespace(order = [], names = [], unresolved = false) -%}
+{%- for off in range(rl.n) -%}
+{%- set tm = messages[here + off] -%}
+{%- set cid = namespace(v = none) -%}
+{%- if 'tool_call_id' in tm -%}
+{%- set cid.v = tm['tool_call_id'] -%}
+{%- else -%}
+{%- set cid.v = tm.get('id') -%}
+{%- endif -%}
+{%- set mt = namespace(pos = none, name = none) -%}
+{%- if cid.v is not none and ns.tool_calls -%}
+{%- for tc in ns.tool_calls -%}
+{%- if mt.pos is none and tc is mapping and tc.get('id') is not none and (tc.get('id') | string) == (cid.v | string) -%}
+{%- set mt.pos = loop.index -%}
+{%- if tc.get('function') is mapping -%}
+{%- set mt.name = tc['function'].get('name') -%}
+{%- else -%}
+{%- set mt.name = tc.get('name') -%}
+{%- endif -%}
+{%- endif -%}
+{%- endfor -%}
+{%- endif -%}
+{%- if mt.pos is none -%}{%- set rr.unresolved = true -%}{%- endif -%}{%- set rr.order = rr.order + [(mt.pos if mt.pos is not none else 0) * 1000000 + off] -%}
+{%- set rr.names = rr.names + [mt.name] -%}
+{%- endfor -%}{%- if rr.unresolved -%}
+{%- for off in range(rl.n) -%}
+{%- set ns.tool_index = ns.tool_index + 1 -%}
+{{- render_tool_msg(messages[here + off], ns.tool_index, none, ns.tool_calls) -}}
+{%- endfor -%}
+{%- else -%}
+{%- for key in rr.order | sort -%}
+{%- set off = key % 1000000 -%}
+{%- set ns.tool_index = ns.tool_index + 1 -%}
+{{- render_tool_msg(messages[here + off], ns.tool_index, rr.names[off], ns.tool_calls) -}}
+{%- endfor -%}
+{%- endif -%}
+{%- endif -%}
+{%- endif -%}
+{%- endif -%}
+{%- endfor -%}{%- if tool_choice is defined and tool_choice == 'required' -%}
+{{- internal_msg('tool-choice', 'The system is invoked with `tool_choice=required`.\nYou MUST call tools in the next message.') -}}
+{%- elif tool_choice is defined and tool_choice == 'none' -%}
+{{- internal_msg('tool-choice', 'The system is invoked with `tool_choice=none`.\nYou MUST NOT call any tools in the next message.') -}}
+{%- endif -%}{%- set rsx = namespace(v = none) -%}
+{%- if response_schema is defined -%}
+{%- set rsx.v = response_schema -%}
+{%- elif response_format is defined and response_format is mapping and response_format.get('json_schema') is not none -%}
+{%- set js = response_format['json_schema'] -%}
+{%- if js is mapping -%}
+{%- if 'schema' in js -%}
+{%- set rsx.v = js['schema'] -%}
+{%- elif 'json_schema' in js -%}
+{%- set rsx.v = js['json_schema'] -%}
+{%- else -%}
+{%- set rsx.v = js -%}
+{%- endif -%}
+{%- else -%}
+{%- set rsx.v = js -%}
+{%- endif -%}
+{%- endif -%}
+{%- set rft = namespace(v = none) -%}
+{%- if response_format is defined and response_format is not none -%}
+{%- if response_format is mapping -%}
+{%- set rft.v = response_format.get('type', response_format) -%}
+{%- else -%}
+{%- set rft.v = response_format -%}
+{%- endif -%}
+{%- endif -%}
+{%- if rft.v == 'json_object' -%}
+{{- internal_msg('response-format', 'The system is invoked with `response_format=json_object`.\nYour response must be raw JSON data without markdown code blocks (```json) or any additional formatting.') -}}
+{%- elif rft.v == 'json_schema' -%}
+{{- internal_msg('response-format', 'The system is invoked with `response_format=json_schema`.\nYour response must be raw JSON data without markdown code blocks (```json) or any additional formatting.\nThe JSON data must match the following schema:\n```json\n' + jdump(rsx.v) + '\n```') -}}
+{%- endif -%}
+
+{%- if add_generation_prompt -%}
+{{- otag('message', [['role', 'assistant']]) -}}
+{%- if thinking -%}{{- otag('think') -}}{%- else -%}{{- otag('response') -}}{%- endif -%}
+{%- endif -%}{%- if imgns.on and imgns.idx != (image_prompts | length) -%}
+{{- raise_exception('image prompt count ' + (image_prompts | length | string) + ' != consumed placeholder count ' + (imgns.idx | string)) -}}
+{%- endif -%}
+{#- Unsloth translation to jinja #}

config.json

@@ -0,0 +1,338 @@
+{
+  "architectures": [
+    "KimiK3ForConditionalGeneration"
+  ],
+  "auto_map": {
+    "AutoConfig": "configuration_kimi_k3.KimiK3Config",
+    "AutoModel": "modeling_kimi_k3.KimiK3ForConditionalGeneration",
+    "AutoModelForCausalLM": "modeling_kimi_k3.KimiK3ForConditionalGeneration"
+  },
+  "bos_token_id": 163584,
+  "torch_dtype": "bfloat16",
+  "eos_token_id": 163586,
+  "ignore_index": -100,
+  "image_placeholder": "<|kimi_image_placeholder|>",
+  "media_placeholder_token_id": 163605,
+  "model_type": "kimi_k3",
+  "pad_token_id": 163839,
+  "quantization_config": {
+    "config_groups": {
+      "group_0": {
+        "format": "mxfp4-pack-quantized",
+        "input_activations": null,
+        "output_activations": null,
+        "targets": [
+          "Linear"
+        ],
+        "weights": {
+          "actorder": null,
+          "block_structure": null,
+          "dynamic": false,
+          "group_size": 32,
+          "num_bits": 4,
+          "observer": "minmax",
+          "observer_kwargs": {},
+          "scale_dtype": "torch.uint8",
+          "strategy": "group",
+          "symmetric": true,
+          "type": "float",
+          "zp_dtype": null
+        }
+      }
+    },
+    "format": "mxfp4-pack-quantized",
+    "global_compression_ratio": null,
+    "ignore": [
+      "re:.*self_attn.*",
+      "re:.*shared_experts.*",
+      "re:.*mlp\\.(gate|up|gate_up|down)_proj.*",
+      "re:.*lm_head.*",
+      "re:.*vision_tower.*",
+      "re:.*mm_projector.*"
+    ],
+    "kv_cache_scheme": null,
+    "quant_method": "compressed-tensors",
+    "quantization_status": "compressed"
+  },
+  "text_config": {
+    "_name_or_path": "",
+    "activation_situ_beta": 4.0,
+    "activation_situ_linear_beta": 25.0,
+    "add_cross_attention": false,
+    "architectures": [
+      "KimiLinearForCausalLM"
+    ],
+    "attn_res_block_size": 12,
+    "auto_map": {
+      "AutoConfig": "configuration_kimi_k3.KimiLinearConfig",
+      "AutoModel": "modeling_kimi_linear.KimiLinearModel",
+      "AutoModelForCausalLM": "modeling_kimi_linear.KimiLinearForCausalLM"
+    },
+    "bos_token_id": 163584,
+    "chunk_size_feed_forward": 0,
+    "cross_attention_hidden_size": null,
+    "decoder_start_token_id": null,
+    "torch_dtype": "bfloat16",
+    "eos_token_id": 163586,
+    "finetuning_task": null,
+    "first_k_dense_replace": 1,
+    "head_dim": 74,
+    "hidden_act": "situ",
+    "hidden_size": 7168,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "initializer_range": 0.02,
+    "intermediate_size": 33792,
+    "is_decoder": false,
+    "is_encoder_decoder": false,
+    "kv_lora_rank": 512,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "latent_moe_use_norm": true,
+    "linear_attn_config": {
+      "full_attn_layers": [
+        4,
+        8,
+        12,
+        16,
+        20,
+        24,
+        28,
+        32,
+        36,
+        40,
+        44,
+        48,
+        52,
+        56,
+        60,
+        64,
+        68,
+        72,
+        76,
+        80,
+        84,
+        88,
+        92,
+        93
+      ],
+      "gate_lower_bound": -5.0,
+      "head_dim": 128,
+      "kda_layers": [
+        1,
+        2,
+        3,
+        5,
+        6,
+        7,
+        9,
+        10,
+        11,
+        13,
+        14,
+        15,
+        17,
+        18,
+        19,
+        21,
+        22,
+        23,
+        25,
+        26,
+        27,
+        29,
+        30,
+        31,
+        33,
+        34,
+        35,
+        37,
+        38,
+        39,
+        41,
+        42,
+        43,
+        45,
+        46,
+        47,
+        49,
+        50,
+        51,
+        53,
+        54,
+        55,
+        57,
+        58,
+        59,
+        61,
+        62,
+        63,
+        65,
+        66,
+        67,
+        69,
+        70,
+        71,
+        73,
+        74,
+        75,
+        77,
+        78,
+        79,
+        81,
+        82,
+        83,
+        85,
+        86,
+        87,
+        89,
+        90,
+        91
+      ],
+      "num_heads": 96,
+      "short_conv_kernel_size": 4,
+      "use_full_rank_gate": true
+    },
+    "max_position_embeddings": 1048576,
+    "mla_use_nope": true,
+    "mla_use_output_gate": true,
+    "model_type": "kimi_linear",
+    "moe_intermediate_size": 3072,
+    "moe_layer_freq": 1,
+    "moe_renormalize": true,
+    "moe_router_activation_func": "sigmoid",
+    "num_attention_heads": 96,
+    "num_expert_group": 1,
+    "num_experts": 896,
+    "num_experts_per_token": 16,
+    "num_hidden_layers": 93,
+    "num_key_value_heads": 96,
+    "num_nextn_predict_layers": 0,
+    "num_shared_experts": 2,
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "pad_token_id": 163839,
+    "prefix": null,
+    "problem_type": null,
+    "pruned_heads": {},
+    "q_lora_rank": 1536,
+    "qk_nope_head_dim": 128,
+    "qk_rope_head_dim": 64,
+    "quantization_config": {
+      "config_groups": {
+        "group_0": {
+          "format": "mxfp4-pack-quantized",
+          "input_activations": null,
+          "output_activations": null,
+          "targets": [
+            "Linear"
+          ],
+          "weights": {
+            "actorder": null,
+            "block_structure": null,
+            "dynamic": false,
+            "group_size": 32,
+            "num_bits": 4,
+            "observer": "minmax",
+            "observer_kwargs": {},
+            "scale_dtype": "torch.uint8",
+            "strategy": "group",
+            "symmetric": true,
+            "type": "float",
+            "zp_dtype": null
+          }
+        }
+      },
+      "format": "mxfp4-pack-quantized",
+      "global_compression_ratio": null,
+      "ignore": [
+        "re:.*self_attn.*",
+        "re:.*shared_experts.*",
+        "re:.*mlp\\.(gate|up|gate_up|down)_proj.*",
+        "re:.*lm_head.*",
+        "re:.*vision_tower.*",
+        "re:.*mm_projector.*"
+      ],
+      "kv_cache_scheme": null,
+      "quant_method": "compressed-tensors",
+      "quantization_status": "compressed"
+    },
+    "return_dict": true,
+    "rms_norm_eps": 1e-05,
+    "rope_parameters": {
+      "rope_theta": 10000.0,
+      "rope_type": "default"
+    },
+    "rope_theta": 10000.0,
+    "routed_expert_hidden_size": 3584,
+    "routed_scaling_factor": 1.0,
+    "sep_token_id": null,
+    "task_specific_params": null,
+    "tf_legacy_loss": false,
+    "tie_encoder_decoder": false,
+    "tie_word_embeddings": false,
+    "tokenizer_class": null,
+    "topk_group": 1,
+    "topk_method": "noaux_tc",
+    "torchscript": false,
+    "use_bfloat16": false,
+    "use_cache": true,
+    "use_grouped_topk": true,
+    "v_head_dim": 128,
+    "vocab_size": 163840
+  },
+  "tie_word_embeddings": false,
+  "transformers_version": "5.15.0.dev0",
+  "unsloth_fixed": true,
+  "vision_config": {
+    "_name_or_path": "",
+    "activation_func": "gelu_pytorch_tanh",
+    "architectures": null,
+    "attn_bias": false,
+    "chunk_size_feed_forward": 0,
+    "torch_dtype": null,
+    "id2label": {
+      "0": "LABEL_0",
+      "1": "LABEL_1"
+    },
+    "init_pos_emb_height": 64,
+    "init_pos_emb_time": 4,
+    "init_pos_emb_width": 64,
+    "is_encoder_decoder": false,
+    "label2id": {
+      "LABEL_0": 0,
+      "LABEL_1": 1
+    },
+    "linear_bias": false,
+    "merge_kernel_size": [
+      2,
+      2
+    ],
+    "merge_type": "sd2_tpool",
+    "mlp_type": "mlp2",
+    "mm_hidden_size": 1024,
+    "mm_projector_type": "patchmergerv2",
+    "model_type": "",
+    "norm_type": "rmsnorm",
+    "output_attentions": false,
+    "output_hidden_states": false,
+    "patch_embed_proj_bias": false,
+    "patch_size": 14,
+    "pos_emb_interpolation_mode": "bilinear",
+    "pos_emb_type": "divided_fixed",
+    "problem_type": null,
+    "projector_hidden_act": "gelu",
+    "projector_ln_eps": 1e-05,
+    "qkv_hidden_size": 1536,
+    "return_dict": true,
+    "text_hidden_size": 7168,
+    "vt_hidden_size": 1024,
+    "vt_intermediate_size": 4096,
+    "vt_num_attention_heads": 12,
+    "vt_num_hidden_layers": 27
+  }
+}

configuration_kimi_k3.py

@@ -0,0 +1,285 @@
+from typing import Optional
+
+from transformers.configuration_utils import PretrainedConfig
+
+
+class KimiLinearConfig(PretrainedConfig):
+    model_type = "kimi_linear"
+    keys_to_ignore_at_inference = ["past_key_values"]
+
+    def __init__(
+        self,
+        model_type="kimi_linear",
+        vocab_size=163840,
+        hidden_size=4096,
+        head_dim=None,
+        intermediate_size=11008,
+        num_hidden_layers=32,
+        num_attention_heads=32,
+        num_key_value_heads=None,
+        hidden_act="silu",
+        initializer_range=0.02,
+        rms_norm_eps=1e-6,
+        use_cache=True,
+        pad_token_id=0,
+        bos_token_id=1,
+        eos_token_id=2,
+        rope_theta=10000.0,
+        rope_scaling=None,
+        tie_word_embeddings=False,
+        moe_intermediate_size: Optional[int] = None,
+        moe_renormalize: bool = True,
+        moe_router_activation_func: str = "sigmoid",
+        num_experts: Optional[int] = None,
+        num_experts_per_token: Optional[int] = None,
+        num_shared_experts: int = 0,
+        routed_scaling_factor: float = 1.0,
+        first_k_dense_replace: int = 0,
+        moe_layer_freq: int = 1,
+        use_grouped_topk: bool = True,
+        num_expert_group: int = 1,
+        topk_group: int = 1,
+        q_lora_rank: Optional[int] = None,
+        kv_lora_rank: Optional[int] = None,
+        qk_nope_head_dim: Optional[int] = None,
+        qk_rope_head_dim: Optional[int] = None,
+        v_head_dim: Optional[int] = None,
+        mla_use_nope: Optional[bool] = False,
+        mla_use_output_gate: Optional[bool] = False,
+        num_nextn_predict_layers: int = 0,
+        linear_attn_config: Optional[dict] = None,
+        attn_res_block_size: Optional[int] = None,
+        latent_moe_use_norm: bool = False,
+        activation_situ_beta: Optional[float] = None,
+        activation_situ_linear_beta: Optional[float] = None,
+        max_position_embeddings: int = 4096,
+        routed_expert_hidden_size: Optional[int] = None,
+        topk_method: str = "noaux_tc",
+        **kwargs,
+    ):
+        self.model_type = model_type
+        self.vocab_size = vocab_size
+        self.hidden_size = hidden_size
+        self.head_dim = (
+            head_dim if head_dim is not None else hidden_size // num_attention_heads
+        )
+        self.intermediate_size = intermediate_size
+        self.num_hidden_layers = num_hidden_layers
+        self.num_attention_heads = num_attention_heads
+
+        # 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.rope_scaling = rope_scaling
+
+        self.q_lora_rank = q_lora_rank
+        self.kv_lora_rank = kv_lora_rank
+        self.qk_nope_head_dim = qk_nope_head_dim
+        self.qk_rope_head_dim = qk_rope_head_dim
+        self.v_head_dim = v_head_dim
+        self.mla_use_nope = mla_use_nope
+        self.mla_use_output_gate = mla_use_output_gate
+        # moe config
+        self.num_experts = num_experts
+        self.num_experts_per_token = num_experts_per_token
+        self.moe_renormalize = moe_renormalize
+        self.num_shared_experts = num_shared_experts
+        self.routed_scaling_factor = routed_scaling_factor
+        self.moe_router_activation_func = moe_router_activation_func
+        assert self.moe_router_activation_func in ("softmax", "sigmoid")
+        self.moe_intermediate_size = moe_intermediate_size
+        self.first_k_dense_replace = first_k_dense_replace
+        self.moe_layer_freq = moe_layer_freq
+        self.use_grouped_topk = use_grouped_topk
+        self.num_expert_group = num_expert_group
+        self.topk_group = topk_group
+        self.num_nextn_predict_layers = num_nextn_predict_layers
+
+        self.attn_res_block_size = attn_res_block_size
+        self.latent_moe_use_norm = latent_moe_use_norm
+        self.activation_situ_beta = activation_situ_beta
+        self.activation_situ_linear_beta = activation_situ_linear_beta
+        self.max_position_embeddings = max_position_embeddings
+        self.routed_expert_hidden_size = routed_expert_hidden_size
+        self.topk_method = topk_method
+
+        if linear_attn_config is not None:
+            assert linear_attn_config["kda_layers"] is not None
+            assert linear_attn_config["full_attn_layers"] is not None
+        self.linear_attn_config = linear_attn_config
+
+        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,
+        )
+
+    @property
+    def is_mla(self):
+        return (
+            self.q_lora_rank is not None
+            or self.kv_lora_rank is not None
+            or self.qk_nope_head_dim is not None
+            or self.qk_rope_head_dim is not None
+            or self.v_head_dim is not None
+            or self.mla_use_nope is True
+        )
+
+    @property
+    def is_moe(self):
+        return self.num_experts is not None
+
+    @property
+    def is_linear_attn(self) -> bool:
+        return not (
+            self.linear_attn_config is None
+            or (
+                isinstance(self.linear_attn_config, dict)
+                and self.linear_attn_config["kda_layers"] is not None
+                and len(self.linear_attn_config["kda_layers"]) == 0
+            )
+        )
+
+    def is_kda_layer(self, layer_idx: int):
+        return (
+            self.linear_attn_config is not None
+            and (layer_idx + 1) in self.linear_attn_config["kda_layers"]
+        )
+
+
+class KimiK3VisionConfig(PretrainedConfig):
+
+    def __init__(
+            self,
+            patch_size: int = 14,
+            init_pos_emb_height: int = 64,
+            init_pos_emb_width: int = 64,
+            init_pos_emb_time: int = 4,
+            pos_emb_type: str = 'divided_fixed',
+            vt_num_attention_heads: int = 12,
+            vt_num_hidden_layers: int = 27,
+            vt_hidden_size: int = 1024,
+            vt_intermediate_size: int = 4096,
+            merge_kernel_size: tuple = (2, 2),
+            merge_type: str = 'sd2_tpool',
+            _attn_implementation: str = 'flash_attention_2',
+            # MM Projector parameters
+            mm_projector_type: str = 'patchmergerv2',
+            mm_hidden_size: int | None = None,
+            projector_hidden_act: str = "gelu",
+            projector_ln_eps: float = 1e-5,
+            # vision tower parameters
+            qkv_hidden_size: int = 1536,
+            norm_type: str = 'rmsnorm',
+            attn_bias: bool = False,
+            patch_embed_proj_bias: bool = False,
+            mlp_type: str = 'mlp2',
+            linear_bias: bool = False,
+            activation_func: str = 'gelu_pytorch_tanh',
+            pos_emb_interpolation_mode: str = 'bilinear',
+            # Other parameters
+            ignore_index: int = -100,
+            media_placeholder_token_id: int = 163605,
+            pad_token_id: int = 0,
+            text_hidden_size=7168,
+            **kwargs):
+
+        self.patch_size = patch_size
+        self.init_pos_emb_height = init_pos_emb_height
+        self.init_pos_emb_width = init_pos_emb_width
+        self.init_pos_emb_time = init_pos_emb_time
+        self.pos_emb_type = pos_emb_type
+        self.vt_num_attention_heads = vt_num_attention_heads
+        self.vt_num_hidden_layers = vt_num_hidden_layers
+        self.vt_hidden_size = vt_hidden_size
+        self.vt_intermediate_size = vt_intermediate_size
+        self.merge_kernel_size = merge_kernel_size
+        self.merge_type = merge_type
+        self._attn_implementation = _attn_implementation
+
+        # MM Projector config
+        self.mm_projector_type = mm_projector_type
+        self.mm_hidden_size = mm_hidden_size if mm_hidden_size is not None else vt_hidden_size
+        self.projector_hidden_act = projector_hidden_act
+        self.projector_ln_eps = projector_ln_eps
+        self.text_hidden_size = text_hidden_size
+
+        # vision tower parameters
+        self.qkv_hidden_size = qkv_hidden_size
+        self.norm_type = norm_type
+        self.attn_bias = attn_bias
+        self.patch_embed_proj_bias = patch_embed_proj_bias
+        self.mlp_type = mlp_type
+        self.linear_bias = linear_bias
+        self.activation_func = activation_func
+        self.pos_emb_interpolation_mode = pos_emb_interpolation_mode
+
+        super().__init__(**kwargs)
+
+
+class KimiK3Config(PretrainedConfig):
+    """Kimi-K3 model configuration.
+
+    Args:
+        text_config (dict | KimiLinearConfig): Configuration for the text model.
+        
+        Vision Tower Parameters (from MoonViT3dConfig):
+            patch_size (int): Patch size for vision tower.
+            init_pos_emb_height (int): Initial position embedding height.
+            init_pos_emb_width (int): Initial position embedding width.
+            init_pos_emb_time (int): Initial position embedding time dimension.
+            pos_emb_type (str): Type of position embedding.
+            vt_num_attention_heads (int): Number of attention heads in vision tower.
+            vt_num_hidden_layers (int): Number of hidden layers in vision tower.
+            vt_hidden_size (int): Hidden size of vision tower.
+            vt_intermediate_size (int): Intermediate size in vision tower FFN.
+            merge_kernel_size (tuple): Kernel size for patch merging.
+            merge_type (str): Type of merge operation.
+            _attn_implementation (str): Attention implementation type.
+        
+        MM Projector Parameters (from MultiModalProjectorConfig):
+            mm_projector_type (str): Type of multimodal projector.
+            mm_hidden_size (int): Hidden size from vision tower (should match vt_hidden_size).
+            projector_hidden_act (str): Activation function for projector.
+            projector_ln_eps (float): Layer norm epsilon for projector.
+        
+        Other Parameters:
+            ignore_index (int): The ignore index for the loss function.
+            media_placeholder_token_id (int): The token ID to use for media placeholders.
+            pad_token_id (int): The token ID to use for padding.
+    """
+
+    model_type = "kimi_k3"
+
+    def __init__(
+        self,
+        text_config: dict | KimiLinearConfig = None,
+        vision_config: dict | KimiK3VisionConfig = None,
+        # Other parameters
+        ignore_index: int = -100,
+        media_placeholder_token_id: int = 163605,
+        pad_token_id: int = 0,
+        **kwargs,
+    ):
+        if isinstance(text_config, dict):
+            text_config = KimiLinearConfig(**text_config)
+        if isinstance(vision_config, dict):
+            vision_config = KimiK3VisionConfig(**vision_config)
+        self.text_config = text_config
+        self.vision_config = vision_config
+        # Other config
+        self.ignore_index = ignore_index
+        self.media_placeholder_token_id = media_placeholder_token_id
+        if getattr(self.text_config, "quantization_config", None) is not None:
+            self.quantization_config = self.text_config.quantization_config
+
+        super().__init__(pad_token_id=pad_token_id, **kwargs)

encoding_k3.py

@@ -0,0 +1,647 @@
+"""Kimi K3 XTML encoding helpers.
+
+This module keeps chat rendering in Python.
+Callers that need token IDs should consume ``EncodeSegment`` objects directly:
+structural markers may be encoded as tiktoken special tokens, while user/tool
+text and attribute values are encoded as ordinary text.
+"""
+
+from __future__ import annotations
+
+import json
+from dataclasses import dataclass
+from typing import Any, Iterable, Optional
+
+OPEN_TOKEN = "<|open|>"
+CLOSE_TOKEN = "<|close|>"
+SEP_TOKEN = "<|sep|>"
+END_OF_MSG_TOKEN = "<|end_of_msg|>"
+IMAGE_PLACEHOLDER = "<|kimi_image_placeholder|>"
+
+_VALID_THINKING_EFFORTS = {"low", "high", "max"}
+
+
+@dataclass(frozen=True)
+class EncodeSegment:
+    text: str
+    allow_special: bool = False
+
+
+class _ImagePromptState:
+    def __init__(self, image_prompts: Optional[list[str]] = None):
+        self.image_prompts = image_prompts
+        self.index = 0
+
+    def next_prompt(self) -> str:
+        if self.image_prompts is None:
+            return IMAGE_PLACEHOLDER
+        if self.index >= len(self.image_prompts):
+            raise ValueError("More image placeholders than image prompts.")
+        prompt = self.image_prompts[self.index]
+        self.index += 1
+        return prompt
+
+    def assert_consumed(self) -> None:
+        if self.image_prompts is None:
+            return
+        if self.index != len(self.image_prompts):
+            raise ValueError(
+                f"image prompt count {len(self.image_prompts)} != "
+                f"consumed placeholder count {self.index}"
+            )
+
+
+def _segment(text: Any, *, allow_special: bool = False) -> list[EncodeSegment]:
+    text = str(text)
+    if not text:
+        return []
+    return [EncodeSegment(text, allow_special=allow_special)]
+
+
+def _control(text: str) -> list[EncodeSegment]:
+    return _segment(text, allow_special=True)
+
+
+def _text(text: Any) -> list[EncodeSegment]:
+    return _segment(text, allow_special=False)
+
+
+def _append_text(
+    segments: list[EncodeSegment],
+    text: Any,
+    image_state: _ImagePromptState,
+) -> None:
+    text = str(text)
+    if text == "":
+        return
+    if image_state.image_prompts is None or IMAGE_PLACEHOLDER not in text:
+        segments.extend(_text(text))
+        return
+
+    parts = text.split(IMAGE_PLACEHOLDER)
+    for i, part in enumerate(parts):
+        segments.extend(_text(part))
+        if i < len(parts) - 1:
+            segments.extend(_segment(image_state.next_prompt(),
+                                     allow_special=True))
+
+
+def _escape_attr_value(value: Any) -> str:
+    return str(value).replace("&", "&amp;").replace('"', "&quot;")
+
+
+def _attr(key: str, value: Any) -> list[EncodeSegment]:
+    return (
+        _text(f" {key}")
+        + _text('="')
+        + _text(_escape_attr_value(value))
+        + _text('"')
+    )
+
+
+def _open_tag(tag: str, attrs: Iterable[tuple[str, Any]] = ()) -> list[EncodeSegment]:
+    segments: list[EncodeSegment] = []
+    segments.extend(_control(OPEN_TOKEN))
+    segments.extend(_text(tag))
+    for key, value in attrs:
+        segments.extend(_attr(key, value))
+    segments.extend(_control(SEP_TOKEN))
+    return segments
+
+def _close_tag(tag: str) -> list[EncodeSegment]:
+    segments: list[EncodeSegment] = []
+    segments.extend(_control(CLOSE_TOKEN))
+    segments.extend(_text(tag))
+    segments.extend(_control(SEP_TOKEN))
+    return segments
+
+
+def _end_of_msg() -> list[EncodeSegment]:
+    return _control(END_OF_MSG_TOKEN)
+
+
+def _json_compact(value: Any) -> str:
+    return json.dumps(value, ensure_ascii=False, separators=(",", ":"))
+
+
+def _is_mapping(value: Any) -> bool:
+    return isinstance(value, dict)
+
+
+def _xtml_type(value: Any) -> str:
+    if isinstance(value, bool):
+        return "boolean"
+    if value is None:
+        return "null"
+    if isinstance(value, (int, float)) and not isinstance(value, bool):
+        return "number"
+    if isinstance(value, str):
+        return "string"
+    if _is_mapping(value):
+        return "object"
+    return "array"
+
+
+def _xtml_value(value: Any) -> str:
+    if isinstance(value, str):
+        return value
+    return json.dumps(value, ensure_ascii=False)
+
+
+def _get_value(obj: Any, key: str, default: Any = None) -> Any:
+    if isinstance(obj, dict):
+        return obj.get(key, default)
+    return getattr(obj, key, default)
+
+
+def extract_response_schema(response_format: Any) -> Any:
+    if response_format is None:
+        return None
+
+    json_schema = _get_value(response_format, "json_schema")
+    if json_schema is None:
+        return None
+
+    if isinstance(json_schema, dict):
+        return json_schema.get(
+            "schema",
+            json_schema.get("json_schema", json_schema),
+        )
+
+    schema = _get_value(json_schema, "schema")
+    if schema is not None:
+        return schema
+
+    schema = _get_value(json_schema, "json_schema")
+    if schema is not None:
+        return schema
+
+    return json_schema
+
+
+def deep_sort_dict(obj: Any) -> Any:
+    if isinstance(obj, dict):
+        return {k: deep_sort_dict(v) for k, v in sorted(obj.items())}
+    if isinstance(obj, list):
+        return [deep_sort_dict(item) for item in obj]
+    return obj
+
+
+def normalize_tool_arguments(arguments: Any) -> tuple[dict[str, Any], Optional[str]]:
+    if arguments is None:
+        return {}, None
+    if isinstance(arguments, dict):
+        return arguments, None
+    if isinstance(arguments, str):
+        if not arguments.strip():
+            return {}, None
+        try:
+            parsed = json.loads(arguments)
+        except json.JSONDecodeError:
+            return {}, arguments
+        if not isinstance(parsed, dict):
+            raise ValueError("Kimi K3 tool call arguments must be a JSON object.")
+        return parsed, None
+    raise TypeError(
+        "Kimi K3 tool call arguments must be a dict or a JSON object string."
+    )
+
+
+def normalize_message(message: Any) -> Any:
+    if not isinstance(message, dict):
+        return message
+
+    normalized = dict(message)
+
+    tools = normalized.get("tools")
+    if tools is not None:
+        normalized["tools"] = deep_sort_dict(tools)
+
+    tool_calls = normalized.get("tool_calls")
+    if not tool_calls:
+        return normalized
+
+    normalized_calls = []
+    for tool_call in tool_calls:
+        if not isinstance(tool_call, dict):
+            normalized_calls.append(tool_call)
+            continue
+
+        tc = dict(tool_call)
+        function = tc.get("function")
+        if isinstance(function, dict):
+            fn = dict(function)
+            arguments, json_block = normalize_tool_arguments(fn.get("arguments"))
+            fn["arguments"] = arguments
+            if json_block is None:
+                fn.pop("_xtml_json_block", None)
+            else:
+                fn["_xtml_json_block"] = json_block
+            tc["function"] = fn
+        else:
+            arguments, json_block = normalize_tool_arguments(tc.get("arguments"))
+            tc["arguments"] = arguments
+            if json_block is None:
+                tc.pop("_xtml_json_block", None)
+            else:
+                tc["_xtml_json_block"] = json_block
+        normalized_calls.append(tc)
+
+    normalized["tool_calls"] = normalized_calls
+    return normalized
+
+
+def normalize_conversation(conversation: Any) -> Any:
+    if not isinstance(conversation, list):
+        return conversation
+
+    def normalize_messages(messages: list[Any]) -> list[Any]:
+        return [normalize_message(message) for message in messages]
+
+    if conversation and isinstance(conversation[0], list):
+        return [normalize_messages(messages) for messages in conversation]
+    return normalize_messages(conversation)
+
+
+def _tool_call_id_index(tool_calls: Any) -> dict:
+    """Map assistant ``tool_calls[].id`` to ``(1-based position, function name)``.
+
+    The position mirrors the chat template's enumeration over ``tool_calls``
+    (every entry advances the position, even an id-less one). Duplicate ids keep
+    their first occurrence.
+    """
+    index: dict = {}
+    if not isinstance(tool_calls, list):
+        return index
+    for position, tool_call in enumerate(tool_calls, start=1):
+        if not isinstance(tool_call, dict):
+            continue
+        call_id = tool_call.get("id")
+        if call_id is None:
+            continue
+        key = str(call_id)
+        if key in index:
+            continue
+        function = tool_call.get("function")
+        name = (
+            function.get("name") if isinstance(function, dict) else tool_call.get("name")
+        )
+        index[key] = (position, name)
+    return index
+
+
+def normalize_xtml_tool_result_messages(messages: list[Any]) -> list[Any]:
+    """Re-sort K3 XTML tool results into assistant ``tool_calls`` order.
+
+    Serving frameworks generally deliver tool results already in call order. A
+    direct Transformers caller, however, may pass OpenAI-style tool messages in any
+    order, so each run of consecutive tool messages is matched against the most
+    recent preceding assistant ``tool_calls`` by opaque ``tool_call_id`` ==
+    ``tool_calls[].id`` (K3 drops the ``func:index`` format requirement) and
+    sorted by the matched 1-based position. The matched call is authoritative,
+    so each matched message's ``tool`` is set to that call's function name --
+    this keeps an explicit (and possibly stale) ``tool``/``name`` from drifting
+    out of sync with the reordered position. ``index`` is still derived from the
+    rendered position by the chat template. A run that cannot be fully matched is
+    left untouched. Re-running is idempotent.
+
+    This function is side-effect free: matched tool messages are shallow-copied
+    before their ``tool``/``name`` is rewritten, and every other message is
+    appended to the output as-is. The input list and its message objects are
+    never mutated.
+    """
+    if not isinstance(messages, list):
+        return messages
+
+    output: list[Any] = []
+    current_index: dict = {}
+    i = 0
+    n = len(messages)
+
+    while i < n:
+        message = messages[i]
+
+        if isinstance(message, dict) and message.get("role") == "assistant":
+            tool_calls = message.get("tool_calls")
+            current_index = _tool_call_id_index(tool_calls) if tool_calls else {}
+            output.append(message)
+            i += 1
+            continue
+
+        if not isinstance(message, dict) or message.get("role") != "tool":
+            output.append(message)
+            i += 1
+            continue
+
+        run: list[tuple] = []  # (position, original_offset, message, name)
+        unresolved = False
+        offset = 0
+        while (
+            i < n and isinstance(messages[i], dict) and messages[i].get("role") == "tool"
+        ):
+            tool_message = messages[i]
+            call_id = tool_message.get("tool_call_id", tool_message.get("id"))
+            matched = current_index.get(str(call_id)) if call_id is not None else None
+            if matched is None:
+                unresolved = True
+                run.append((None, offset, tool_message, None))
+            else:
+                position, name = matched
+                run.append((position, offset, tool_message, name))
+            offset += 1
+            i += 1
+
+        if unresolved:
+            output.extend(item[2] for item in run)
+        else:
+            run.sort(key=lambda item: (item[0], item[1]))
+            for _, _, tool_message, name in run:
+                if name is None:
+                    output.append(tool_message)
+                    continue
+                # The id-matched call is authoritative: align tool (and any
+                # explicit name) so the rendered XTML tool attribute cannot
+                # disagree with the reordered position. Copy first so the
+                # caller's message object is never mutated.
+                resolved = dict(tool_message)
+                resolved["tool"] = name
+                if "name" in resolved:
+                    resolved["name"] = name
+                output.append(resolved)
+
+    return output
+
+
+def is_batched_conversation(conversation: Any) -> bool:
+    return (
+        isinstance(conversation, list)
+        and bool(conversation)
+        and isinstance(conversation[0], list)
+    )
+
+
+def _render_content_segments(
+    content: Any,
+    image_state: _ImagePromptState,
+) -> list[EncodeSegment]:
+    segments: list[EncodeSegment] = []
+    if isinstance(content, str):
+        _append_text(segments, content, image_state)
+    elif content is not None:
+        for part in content:
+            if part["type"] in ["image", "image_url"]:
+                segments.extend(
+                    _segment(image_state.next_prompt(), allow_special=True))
+            else:
+                _append_text(segments, part["text"], image_state)
+    return segments
+
+
+def _internal_system_message(message_type: str, body: str) -> list[EncodeSegment]:
+    segments: list[EncodeSegment] = []
+    segments.extend(_open_tag("message", [("role", "system"), ("type", message_type)]))
+    segments.extend(_text(body.strip()))
+    segments.extend(_close_tag("message"))
+    segments.extend(_end_of_msg())
+    return segments
+
+
+def _render_assistant_segments(
+    message: dict[str, Any],
+    image_state: _ImagePromptState,
+    thinking: bool = True,
+) -> list[EncodeSegment]:
+    segments: list[EncodeSegment] = []
+    # The <think> channel is structural: in thinking mode every assistant
+    # message carries the open/close tags even when there is no reasoning
+    # content to fill in. In non-thinking mode the channel is dropped
+    # entirely.
+    if thinking:
+        reasoning_content = message.get("reasoning_content") or message.get(
+            "reasoning"
+        )
+        segments.extend(_open_tag("think"))
+        if reasoning_content is not None and str(reasoning_content).strip():
+            _append_text(segments, reasoning_content, image_state)
+        segments.extend(_close_tag("think"))
+
+    segments.extend(_open_tag("response"))
+    segments.extend(_render_content_segments(message.get("content"), image_state))
+    segments.extend(_close_tag("response"))
+
+    tool_calls = message.get("tool_calls")
+    if tool_calls:
+        segments.extend(_open_tag("tools"))
+        for index, tool_call in enumerate(tool_calls, start=1):
+            fn = tool_call.get("function", tool_call)
+            segments.extend(
+                _open_tag("call", [("tool", fn["name"]), ("index", index)])
+            )
+            args = fn.get("arguments", {})
+            json_block = fn.get("_xtml_json_block")
+            if json_block is not None:
+                segments.extend(_open_tag("json", [("type", "object")]))
+                _append_text(segments, json_block, image_state)
+                segments.extend(_close_tag("json"))
+            elif _is_mapping(args):
+                for key, value in args.items():
+                    segments.extend(
+                        _open_tag(
+                            "argument",
+                            [("key", key), ("type", _xtml_type(value))],
+                        )
+                    )
+                    _append_text(segments, _xtml_value(value), image_state)
+                    segments.extend(_close_tag("argument"))
+            segments.extend(_close_tag("call"))
+        segments.extend(_close_tag("tools"))
+
+    return segments
+
+
+def _render_tool_declare(tools: Any, *, dynamic: bool = False) -> list[EncodeSegment]:
+    if dynamic:
+        body = (
+            "## New Tools Available\n"
+            "The system dynamically extends the toolset via lazy-loading.\n"
+            "You have access to all existing and extended tools.\n"
+            "Here are the specs for the extended tools.\n\n"
+            "```json\n"
+            f"{_json_compact(tools)}\n"
+            "```"
+        )
+    else:
+        body = (
+            "# Tools\n"
+            "Here are the available tools, described in JSONSchema.\n\n"
+            "```json\n"
+            f"{_json_compact(tools)}\n"
+            "```"
+        )
+    segments: list[EncodeSegment] = []
+    segments.extend(_open_tag("message", [("role", "system"), ("type", "tool-declare")]))
+    segments.extend(_text(body))
+    segments.extend(_close_tag("message"))
+    segments.extend(_end_of_msg())
+    return segments
+
+
+def build_chat_segments(
+    messages: list[Any],
+    tools: Optional[list[dict]] = None,
+    *,
+    add_generation_prompt: bool = True,
+    thinking: bool = True,
+    image_prompts: Optional[list[str]] = None,
+    **kwargs: Any,
+) -> list[EncodeSegment]:
+    # Re-sort tool results by tool_call_id at the lowest layer so every caller
+    # (processor or direct tokenizer) gets correctly ordered XTML. The helper is
+    # side-effect free, so the caller's message objects are left untouched.
+    messages = normalize_xtml_tool_result_messages(messages)
+    messages = normalize_conversation(messages)
+    tools = deep_sort_dict(tools)
+
+    kwargs = dict(kwargs)
+    response_format = kwargs.get("response_format")
+    if "response_schema" not in kwargs:
+        response_schema = extract_response_schema(response_format)
+        if response_schema is not None:
+            kwargs["response_schema"] = response_schema
+    if kwargs.get("response_schema") is not None:
+        kwargs["response_schema"] = deep_sort_dict(kwargs["response_schema"])
+
+    image_state = _ImagePromptState(image_prompts)
+    segments: list[EncodeSegment] = []
+
+    tool_calls = None
+    tool_index = 0
+
+    if tools:
+        segments.extend(_render_tool_declare(tools))
+
+    thinking_effort = kwargs.get("thinking_effort")
+    if thinking and thinking_effort is not None:
+        assert thinking_effort in _VALID_THINKING_EFFORTS, (
+            f"Unsupported thinking_effort={thinking_effort!r}; "
+            f"supported values are {sorted(_VALID_THINKING_EFFORTS)}."
+        )
+    if thinking and thinking_effort in _VALID_THINKING_EFFORTS:
+        segments.extend(
+            _internal_system_message(
+                "thinking-effort",
+                "`thinking_effort` guides on how much to think in your "
+                "thinking channel (not including the response channel), "
+                "supported values include `low`, `medium`, `high`, and `max`.\n"
+                f"Now the system is invoked with `thinking_effort={thinking_effort}`.",
+            )
+        )
+
+    for message_index, message in enumerate(messages):
+        if not isinstance(message, dict):
+            continue
+
+        role = message["role"]
+        if role == "user":
+            attrs = [("role", "user")]
+            if message.get("name"):
+                attrs.append(("name", message["name"]))
+            segments.extend(_open_tag("message", attrs))
+            segments.extend(_render_content_segments(message.get("content"), image_state))
+            segments.extend(_close_tag("message"))
+            segments.extend(_end_of_msg())
+        elif role == "system" and message.get("tools"):
+            segments.extend(_render_tool_declare(message["tools"], dynamic=True))
+        elif role == "system":
+            attrs = [("role", "system")]
+            if message.get("name"):
+                attrs.append(("name", message["name"]))
+            segments.extend(_open_tag("message", attrs))
+            segments.extend(_render_content_segments(message.get("content"), image_state))
+            segments.extend(_close_tag("message"))
+            segments.extend(_end_of_msg())
+        elif role == "tool":
+            tool_index += 1
+            tool_name = message.get("tool", message.get("name"))
+            if (
+                tool_name is None
+                and tool_calls is not None
+                and tool_index <= len(tool_calls)
+            ):
+                tc = tool_calls[tool_index - 1]
+                fn = tc.get("function", tc)
+                tool_name = fn["name"]
+            if tool_name is None:
+                raise ValueError(
+                    "Kimi K3 tool messages need a resolvable tool name: "
+                    "carry `tool`/`name`, or match a preceding assistant "
+                    "tool_call by order."
+                )
+            segments.extend(
+                _open_tag(
+                    "message",
+                    [("role", "tool"), ("tool", tool_name), ("index", tool_index)],
+                )
+            )
+            segments.extend(_render_content_segments(message.get("content"), image_state))
+            segments.extend(_close_tag("message"))
+            segments.extend(_end_of_msg())
+        elif role == "assistant":
+            tool_calls = message.get("tool_calls")
+            tool_index = 0
+            attrs = [("role", "assistant")]
+            if message.get("name"):
+                attrs.append(("name", message["name"]))
+            segments.extend(_open_tag("message", attrs))
+            segments.extend(_render_assistant_segments(message, image_state, thinking))
+            segments.extend(_close_tag("message"))
+            segments.extend(_end_of_msg())
+
+    tool_choice = kwargs.get("tool_choice")
+    if tool_choice == "required":
+        segments.extend(
+            _internal_system_message(
+                "tool-choice",
+                "The system is invoked with `tool_choice=required`.\n"
+                "You MUST call tools in the next message.",
+            )
+        )
+    elif tool_choice == "none":
+        segments.extend(
+            _internal_system_message(
+                "tool-choice",
+                "The system is invoked with `tool_choice=none`.\n"
+                "You MUST NOT call any tools in the next message.",
+            )
+        )
+
+    rf = kwargs.get("response_format")
+    rf_type = _get_value(rf, "type", rf) if isinstance(rf, dict) else rf
+    if rf_type == "json_object":
+        segments.extend(
+            _internal_system_message(
+                "response-format",
+                "The system is invoked with `response_format=json_object`.\n"
+                "Your response must be raw JSON data without markdown code "
+                "blocks (```json) or any additional formatting.",
+            )
+        )
+    elif rf_type == "json_schema":
+        schema = _json_compact(kwargs.get("response_schema"))
+        segments.extend(
+            _internal_system_message(
+                "response-format",
+                "The system is invoked with `response_format=json_schema`.\n"
+                "Your response must be raw JSON data without markdown code "
+                "blocks (```json) or any additional formatting.\n"
+                "The JSON data must match the following schema:\n"
+                f"```json\n{schema}\n```",
+            )
+        )
+
+    if add_generation_prompt:
+        segments.extend(_open_tag("message", [("role", "assistant")]))
+        segments.extend(_open_tag("think" if thinking else "response"))
+
+    image_state.assert_consumed()
+    return segments

generation_config.json

@@ -0,0 +1,4 @@
+{
+  "max_length": 1048576,
+  "eos_token_id": 163586
+}

kimi_k3_processor.py

@@ -0,0 +1,187 @@
+"""Kimi-K3 processor: wraps vision processor + tokenizer into a single interface.
+
+Chat rendering (including XTML tool-result ordering) is handled by the
+tokenizer's Python encoder; this processor adds multimodal media preprocessing.
+"""
+
+from transformers.feature_extraction_utils import BatchFeature
+from transformers.processing_utils import ProcessorMixin
+from transformers.utils import logging
+
+from .media_utils import ensure_media_type
+
+logger = logging.get_logger(__name__)
+
+# ── KimiK3Processor ───────────────────────────────────────────────────
+
+
+class KimiK3Processor(ProcessorMixin):
+    r"""
+    Constructs a KimiK3 processor which wraps a KimiK3 image processor
+    and a tokenizer into a single processor.
+
+    [`KimiK3Processor`] offers all the functionalities of
+    [`KimiK3VisionProcessor`] and [`TikTokenTokenizer`].
+
+    Args:
+        image_processor ([`KimiK3VisionProcessor`], *optional*):
+            The image processor is a required input.
+        tokenizer ([`TikTokenTokenizer`], *optional*):
+            The tokenizer is a required input.
+        chat_template (`str`, *optional*): Kept for ProcessorMixin
+            compatibility. Kimi K3 chat encoding is implemented in Python by
+            the tokenizer.
+    """
+
+    attributes = ["image_processor", "tokenizer"]
+    valid_kwargs = ["chat_template"]
+    image_processor_class = "AutoImageProcessor"
+    tokenizer_class = "AutoTokenizer"
+
+    def __init__(
+        self,
+        image_processor=None,
+        tokenizer=None,
+        chat_template=None,
+        **kwargs,
+    ):
+        super().__init__(image_processor,
+                         tokenizer,
+                         chat_template=chat_template)
+        self.media_processor = image_processor
+        self.image_placeholder = "<|kimi_image_placeholder|>"
+
+    # ── Media preprocessing ────────────────────────────────────────────
+
+    def update_raw_text(self, text: str, image_prompts: list[str]) -> str:
+        # Replace image placeholders
+        image_count = text.count(self.image_placeholder)
+        if image_count > 0:
+            assert image_count == len(image_prompts), (
+                f"image placeholder count {image_count} != "
+                f"image_prompts count {len(image_prompts)}")
+            text_parts = text.split(self.image_placeholder)
+            assert len(text_parts) == len(image_prompts) + 1
+            text = "".join([
+                text_parts[i] + image_prompts[i]
+                for i in range(len(image_prompts))
+            ])
+            text += text_parts[-1]
+
+        return text
+
+    def preprocess_medias(self,
+                          medias: list[dict]) -> tuple[list[dict], list[str]]:
+        """Process media items and generate corresponding prompts.
+
+        Returns:
+            A tuple of (updated_medias, image_prompts).
+        """
+        updated_medias = []
+        image_prompts = []
+        for media in medias:
+            if media['type'] == 'image':
+                updated_medias.append(media)
+                img = ensure_media_type(
+                    media,
+                    transparent_bg_config=self.media_processor.
+                    _transparent_bg_config,
+                    transparent_bg_fill_stage=self.media_processor.
+                    _transparent_bg_fill_stage,
+                )['image']
+                w, h = img.size
+                image_prompts.append(
+                    self.media_processor.make_image_prompt(w, h))
+            else:
+                raise ValueError(f"unsupported media type: {media['type']}")
+        return updated_medias, image_prompts
+
+    # ── Main entry points ──────────────────────────────────────────────
+
+    def __call__(self,
+                 messages: list[dict] = None,
+                 medias: list[dict] = None,
+                 text: str = None,
+                 return_tensors: str = "pt",
+                 **kwargs) -> BatchFeature:
+        """
+        Process multimodal inputs for Kimi-K3 model.
+
+        Args:
+            messages: List of message dicts with 'role' and 'content' fields.
+                     If provided, medias and text will be extracted automatically.
+            medias: Pre-extracted list of media dicts.
+            text: Pre-formatted text string.
+            return_tensors: Format of returned tensors. Default: 'pt'.
+            **kwargs: Additional arguments passed to apply_chat_template.
+
+        Returns:
+            BatchFeature with fields: input_ids, attention_mask,
+            pixel_values, grid_thws.
+        """
+        if messages is None and (medias is None or text is None):
+            raise ValueError(
+                "Provide either 'messages' or both 'medias' and 'text'")
+
+        if medias is not None and text is not None:
+            updated_medias, image_prompts = (self.preprocess_medias(medias))
+            preprocessed = self.media_processor.preprocess(
+                updated_medias, return_tensors=return_tensors)
+            text = self.update_raw_text(text, image_prompts)
+            text_inputs = self.tokenizer(text, return_tensors=return_tensors)
+            return BatchFeature(data={**text_inputs, **preprocessed.data})
+
+        if medias is None:
+            medias = self._extract_medias_from_messages(messages)
+        updated_medias, image_prompts = (self.preprocess_medias(medias))
+        preprocessed = self.media_processor.preprocess(
+            updated_medias, return_tensors=return_tensors)
+
+        if text is None:
+            text_inputs = self.tokenizer.apply_chat_template(
+                messages,
+                tokenize=True,
+                return_tensors=return_tensors,
+                return_dict=True,
+                image_prompts=image_prompts,
+                **kwargs)
+            return BatchFeature(data={**text_inputs, **preprocessed.data})
+
+        text = self.update_raw_text(text, image_prompts)
+        text_inputs = self.tokenizer(text, return_tensors=return_tensors)
+        return BatchFeature(data={**text_inputs, **preprocessed.data})
+
+    @staticmethod
+    def _extract_medias_from_messages(messages: list[dict]) -> list[dict]:
+        """Extract media items from messages in a single pass."""
+        medias = []
+        for msg in messages:
+            if msg['role'] != 'user' or not msg.get('content'):
+                continue
+
+            for content_part in msg['content']:
+                if not isinstance(content_part, dict):
+                    continue
+
+                content_type = content_part.get('type')
+                if content_type in ['image_url', 'image']:
+                    image_data = content_part.get(content_type)
+                    assert image_data is not None, f"image data is missing for content part: {content_part}"
+                    medias.append({
+                        'type': 'image',
+                        'image': image_data,
+                    })
+        return medias
+
+    def apply_chat_template(self, messages, **kwargs):
+        return self.tokenizer.apply_chat_template(messages, **kwargs)
+
+    def batch_decode(self, *args, **kwargs):
+        return self.tokenizer.batch_decode(*args, **kwargs)
+
+    def decode(self, *args, **kwargs):
+        return self.tokenizer.decode(*args, **kwargs)
+
+    @property
+    def model_input_names(self):
+        return ['input_ids', 'attention_mask', 'pixel_values', 'grid_thws']

kimi_k3_vision_processing.py

@@ -0,0 +1,179 @@
+"""Image processor class for Kimi-K3.
+"""
+
+import json
+from typing import Any, Dict, Optional, Union
+
+import numpy as np
+import torch
+from PIL import Image
+from transformers.image_processing_utils import (BaseImageProcessor,
+                                                 BatchFeature)
+from transformers.utils import TensorType
+
+from .media_utils import (MediaInput, TransparentBgConfig, _to_tensor,
+                          ensure_media_type, image_to_np, navit_patchify,
+                          navit_resize_image, normalize)
+
+
+class KimiK3VisionProcessor(BaseImageProcessor):
+    model_type = "kimi_k3"
+
+    def __init__(
+        self,
+        media_proc_cfg: dict,
+        **kwargs,
+    ):
+        super().__init__(**kwargs)
+        self.media_proc_cfg = media_proc_cfg
+
+    @property
+    def _transparent_bg_config(self) -> Optional[TransparentBgConfig]:
+        cfg = self.media_proc_cfg.get("transparent_bg_config")
+        if cfg is None:
+            return None
+        if isinstance(cfg, TransparentBgConfig):
+            return cfg
+        return TransparentBgConfig(**cfg)
+
+    @property
+    def _transparent_bg_fill_stage(self) -> str:
+        return self.media_proc_cfg.get("transparent_bg_fill_stage",
+                                       "before_resize")
+
+    def media_tokens_calculator(self, media: MediaInput):
+        media = ensure_media_type(
+            media,
+            transparent_bg_config=self._transparent_bg_config,
+            transparent_bg_fill_stage=self._transparent_bg_fill_stage,
+        )
+        ret = self.get_resize_config(media)
+        return ret['num_tokens']
+
+    @classmethod
+    def make_image_prompt(cls, width: int, height: int) -> str:
+        """Build the K3 image placeholder with resolution info."""
+        return (f"<|media_begin|>image {width}x{height}"
+                f"<|media_content|><|media_pad|><|media_end|>")
+
+    def get_resize_config(self, media_input: MediaInput) -> dict:
+        if media_input['type'] == 'image':
+            w, h = media_input['image'].size
+            ret = navit_resize_image(
+                w, h, self.media_proc_cfg['patch_size'],
+                self.media_proc_cfg['merge_kernel_size'],
+                self.media_proc_cfg['in_patch_limit'],
+                self.media_proc_cfg['patch_limit_on_one_side'],
+                self.media_proc_cfg['fixed_output_tokens'])
+            return ret
+        else:
+            raise ValueError("Unsupported type: {}".format(
+                media_input['type']))
+
+    def resize_image(self, image: Image.Image, new_width: int, new_height: int,
+                     pad_width: int, pad_height: int) -> np.ndarray:
+        image_np = image_to_np(
+            image,
+            (new_width, new_height),
+            "resize",
+            transparent_bg_config=self._transparent_bg_config,
+            transparent_bg_fill_stage=self._transparent_bg_fill_stage,
+        )
+        image_np = np.pad(
+            image_np,
+            ((0, pad_height), (0, pad_width), (0, 0)),
+            mode="constant",
+            constant_values=0,
+        )
+        return image_np
+
+    def preprocess(
+        self,
+        medias: list[MediaInput],
+        return_tensors: Optional[Union[str, TensorType]] = None,
+    ) -> BatchFeature:
+        """
+        Preprocess a atom vision input (images) into model-ready tensors.
+
+        Args:
+            medias: List of MediaInput.
+            return_tensors: Desired output format ('pt', 'np', 'tf', or None).
+
+        Returns:
+            BatchFeature containing 'pixel_values' and 'grid_thws' tensors.
+        """
+        if not isinstance(medias, list):
+            medias = [medias]
+        if medias:
+            pixel_values = []
+            for item in medias:
+                item = ensure_media_type(
+                    item,
+                    transparent_bg_config=self._transparent_bg_config,
+                    transparent_bg_fill_stage=self._transparent_bg_fill_stage,
+                )
+                resize_config = self.get_resize_config(item)
+                new_width, new_height, pad_width, pad_height = resize_config[
+                    'new_width'], resize_config['new_height'], resize_config[
+                        'pad_width'], resize_config['pad_height']
+                if item['type'] == 'image':
+                    image = item['image']
+                    image_np = self.resize_image(image, new_width, new_height,
+                                                 pad_width, pad_height)
+                    pixel_values.append(np.expand_dims(image_np, axis=0))
+                else:
+                    raise ValueError("Unsupported type: {}".format(
+                        item['type']))
+            normalized_pixel_values = []
+            image_std_inv = 1.0 / np.array(self.media_proc_cfg['image_std'])
+            image_mean = np.array(self.media_proc_cfg['image_mean'])
+            for pixels in pixel_values:
+                pixels = normalize(pixels, image_mean, image_std_inv)
+                pixels_and_thw = navit_patchify(
+                    pixels,
+                    self.media_proc_cfg['patch_size'],
+                )
+                normalized_pixel_values.append(pixels_and_thw)
+
+            pixel_values = torch.cat([
+                _to_tensor(pixel_value['pixel_values'])
+                for pixel_value in normalized_pixel_values
+            ])
+            grid_thws = torch.cat([
+                _to_tensor(pixel_value['grid_thw'],
+                           dtype=torch.int64).unsqueeze(0)
+                for pixel_value in normalized_pixel_values
+            ])
+
+            data = {
+                'pixel_values': pixel_values,
+                'grid_thws': grid_thws,
+            }
+
+        else:
+            data = {}
+
+        return BatchFeature(data=data, tensor_type=return_tensors)
+
+    def __repr__(self):
+        return f"KimiK3VisionProcessor(media_proc_cfg={self.media_proc_cfg})"
+
+    def to_dict(self) -> Dict[str, Any]:
+        output = super().to_dict()
+        output["media_proc_cfg"] = self.media_proc_cfg
+        if "media_processor" in output:
+            del output["media_processor"]
+        return output
+
+    @classmethod
+    def from_dict(cls, config_dict: Dict[str, Any], **kwargs):
+        config = config_dict.copy()
+        media_proc_cfg = config.pop("media_proc_cfg", {})
+        return cls(media_proc_cfg=media_proc_cfg, **config, **kwargs)
+
+    def to_json_string(self):
+        dictionary = self.to_dict()
+        for key, value in dictionary.items():
+            if hasattr(value, 'tolist'):
+                dictionary[key] = value.tolist()
+        return json.dumps(dictionary, indent=2, sort_keys=True) + "\n"

media_utils.py

@@ -0,0 +1,376 @@
+import base64
+import functools
+import io
+import math
+from dataclasses import dataclass
+from typing import Literal, TypedDict
+
+import numpy as np
+from PIL import Image
+
+
+class ImageInput(TypedDict):
+    type: Literal['image']
+    image: Image.Image
+
+
+MediaInput = ImageInput
+
+
+@dataclass
+class TransparentBgConfig:
+    """The config of the transparent background."""
+
+    pattern: Literal["white", "black", "gray", "chessboard"] = "black"
+    """The pattern of the transparent background."""
+
+    chessboard_square_size: int = 16
+    """The size of the squares in the chessboard background."""
+
+    chessboard_square_on_top_left: bool = True
+    """Whether to start the chessboard with a white square on the top left."""
+
+    chessboard_white_value: int = 255
+    """The value of the white pixels in the background."""
+
+    chessboard_gray_value: int = 200
+    """The value of the gray pixels in the background."""
+
+
+@functools.lru_cache(maxsize=256)
+def _create_chessboard_background(
+    height: int,
+    width: int,
+    square_size: int,
+    square_on_top_left: bool,
+    white_value: int,
+    gray_value: int,
+) -> np.ndarray:
+    """Create a chessboard background."""
+    bg = np.ones((height, width, 3), dtype=np.uint8) * white_value
+    for y in range(0, height, square_size):
+        for x in range(0, width, square_size):
+            if (y // square_size + x // square_size) % 2 == (
+                    1 if square_on_top_left else 0):
+                bg[y:y + square_size, x:x + square_size] = gray_value
+    return bg
+
+
+def fill_transparent_bg_with(
+    image: Image.Image,
+    transparent_bg_config: TransparentBgConfig | None = None,
+) -> Image.Image:
+    """Composite a (possibly) transparent image onto a configured background.
+
+    When ``transparent_bg_config`` is ``None``, the image is simply converted
+    to RGB (preserving the historical behavior). Otherwise the alpha channel
+    is alpha-composited over a background generated according to the config.
+    """
+    if transparent_bg_config is None:
+        return image.convert("RGB")
+
+    if image.mode == "RGB":
+        return image
+
+    has_alpha = "A" in image.getbands() or "transparency" in image.info
+    if not has_alpha:
+        return image.convert("RGB")
+
+    img = np.array(image.convert("RGBA"))
+    height, width = img.shape[:2]
+    bg_pattern = transparent_bg_config.pattern
+    if bg_pattern == "white":
+        bg = np.full((height, width, 3), 255, dtype=np.uint8)
+    elif bg_pattern == "black":
+        bg = np.zeros((height, width, 3), dtype=np.uint8)
+    elif bg_pattern == "gray":
+        bg = np.full((height, width, 3), 128, dtype=np.uint8)
+    elif bg_pattern == "chessboard":
+        bg = _create_chessboard_background(
+            height,
+            width,
+            transparent_bg_config.chessboard_square_size,
+            transparent_bg_config.chessboard_square_on_top_left,
+            transparent_bg_config.chessboard_white_value,
+            transparent_bg_config.chessboard_gray_value,
+        )
+    else:
+        raise ValueError(f"Invalid background pattern: {bg_pattern}")
+
+    alpha = img[:, :, 3]
+    img_rgb = img[:, :, :3]
+    alpha_normalized = alpha.astype(np.float32) / 255.0
+    alpha_3d = np.stack([alpha_normalized] * 3, axis=2)
+    result = alpha_3d * img_rgb + (1 - alpha_3d) * bg
+    result = result.astype(np.uint8)
+    return Image.fromarray(result)
+
+
+def navit_resize_image(
+    width: int,
+    height: int,
+    patch_size: int,
+    merge_kernel_size: int,
+    in_patch_limit: int,
+    patch_limit_on_one_side: int,
+    fixed_output_tokens: int | None,
+):
+    # Apply the patch limits.
+    s1 = math.sqrt(
+        in_patch_limit /
+        (max(1.0, width // patch_size) * max(1.0, height // patch_size)))
+    s2 = patch_limit_on_one_side * patch_size / width
+    s3 = patch_limit_on_one_side * patch_size / height
+    scale = min(1.0, s1, s2, s3)
+    new_w, new_h = max(1, int(width * scale)), max(1, int(height * scale))
+    new_w = min(new_w, patch_limit_on_one_side * patch_size)
+    new_h = min(new_h, patch_limit_on_one_side * patch_size)
+
+    # Calculate the padding to make the height and width divisible by the merge kernel size and patch size.
+    factor = merge_kernel_size * patch_size
+
+    pad_height = (factor - new_h % factor) % factor
+    pad_width = (factor - new_w % factor) % factor
+
+    if fixed_output_tokens is not None:
+        num_tokens = fixed_output_tokens
+    else:
+        # Calculate new dimensions after padding and patching
+        token_height = (new_h + pad_height) // factor
+        token_width = (new_w + pad_width) // factor
+
+        assert token_height * merge_kernel_size <= patch_limit_on_one_side, (
+            f"token_height {token_height} * merge_kernel_size {merge_kernel_size} > patch_limit_on_one_side {patch_limit_on_one_side}"
+        )
+        assert token_width * merge_kernel_size <= patch_limit_on_one_side, (
+            f"token_width {token_width} * merge_kernel_size {merge_kernel_size} > patch_limit_on_one_side {patch_limit_on_one_side}"
+        )
+
+        num_tokens = token_height * token_width
+    return {
+        "num_tokens": num_tokens,
+        "new_width": new_w,
+        "new_height": new_h,
+        "pad_width": pad_width,
+        "pad_height": pad_height,
+        "sampled_nframes": 1,
+    }
+
+
+def _to_pil(
+    data: str | bytes | Image.Image,
+    transparent_bg_config: TransparentBgConfig | None = None,
+    to_rgb: bool = True,
+) -> Image.Image:
+    """Load an image and (optionally) composite its transparent background.
+
+    Args:
+        data: A PIL Image, a base64 ``data:`` URL, a file path, or raw bytes.
+        transparent_bg_config: The config used to fill the transparent
+            background. ``None`` keeps the historical behavior of converting
+            to RGB without compositing.
+        to_rgb: If ``False`` the image is returned as-is (the
+            ``transparent_bg_config`` is ignored). The caller is then
+            expected to call :func:`fill_transparent_bg_with` later — e.g.
+            after a resize.
+    """
+    if isinstance(data, Image.Image):
+        image = data
+    elif isinstance(data, str):
+        if data.startswith("data:"):
+            raw_base64 = data.split(",")[1]
+            image = Image.open(io.BytesIO(base64.b64decode(raw_base64)))
+        else:
+            image = Image.open(data)
+    elif isinstance(data, bytes):
+        image = Image.open(io.BytesIO(data))
+    else:
+        raise ValueError(f"Unsupported data type: {type(data)}")
+
+    if not to_rgb:
+        return image
+
+    return fill_transparent_bg_with(image, transparent_bg_config)
+
+
+def ensure_media_type(
+    media: MediaInput,
+    transparent_bg_config: TransparentBgConfig | None = None,
+    transparent_bg_fill_stage: Literal["before_resize",
+                                       "after_resize"] = "before_resize",
+) -> MediaInput:
+    if media['type'] == 'image':
+        media['image'] = _to_pil(
+            media['image'],
+            transparent_bg_config=transparent_bg_config,
+            to_rgb=transparent_bg_fill_stage == "before_resize",
+        )
+        return media
+    else:
+        raise ValueError(f"Unsupported media type: {media['type']}")
+
+
+def image_to_np(
+    image: Image.Image,
+    resize_to: tuple[int, int] | None = None,
+    mode: str = "resize",
+    raise_error_for_ill_resize: bool = True,
+    transparent_bg_config: TransparentBgConfig | None = None,
+    transparent_bg_fill_stage: Literal["before_resize",
+                                       "after_resize"] = "before_resize",
+) -> np.ndarray:
+    """Convert an image to a numpy array.
+
+    Args:
+        content: The image to convert.
+        resize_to: The size to resize the image to.
+        mode: The mode to resize the image to.
+        raise_error_for_ill_resize: Whether to raise an error for ill-sized resize.
+        transparent_bg_config: The config of the transparent background. Only
+            used when ``transparent_bg_fill_stage == "after_resize"`` (the
+            caller is responsible for filling before resize otherwise).
+        transparent_bg_fill_stage: When to composite the transparent
+            background — before or after the resize step.
+
+    Returns:
+        A numpy array.
+    """
+    assert isinstance(image, Image.Image), "image must be a PIL Image"
+    if resize_to is not None:
+        if mode == "resize":
+            image = image.resize(resize_to, resample=Image.Resampling.BICUBIC)
+            if transparent_bg_fill_stage == "after_resize":
+                image = fill_transparent_bg_with(image, transparent_bg_config)
+
+        elif mode == "rescale_and_pad_to_center":
+            scale = min(resize_to[0] / image.width,
+                        resize_to[1] / image.height, 1.0)
+            new_width = round(image.width * scale)
+            new_height = round(image.height * scale)
+            if new_width == 0 or new_height == 0:
+                if raise_error_for_ill_resize:
+                    raise ValueError(
+                        f"Invalid resize to: {resize_to}, from image size: {image.size}"
+                    )
+                else:
+                    return np.zeros((resize_to[1], resize_to[0], 3),
+                                    dtype=np.uint8)
+
+            image = image.resize((new_width, new_height),
+                                 resample=Image.Resampling.BICUBIC)
+            if transparent_bg_fill_stage == "after_resize":
+                image = fill_transparent_bg_with(image, transparent_bg_config)
+            padding_left = (resize_to[0] - new_width) // 2
+            padding_right = resize_to[0] - new_width - padding_left
+            padding_top = (resize_to[1] - new_height) // 2
+            padding_bottom = resize_to[1] - new_height - padding_top
+            image = np.asarray(image)
+            image = np.pad(
+                image,
+                ((padding_top, padding_bottom), (padding_left, padding_right),
+                 (0, 0)),
+                mode="constant",
+                constant_values=0,
+            )
+            assert image.shape == (resize_to[1], resize_to[0], 3)
+
+        elif mode == "rescale_and_pad_to_rightbottom":
+            scale = min(resize_to[0] / image.width,
+                        resize_to[1] / image.height, 1.0)
+            new_width = round(image.width * scale)
+            new_height = round(image.height * scale)
+            if new_width == 0 or new_height == 0:
+                if raise_error_for_ill_resize:
+                    raise ValueError(
+                        f"Invalid resize to: {resize_to}, from image size: {image.size}"
+                    )
+                else:
+                    return np.zeros((resize_to[1], resize_to[0], 3),
+                                    dtype=np.uint8)
+
+            image = image.resize((new_width, new_height),
+                                 resample=Image.Resampling.BICUBIC)
+            if transparent_bg_fill_stage == "after_resize":
+                image = fill_transparent_bg_with(image, transparent_bg_config)
+            padding_right = resize_to[0] - new_width
+            padding_bottom = resize_to[1] - new_height
+            image = np.asarray(image)
+            image = np.pad(
+                image,
+                ((0, padding_bottom), (0, padding_right), (0, 0)),
+                mode="constant",
+                constant_values=0,
+            )
+            assert image.shape == (resize_to[1], resize_to[0], 3)
+
+        else:
+            raise ValueError(f"Invalid mode: {mode}")
+
+    if isinstance(image, Image.Image):
+        return np.asarray(image)
+    else:
+        return image
+
+
+def navit_patchify(pixel_values: np.ndarray,
+                   patch_size: int) -> dict[str, np.ndarray]:
+    """Reshape the pixel values to a navit shape.
+
+    Args:
+        pixel_values: np.ndarray, shape (t, h, w, c)
+        patch_size: int
+
+    Returns:
+        dict[str, np.ndarray]
+        - patches: np.ndarray, shape (t * h//patch_size * w//patch_size, c, patch_size, patch_size)
+        - grid_thw: np.ndarray, (t, h//patch_size, w//patch_size)
+    """
+    T, H, W, C = pixel_values.shape
+    assert C == 3, "pixel_values must have 3 channels"
+
+    patches = pixel_values.reshape(T, H // patch_size, patch_size,
+                                   W // patch_size, patch_size, C)
+    # (T, H//patch_size, W//patch_size, C, patch_size, patch_size)
+    patches = patches.transpose(0, 1, 3, 5, 2, 4)
+    patches = patches.reshape(-1, C, patch_size, patch_size)
+    grid_thw = np.array([T, H // patch_size, W // patch_size])
+    return {"pixel_values": patches, "grid_thw": grid_thw}
+
+
+def normalize(x: np.ndarray,
+              mean,
+              std_inv,
+              pixels_dtype: np.dtype = np.float32) -> np.ndarray:
+    """Normalize the image.
+
+    Args:
+        x: The image to normalize. The shape is (..., 3). The dtype is uint8. The range is [0, 255].
+        mean: The mean of the image.
+        std_inv: The inverse of the std of the image.
+        pixels_dtype: The dtype of the image.
+    Returns:
+        The normalized image. The shape is (..., 3). The dtype is determined by the pixels_dtype.
+    """
+    x = (x / 255.0).astype(pixels_dtype)
+    x -= mean
+    x *= std_inv
+    return x
+
+
+def _to_tensor(data, **kwargs):
+    import torch
+
+    if isinstance(data, np.ndarray):
+        return torch.from_numpy(data).to(**kwargs)
+    elif isinstance(data, torch.Tensor):
+        return data.to(**kwargs)
+    elif isinstance(data, list):
+        return [_to_tensor(item, **kwargs) for item in data]
+    elif isinstance(data, tuple):
+        return tuple(_to_tensor(item, **kwargs) for item in data)
+    elif isinstance(data, dict):
+        return {k: _to_tensor(v, **kwargs) for k, v in data.items()}
+    elif data is None:
+        return None
+    else:
+        raise ValueError(f"Unsupported data type: {type(data)}")

model-00001-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:975584c00f85a95fce8ae0f840af8cef69c2ef4db00d34cab3e2cbdfc60f6e51
+size 2341216112

model-00002-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:26a3284e1d2cb567934ebef002e6a1813551d646739e8bcb1e9e3fe7f878e0f5
+size 16990911504

model-00003-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e54af9de4c554956082364010f732443bcd5097390f0121a33fb35e37280b5a9
+size 16990911504

model-00004-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5955fd8feda89b1af8400c25e885e7177d47edff155f54b318beb8dd1cec5c05
+size 16567501776

model-00005-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d60d68ad0381ffd2d8716d5991a65993cc81837455bf05d1c76ea3a75662e268
+size 16990911504

model-00006-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b1d4805767471a9721cd087d2047843ab9262d4f9bbe0d4a306e72c07179f939
+size 16990911504

model-00007-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb1120fef34c0416e73a3259ae6114b9a82b6081871b368a199e0b265cac7df7
+size 16990911504

model-00008-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2318dda54fc1985b63d6068de54fbe312ac87317c902666c9075aa5e10e8a1c9
+size 16567501776

model-00009-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b66cdde34f5130cf4f99ec35b16f1f6d33bf0ec946cf2a5612004f5d399f998
+size 16990911504

model-00010-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d34f55f7b734a1eaecc986fc7f92bcaa1b7d18fad90a33befd3fae4f1b2771a9
+size 16990911504

model-00011-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1738856a4cca2e356510892de4accf3c3f94b9f598b0394a87e7951740ba4a21
+size 16990916912

model-00012-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6b9bef38415f509898fa7b5c04d20c6fee84b0e02c05d74259d201fe6f3868b
+size 16567507176

model-00013-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3cbf43d56d9c80be0f848a4866e290741c697e09d885dc2351b0a11de514d9c8
+size 16990916912

model-00014-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce5f343f07d408c64fb247bbe8691edb84d54e3976d3ff586ebcf5929fd143ad
+size 16990916912

model-00015-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b55caa8013348498525eef2b1416a657f0c28264ada9b245df19d2eea168b4c1
+size 16990916912

model-00016-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5a63e63ced659c1996b02ccabdea977546a39ecce025af1e54552de3bfc0489e
+size 16567507176

model-00017-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:622bfa605205f4ef65e44dac07246bea79e18123ae8b84ab33eec89b6364c3a9
+size 16990916912

model-00018-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:838b265ebb86a1320e589685779125d224663160823873b9e5930cf63fafe796
+size 16990916912

model-00019-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:599a8ecf88f21d4c91295b4160e7564e501807535e522195ee45050a04f17ded
+size 16990916912

model-00020-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e01b61ab76655ca0c5bd6645303a00c05fb58da94e00f978a9ef1e8114a6f0f
+size 16567507176

model-00021-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1944265bd024ea6caa23796326e420260a801e5c52e6253155f22ce97e9578d0
+size 16990916912

model-00022-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d32d3e3c8da0764f73830289c6a22224f755c0cf058df0262fb0f83b3777029
+size 16990916912

model-00023-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c291f2ec15788cb24cdf791c6d68dbf8803b26d01415cff3fab367cba581a169
+size 16990916912

model-00024-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:278855ab81d42337e2a1857f136d740911fe3f5b1813dc21ce2cb03dc5c3cde3
+size 16567507176

model-00025-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6e3ae6ad868f5b38fa7ddface520796ea251568b07447deee3d164a0a32d862b
+size 16990916912

model-00026-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cdd79fb52c7a3350926c4b947ae82cccd0af17fa3fb537c5694a4cedf2b59811
+size 16990916912

model-00027-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1d974b40c4aeb5cc0adeba6f9fbad25fc99f8cd39f9dd3513aeb2fb30a9c4188
+size 16990916912

model-00028-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1bec58e89ab7835d3ca40e581cc6f047cb93de47fda804eaea4e2407e01c3ed3
+size 16567507176

model-00029-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c3bf0e738aa5fbd6ca3ecad982aba8d09efffd58b40cd8fb62d02fe0492f5ceb
+size 16990916912

model-00030-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b996410482ad3c21700ed35d3f2b36282540db92b5631a31aeda1c5e90d6343
+size 16990916912

model-00031-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c26689540a2471fe690c441a6cf9bec8f00471930e3131962206fd992e0d377c
+size 16990916912

model-00032-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7dc9e726d46d5c7c2c35af2eea4860cc89dd38a52b60591358a5ca7899c5159
+size 16567507176

model-00033-of-000096.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:615afa33b69cbaa42122bb3bcce92e40d6de86713a301c43c5d1bc2ac4342c1b
+size 16990916912