diff --git a/README.md b/README.md
index 31b3d8eb..19684b80 100644
--- a/README.md
+++ b/README.md
@@ -17,6 +17,7 @@ KTransformers is a research project focused on efficient inference and fine-tuni
 
 ## 🔥 Updates
 
+* **Jan 27, 2026**: Kimi-K2.5 Day0 Support! ([Tutorial](./doc/en/Kimi-K2.5.md)) ([SFT Tutorial](./doc/en/SFT_Installation_Guide_KimiK2.5.md))
 * **Jan 22, 2026**: Support [CPU-GPU Expert Scheduling](./doc/en/kt-kernel/experts-sched-Tutorial.md), [Native BF16 and FP8 per channel Precision](./doc/en/kt-kernel/Native-Precision-Tutorial.md) and [AutoDL unified fine-tuning and inference](./doc/zh/【云端低价训推】%20KTransformers%2BAutoDL%2BLlamaFactory：随用随租的低成本超大模型「微调%2B推理」一体化流程.pdf)
 * **Dec 24, 2025**: Support Native MiniMax-M2.1 inference. ([Tutorial](./doc/en/kt-kernel/MiniMax-M2.1-Tutorial.md))
 * **Dec 22, 2025**: Support RL-DPO fine-tuning with LLaMA-Factory. ([Tutorial](./doc/en/SFT/DPO_tutorial.md))
diff --git a/doc/en/Kimi-K2.5.md b/doc/en/Kimi-K2.5.md
new file mode 100644
index 00000000..e75c22d3
--- /dev/null
+++ b/doc/en/Kimi-K2.5.md
@@ -0,0 +1,154 @@
+# Running Kimi-K2.5 with SGLang and KT-Kernel
+
+This tutorial demonstrates how to run Kimi-K2.5 model inference using SGLang integrated with KT-Kernel for CPU-GPU heterogeneous inference. This setup enables efficient deployment of large MoE models by offloading experts to CPU.
+
+## Table of Contents
+
+- [Hardware Requirements](#hardware-requirements)
+- [Prerequisites](#prerequisites)
+- [Step 1: Download Model Weights](#step-1-download-model-weights)
+- [Step 2: Launch SGLang Server](#step-2-launch-sglang-server)
+- [Step 3: Send Inference Requests](#step-3-send-inference-requests)
+
+## Hardware Requirements
+
+**Minimum Configuration:**
+- **GPU**: NVIDIA RTX 2x4090 48GB (or equivalent with at least total 48GB VRAM available)
+- **CPU**: x86 CPU with AVX512F support (e.g., Intel Sapphire Rapids)
+- **RAM**: At least 600GB system memory
+- **Storage**: ~600GB for model weights (native INT4 weight, same weight folder for CPU and GPU)
+
+## Prerequisites
+
+Before starting, ensure you have:
+
+1. **KT-Kernel installed**:
+
+   Note: Latest KTransformers' EPLB feature for Kimi-K2.5 will be supported soon.
+
+```
+git clone https://github.com/kvcache-ai/ktransformers.git
+git checkout kimi_k2.5
+git submodule update --init --recursive
+cd kt-kernel && ./install.sh
+```
+
+2. **SGLang installed** - Follow [SGLang integration steps](./kt-kernel_intro.md#integration-with-sglang)
+
+Note: Currently, please clone our custom SGLang repository:
+
+```
+git clone https://github.com/kvcache-ai/sglang.git
+git checkout kimi_k2.5
+cd sglang && pip install -e "python[all]"
+// maybe need to reinstall cudnn according to the issue when launching SGLang
+pip install nvidia-cudnn-cu12==9.16.0.29
+```
+
+3. **CUDA toolkit** - Compatible with your GPU (CUDA 12.8+ recommended)
+4. **Hugging Face CLI** - For downloading models:
+   
+   ```bash
+   pip install huggingface-hub
+   ```
+
+## Step 1: Download Model Weights
+
+```bash
+# Create a directory for models
+mkdir -p /path/to/models
+cd /path/to/models
+
+# Download Kimi-K2.5 (RAW-INT4 for both CPU and GPU)
+huggingface-cli download moonshotai/Kimi-K2.5 \
+  --local-dir /path/to/kimi-k2.5
+```
+
+**Note:** Replace `/path/to/models` with your actual storage path throughout this tutorial.
+
+## Step 2: Launch SGLang Server
+
+Start the SGLang server with KT-Kernel integration for CPU-GPU heterogeneous inference.
+
+
+### Launch Command (4x RTX 4090 Example)
+
+```bash
+python -m sglang.launch_server \
+  --host 0.0.0.0 \
+  --port 31245 \
+  --model /path/to/kimi-k2.5 \
+  --kt-weight-path /path/to/kimi-k2.5 \
+  --kt-cpuinfer 96 \
+  --kt-threadpool-count 2 \
+  --kt-num-gpu-experts 30 \
+  --kt-method RAWINT4 \
+  --kt-gpu-prefill-token-threshold 400 \
+  --trust-remote-code \
+  --mem-fraction-static 0.94 \
+  --served-model-name Kimi-K2.5 \
+  --enable-mixed-chunk \
+  --tensor-parallel-size 4 \
+  --enable-p2p-check \
+  --disable-shared-experts-fusion \
+  --chunked-prefill-size 32658 \
+  --max-total-tokens 50000 \
+  --attention-backend flashinfer
+```
+
+It takes about 2~3 minutes to start the server.
+
+See [KT-Kernel Parameters](https://github.com/kvcache-ai/ktransformers/tree/main/kt-kernel#kt-kernel-parameters) for detailed parameter tuning guidelines.
+
+## Step 3: Send Inference Requests
+
+Once the server is running, you can send inference requests using the OpenAI-compatible API.
+
+### Basic Chat Completion Request
+
+```bash
+curl -s http://localhost:31245/v1/chat/completions \
+  -H "Content-Type: application/json" \
+  -d '{
+    "model": "Kimi-K2.5",
+    "stream": false,
+    "messages": [
+      {"role": "user", "content": "hi, who are you?"}
+    ]
+  }'
+```
+
+### Example Response
+
+```json
+{
+    "id": "2a4e83f8a79b4b57b103b0f298fbaa7d",
+    "object": "chat.completion",
+    "created": 1769333912,
+    "model": "Kimi-K2.5",
+    "choices": [
+        {
+            "index": 0,
+            "message": {
+                "role": "assistant",
+                "content": " The user is asking \"hi, who are you?\" which is a simple greeting and identity question. I need to respond appropriately by introducing myself clearly and concisely.\n\nI am Kimi, a large language model trained by Moonshot AI. I should state my name, my nature (AI assistant), and my developer (Moonshot AI). I should keep it friendly and helpful.\n\nKey points to include:\n- Greet them back (\"hi\" or \"hello\")\n- State my name: Kimi\n- State what I am: an AI assistant/language model\n- Mention my developer: Moonshot AI\n- Briefly describe my purpose: to help answer questions, provide information, and assist with various tasks\n- Keep it concise but informative\n- Use a friendly, professional tone\n\nI should avoid overly technical jargon while being accurate. The response should be welcoming and set the stage for further interaction.\n\nPossible response:\n\"Hi! I'm Kimi, an AI assistant created by Moonshot AI. I'm designed to help answer questions, provide information, and assist with a wide range of tasks. How can I help you today?\"\n\nThis covers all the necessary points and invites the user to continue the conversation. </think> Hi! I'm Kimi, an AI assistant created by Moonshot AI. I'm designed to help answer questions, provide information, and assist with a wide range of tasks. How can I help you today?",
+                "reasoning_content": null,
+                "tool_calls": null
+            },
+            "logprobs": null,
+            "finish_reason": "stop",
+            "matched_stop": 163586
+        }
+    ],
+    "usage": {
+        "prompt_tokens": 32,
+        "total_tokens": 317,
+        "completion_tokens": 285,
+        "prompt_tokens_details": null,
+        "reasoning_tokens": 0
+    },
+    "metadata": {
+        "weight_version": "default"
+    }
+}
+```
diff --git a/doc/en/SFT_Installation_Guide_KimiK2.5.md b/doc/en/SFT_Installation_Guide_KimiK2.5.md
new file mode 100644
index 00000000..0b90d0ba
--- /dev/null
+++ b/doc/en/SFT_Installation_Guide_KimiK2.5.md
@@ -0,0 +1,222 @@
+# Kimi-K2.5 LoRA SFT Tutorial
+
+This tutorial demonstrates how to perform **LoRA Supervised Fine-Tuning (SFT)** on **Kimi-K2.5** using **LlamaFactory** with **KTransformers** as the backend, and then serve the fine-tuned model using **SGLang**.
+
+The workflow is:
+
+```txt
+KTransformers + LlamaFactory LoRA SFT → (Optional) LlamaFactory Verification → SGLang Serving
+```
+
+## Table of Contents
+
+- [Hardware Requirements](https://chatgpt.com/c/6975bb7f-52e0-839c-a727-ec4b5d6723b5#hardware-requirements)
+- [Prerequisites](https://chatgpt.com/c/6975bb7f-52e0-839c-a727-ec4b5d6723b5#prerequisites)
+- [Step 0: Environment Setup (Method 1: Source Install)](https://chatgpt.com/c/6975bb7f-52e0-839c-a727-ec4b5d6723b5#step-0-environment-setup-method-1-source-install)
+- [Step 1: Prepare Model Weights (BF16 for SFT)](https://chatgpt.com/c/6975bb7f-52e0-839c-a727-ec4b5d6723b5#step-1-prepare-model-weights-bf16-for-sft)
+- [Step 2: Prepare YAML for LoRA SFT (KTransformers Backend)](https://chatgpt.com/c/6975bb7f-52e0-839c-a727-ec4b5d6723b5#step-2-prepare-yaml-for-lora-sft-ktransformers-backend)
+- [Step 3: Run LoRA SFT](https://chatgpt.com/c/6975bb7f-52e0-839c-a727-ec4b5d6723b5#step-3-run-lora-sft)
+- [Step 4: Post-SFT Quick Verification with LlamaFactory (Optional)](https://chatgpt.com/c/6975bb7f-52e0-839c-a727-ec4b5d6723b5#step-4-post-sft-quick-verification-with-LlamaFactory-optional)
+- [Step 5: SGLang Serving with LoRA (Recommended Delivery Path)](https://chatgpt.com/c/6975bb7f-52e0-839c-a727-ec4b5d6723b5#step-5-sglang-serving-with-lora-recommended-delivery-path)
+
+## Hardware Requirements
+
+### Training (LoRA SFT)
+
+- **LlamaFactory + KTransformers**
+- **GPU**: 4 * NVIDIA RTX 4090 24GB (or equivalent with at least total 48GB VRAM available)
+- **CPU**: x86 CPU with AMX support
+- **RAM**: At least 2TGB system memory
+- Swap can be used if CPU memory is insufficient
+
+### Inference (LoRA Adapter + Original Model)
+
+- **SGLang + KTransformers**
+- **GPU**: 2 * NVIDIA RTX 4090 24GB (or equivalent with at least total 48GB VRAM available)
+- **CPU**: x86 CPU with AVX512F support (e.g., Intel Sapphire Rapids)
+- **RAM**: At least 600GB system memory
+- **Storage**: ~600GB for model weights (native INT4 weight, same weight dir for CPU and GPU)
+
+
+
+## Step 0: Environment Setup
+
+We recommend to separate **two conda environments**:
+
+| Environment | Purpose                                             |
+| ----------- | --------------------------------------------------- |
+| `kt-kernel` | Inference & serving (KTransformers + SGLang)        |
+| `kt-sft`    | Training (LlamaFactory + KTransformers SFT backend) |
+
+### 0.1 Inference Environment: `kt-kernel`
+
+```bash
+conda create -n kt-kernel python=3.11
+conda activate kt-kernel
+
+git clone https://github.com/kvcache-ai/ktransformers.git
+git checkout kimi_k2.5
+git submodule update --init --recursive
+cd kt-kernel && ./install.sh
+```
+
+### 0.2 Install SGLang (Inference / Serving)
+
+**Recommended for Kimi-K2.5:**
+
+```bash
+git clone https://github.com/kvcache-ai/sglang.git
+cd sglang
+git checkout kimi_k2.5
+pip install -e "python[all]"
+```
+
+### 0.3 Training Environment: `kt-sft`
+
+```bash
+conda create -n kt-sft python=3.11
+conda activate kt-sft
+
+git clone https://github.com/hiyouga/LlamaFactory.git
+cd LlamaFactory
+pip install -e .
+```
+
+### 0.4 Install KTransformers SFT Dependencies
+
+```bash
+conda install -y -c conda-forge libstdcxx-ng gcc_impl_linux-64
+conda install -y -c nvidia/label/cuda-11.8.0 cuda-runtime
+
+# Install matching wheels (recommended), from https://github.com/kvcache-ai/ktransformers/releases
+pip install ktransformers-<matching-version>.whl
+pip install flash_attn-<matching-version>.whl
+```
+
+## Step 1: Prepare Model Weights (BF16 for SFT)
+
+### 1.1 Download INT4 Weights
+
+KTransformers **requires BF16 weights for SFT**.
+
+```bash
+# Download Kimi-K2.5 (RAW-INT4 for both CPU and GPU)
+huggingface-cli download moonshotai/Kimi-K2.5 \
+  --local-dir /path/to/kimi-k2.5
+```
+
+### 1.2 Convert INT4 → BF16
+
+Kimi-K2.5 base model is in **INT4** format, convert it to **BF16** before SFT.
+
+## Step 2: Prepare YAML for LoRA SFT (KTransformers Backend)
+
+### 2.1 Training YAML (LoRA SFT)
+
+Example file:
+`examples/train_lora/kimik2_lora_sft_kt.yaml`
+
+Required fields:
+
+```yaml
+stage: sft
+finetuning_type: lora
+bf16: true
+
+use_kt: true
+kt_optimize_rule: <rule.yaml>
+cpu_infer: 32
+chunk_size: 8192
+```
+
+Other fields (dataset, output_dir, learning rate, epochs) can be adjusted as usual.
+
+### 2.2 Inference YAML (LlamaFactory Verification)
+
+Key requirements:
+
+- `adapter_name_or_path`: LoRA output directory
+- `infer_backend: ktransformers`
+- **Same `use_kt` and `kt_optimize_rule` as training**
+
+This YAML is used only for **quick verification**, not production serving.
+
+## Step 3: Run LoRA SFT
+
+```bash
+conda activate kt-sft
+cd LlamaFactory
+
+USE_KT=1 llamafactory-cli train examples/train_lora/kimik2_lora_sft_kt.yaml
+```
+
+After training, the LoRA adapter is saved to `output_dir`.
+
+## Step 4: Post-SFT Quick Verification with LlamaFactory (Optional)
+
+Before production deployment, the new PDF recommends a **lightweight sanity check**.
+
+```bash
+conda activate kt-sft
+cd LlamaFactory
+
+llamafactory-cli chat examples/inference/kimik2_lora_sft_kt.yaml
+```
+
+Purpose:
+
+- Validate LoRA correctness
+- Ensure reproducibility
+- Not for throughput benchmarking
+
+## Step 5: SGLang Serving with LoRA (Recommended Delivery Path)
+
+This is the **major runtime update** introduced by the new PDF.
+
+### 5.1 Convert LoRA for SGLang
+
+```bash
+python ktransformers/kt-kernel/scripts/convert_lora.py \
+  --base_path /path/to/kimi-base-model \
+  --lora_path /path/to/llamafactory/output_dir \
+  --output_path /path/to/lora_converted
+```
+
+### 5.2 (Optional) Convert CPU Weights to INT8
+
+To reduce CPU memory usage:
+
+```bash
+python ktransformers/kt-kernel/scripts/convert_cpu_weights.py \
+  --base_path /path/to/kimi-base-model \
+  --output_dir /path/to/kimi-base-model-int8
+```
+
+This produces:
+
+```text
+/path/to/kimi-base-model-int8/int8
+```
+
+### 5.3 Launch SGLang Server with LoRA
+
+```bash
+conda activate kt-kernel
+
+python -m sglang.launch_server \
+  --enable-lora \
+  --lora-paths lora1=/path/to/lora_converted \
+  --lora-backend triton \
+  --model-path /path/to/kimi-base-model \
+  --tp 1 \
+  --trust-remote-code \
+  --context-length 4096 \
+  --kt-weight-path /path/to/kimi-base-model-int8/int8 \
+  --mem-fraction-static 0.9
+```
+
+Notes:
+
+- `--kt-weight-path` points to CPU INT8 weights
+- Adjust `tp`, `context-length`, and memory parameters per machine
+- RAWINT4 inference paths can follow **Kimi-K2.5-Native** directly
\ No newline at end of file