Instructions to use z-lab/Kimi-K2.5-DFlash with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use z-lab/Kimi-K2.5-DFlash with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="z-lab/Kimi-K2.5-DFlash", trust_remote_code=True)

# pip install -U transformers accelerate
# Load model directly
from transformers import AutoTokenizer, AutoModel

tokenizer = AutoTokenizer.from_pretrained("z-lab/Kimi-K2.5-DFlash", trust_remote_code=True)
model = AutoModel.from_pretrained("z-lab/Kimi-K2.5-DFlash", trust_remote_code=True, device_map="auto")

Notebooks
Google Colab
Kaggle
Local Apps Settings

vLLM

How to use z-lab/Kimi-K2.5-DFlash with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "z-lab/Kimi-K2.5-DFlash"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "z-lab/Kimi-K2.5-DFlash",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/z-lab/Kimi-K2.5-DFlash

SGLang

How to use z-lab/Kimi-K2.5-DFlash with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "z-lab/Kimi-K2.5-DFlash" \
    --host 0.0.0.0 \
    --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "z-lab/Kimi-K2.5-DFlash",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

docker run --gpus all \
    --shm-size 32g \
    -p 30000:30000 \
    -v ~/.cache/huggingface:/root/.cache/huggingface \
    --env "HF_TOKEN=<secret>" \
    --ipc=host \
    lmsysorg/sglang:latest \
    python3 -m sglang.launch_server \
        --model-path "z-lab/Kimi-K2.5-DFlash" \
        --host 0.0.0.0 \
        --port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "z-lab/Kimi-K2.5-DFlash",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use z-lab/Kimi-K2.5-DFlash with Docker Model Runner:
```
docker model run hf.co/z-lab/Kimi-K2.5-DFlash
```

jianchen0311 commited on Apr 17

Commit

55bf596

verified ·

1 Parent(s): b121bd8

Update README.md

Browse files

Files changed (1) hide show

README.md +10 -7

README.md CHANGED Viewed

@@ -20,7 +20,7 @@ tags:
 **DFlash** is a novel speculative decoding method that utilizes a lightweight **block diffusion** model for drafting. It enables efficient, high-quality parallel drafting that pushes the limits of inference speed.
-This model is the **drafter** component. It must be used in conjunction with the target model `moonshotai/Kimi-K2.5`. It was trained with a context length of 4096 tokens.
 <div align="center">
   <img src="assets/dflash_system.png" alt="DFlash Architecture" width="100%">
@@ -36,6 +36,8 @@ uv pip install vllm
 uv pip install -U vllm --torch-backend=auto --extra-index-url https://wheels.vllm.ai/nightly
 ```
 SGLang:
 ```bash
 uv pip install "git+https://github.com/sgl-project/sglang.git@refs/pull/20547/head#subdirectory=python"
@@ -46,8 +48,8 @@ uv pip install "git+https://github.com/sgl-project/sglang.git@refs/pull/20547/he
 vLLM:
 ```bash
 vllm serve moonshotai/Kimi-K2.5 \
-  --speculative-config '{"method": "dflash", "model": "z-lab/Kimi-K2.5-DFlash", "num_speculative_tokens": 15}' \
-  --attention-backend flash_attn \
   --max-num-batched-tokens 32768
 ```
@@ -62,10 +64,12 @@ python -m sglang.launch_server \
     --model-path moonshotai/Kimi-K2.5 \
     --speculative-algorithm DFLASH \
     --speculative-draft-model-path z-lab/Kimi-K2.5-DFlash \
-    --speculative-num-draft-tokens 16 \
     --tp-size 8 \
-    --attention-backend fa3 \
     --mem-fraction-static 0.9 \
     --trust-remote-code
 ```
 > **Tip:** For long-context or agentic workloads, add `--speculative-dflash-draft-window-size WINDOW_SIZE` to enable sliding-window attention for the drafter.
@@ -88,9 +92,8 @@ print(response.choices[0].message.content)
 ## Early Results
 - Thinking: enabled
 - Max new tokens: 4096
-- Block size: 16
 - SGLang results. vLLM results might be different.
-- Epoch 1.8
 | Dataset   | Accept Length |
 |-----------|---------------|
 | GSM8K     | 5.8 |

 **DFlash** is a novel speculative decoding method that utilizes a lightweight **block diffusion** model for drafting. It enables efficient, high-quality parallel drafting that pushes the limits of inference speed.
+This model is the **drafter** component. It must be used in conjunction with the target model `moonshotai/Kimi-K2.5`.
 <div align="center">
   <img src="assets/dflash_system.png" alt="DFlash Architecture" width="100%">
 uv pip install -U vllm --torch-backend=auto --extra-index-url https://wheels.vllm.ai/nightly
 ```
+Please refer to [PR39930](https://github.com/vllm-project/vllm/pull/39930) to see how to use DFlash on vLLM.
 SGLang:
 ```bash
 uv pip install "git+https://github.com/sgl-project/sglang.git@refs/pull/20547/head#subdirectory=python"
 vLLM:
 ```bash
 vllm serve moonshotai/Kimi-K2.5 \
+  --speculative-config '{"method": "dflash", "model": "z-lab/Kimi-K2.5-DFlash", "num_speculative_tokens": 7}' \
+  --attention-backend flashinfer \
   --max-num-batched-tokens 32768
 ```
     --model-path moonshotai/Kimi-K2.5 \
     --speculative-algorithm DFLASH \
     --speculative-draft-model-path z-lab/Kimi-K2.5-DFlash \
+    --speculative-num-draft-tokens 8 \
     --tp-size 8 \
+    --attention-backend trtllm_mla \
+    --speculative-draft-attention-backend fa4 \
     --mem-fraction-static 0.9 \
+    --speculative-dflash-draft-window-size 4096 \
     --trust-remote-code
 ```
 > **Tip:** For long-context or agentic workloads, add `--speculative-dflash-draft-window-size WINDOW_SIZE` to enable sliding-window attention for the drafter.
 ## Early Results
 - Thinking: enabled
 - Max new tokens: 4096
+- Block size: 8
 - SGLang results. vLLM results might be different.
 | Dataset   | Accept Length |
 |-----------|---------------|
 | GSM8K     | 5.8 |