Text Generation
Transformers
Safetensors
clover_moe
clover
mixture-of-experts
diffusion
custom-code
conversational
custom_code
Instructions to use Diffuspace/Clover-35B-A3B-Preview with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use Diffuspace/Clover-35B-A3B-Preview with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("text-generation", model="Diffuspace/Clover-35B-A3B-Preview", trust_remote_code=True) messages = [ {"role": "user", "content": "Who are you?"}, ] pipe(messages)# pip install -U transformers accelerate # Load model directly from transformers import AutoModelForCausalLM model = AutoModelForCausalLM.from_pretrained("Diffuspace/Clover-35B-A3B-Preview", trust_remote_code=True, device_map="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use Diffuspace/Clover-35B-A3B-Preview with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "Diffuspace/Clover-35B-A3B-Preview" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Diffuspace/Clover-35B-A3B-Preview", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'Use Docker
docker model run hf.co/Diffuspace/Clover-35B-A3B-Preview
- SGLang
How to use Diffuspace/Clover-35B-A3B-Preview 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 "Diffuspace/Clover-35B-A3B-Preview" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Diffuspace/Clover-35B-A3B-Preview", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }'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 "Diffuspace/Clover-35B-A3B-Preview" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/chat/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "Diffuspace/Clover-35B-A3B-Preview", "messages": [ { "role": "user", "content": "What is the capital of France?" } ] }' - Docker Model Runner
How to use Diffuspace/Clover-35B-A3B-Preview with Docker Model Runner:
docker model run hf.co/Diffuspace/Clover-35B-A3B-Preview
The level of thought is too low, resulting in quality not being very good.
#5
by nocmt - opened
1、The generation_config.json configuration is as follows:
{
"max_new_tokens": 24576,
"do_sample": false,
"use_cache": true,
"eos_token_id": [
248046,
248044
],
"pad_token_id": 248044,
"transformers_version": "5.19.0",
"bos_token_id": 248044,
"presence_penalty": 0.9,
"frequency_penalty": 0.0,
"repetition_penalty": 1.05,
"dllm_length_penalty": 10.0,
"dllm_length_penalty_start": 32768,
"dllm_length_penalty_window": 16384,
"dllm_think_end_bias": 10.0,
"dllm_think_end_start": 16384,
"dllm_think_end_window": 8192,
"dllm_think_end_token_id": 248069,
"temperature": 1.0,
"top_k": null,
"top_p": 1.0
}
But it completed the thinking in 3.1 seconds, with a TPS of about 130.
2、Attempt to increase the constraint of prompt words:
Task: Create an HTML with an SVG that draws a pelican riding a bicycle in a 2D animation
Requirements: Think thoroughly and do your best to improve, beautify, and enrich this 2D animation HTML.
Better results:
3、Enhancing AI to assist in generating prompt words
You are a skilled graphic engineer proficient in SVG hand-drawing and front-end dynamic effects.
Create a 2D horizontal scrolling animation of a pelican riding a bicycle on a country road using a single HTML file (inline CSS + inline JS + inline SVG).
Output Constraints:
- Only output a single complete HTML file. No external resources (no CDN, no images, no font files) should be referenced.
- All graphics must be drawn using SVG primitives/path, and emojis are not allowed to replace characters or key elements.
- Do not leave TODO, placeholders, or "omitted here" notes.
- Code should be divided into sections with comments: CONFIG / Scene / Character / Animation Loop / Utility Functions.
Visual Requirements (must have):
1. Sky: multiple linear gradients + sun (with halo and slowly rotating light) + 3~5 clouds.
2. Background: silhouette of mountains (double layer, different opacity).
3. Middle ground: trees/shrubs.
4. Foreground: grass, a tarmac road with yellow dashed lines, scattered flowers.
5. Main Character: a pelican (white body, long beak + yellow throat pouch, webbed feet, optional scarf), sitting on the bicycle.
6. Bicycle: two wheels with spokes, red frame, handlebars, saddle, cranks + pedals on both sides.
7. Atmosphere: speed lines, floating particles, dust raised by the rear wheel.
Animation and Physics (mandatory requirements, check each item individually):
1. **Time Step**: Driven by `requestAnimationFrame` + `deltaTime`, all motion speeds are defined in `px/second`, and fixed position values for each frame are not allowed.
2. **Wheel Rolling without Sliding**: Wheel angular velocity ω = Linear Speed / Wheel Radius, so the wheel rotation must match the ground displacement strictly.
3. **Crank and Pedals**: The phase difference between the left and right cranks is π, and the pedals must rotate with the crank arm (using transform rotate), not just translate.
4. **Leg IK**: The pelican's legs use a two-stage inverse kinematics - given the hip point and foot point (foot point = center of pedal), solve the knee using the cosine rule, and ensure that the **thigh and shin lengths are constant**; the foot sole always sticks to the pedal.
5. **Parallax Scrolling Seamless Loop**: Each decorative layer uses 2~3 copies + modulo offset to achieve, and at any moment, there must not be any blank areas on the screen. Parallax speed levels: clouds < distant mountains < trees < road markings < foreground flowers.
6. **Main Character Secondary Animation**: The body rises and falls with the pedaling rhythm, the wings flutter slightly, and the throat pouch/scarf swings with speed.
7. **Dust Particles**: Dust particles are generated at the point of contact of the rear wheel, with speed, lifespan, and opacity decay; arrays and DOM nodes must correspond one-to-one, and the creation/destruction of particles must not delete other particles by mistake.
Code Quality (mandatory requirements):
- All DOM nodes are cached to variables in the initialization stage, and `getElementById` should not be called within the animation loop.
- Magic numbers are centralized in the CONFIG object with clear naming.
- Avoid unused variables and repeated `setAttribute` within the animation loop.
- Support `prefers-reduced-motion: reduce` by reducing or turning off animations.
- The page is responsive between 320px ~ 1920px width (use viewBox + percentage height for SVG).
Aesthetic Requirements:
- A unified color palette (not more than 12 main colors), use gradients and soft shadows to create depth.
- Character proportions should be coordinated, and the contact points of the pelican and bicycle (hip-saddle, foot-pedal, wing-handlebar) must be visually aligned.
Self-Check List (confirm each item individually after outputting the code and provide a brief report):
□ Do the wheels visually fit the ground?
□ Do the feet always stay on the pedals and the leg length remain unchanged during a full pedal revolution?
□ Does the background have any frame gaps or jitters during a 60-second camera roll?
□ Are all decorative layers implemented with parallax?
□ Is there any `getElementById` left in the animation loop?
Thought for 337.3 s
Effect:
The potential upper limit of intelligence may be here.
已经很好了,吹毛求疵没必要了
Try this for graphics. It works a lot better than this 35B.
https://huggingface.co/ccore/Qwen3.5-4B-3D-Coder



