Image-to-Video
Diffusers
Safetensors
MLX
cosmos3
video-generation
quantization
int8
w8a16
sdnq
apple-silicon
cuda
8-bit precision
Instructions to use JuliaML/Cosmos3-Super-Image2Video-4Step-INT8-G64-BF16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use JuliaML/Cosmos3-Super-Image2Video-4Step-INT8-G64-BF16 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline from diffusers.utils import load_image, export_to_video # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("JuliaML/Cosmos3-Super-Image2Video-4Step-INT8-G64-BF16", dtype=torch.bfloat16, device_map="cuda") pipe.to("cuda") prompt = "A man with short gray hair plays a red electric guitar." image = load_image( "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/guitar-man.png" ) output = pipe(image=image, prompt=prompt).frames[0] export_to_video(output, "output.mp4") - MLX
How to use JuliaML/Cosmos3-Super-Image2Video-4Step-INT8-G64-BF16 with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir Cosmos3-Super-Image2Video-4Step-INT8-G64-BF16 JuliaML/Cosmos3-Super-Image2Video-4Step-INT8-G64-BF16
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Atomic Chat
| { | |
| "format": "cosmos3-sdnq-verification-v1", | |
| "source": "models/Cosmos3-Super-Image2Video-4Step/transformer", | |
| "checkpoint": "checkpoints/Cosmos3-Super-Image2Video-4Step-SDNQ-INT8-G64-BF16IO/transformer", | |
| "selection_report": "benchmarks/i2v/verification-selection.json", | |
| "weights_dtype": "int8", | |
| "group_size": 64, | |
| "selected_tensors": [ | |
| "embed_tokens.weight", | |
| "layers.0.self_attn.to_q.weight", | |
| "layers.0.mlp.up_proj.weight", | |
| "layers.21.self_attn.to_k.weight", | |
| "layers.21.mlp_moe_gen.down_proj.weight", | |
| "layers.42.self_attn.add_v_proj.weight", | |
| "layers.42.mlp.gate_proj.weight", | |
| "layers.63.self_attn.to_out.weight", | |
| "layers.63.mlp.down_proj.weight", | |
| "norm.weight", | |
| "proj_in.weight", | |
| "proj_out.weight" | |
| ], | |
| "packing_exact": true, | |
| "global_relative_mse": 4.091530100791147e-05, | |
| "core_global_relative_mse": 4.091530100791147e-05, | |
| "elapsed_seconds": 2.059502917000003, | |
| "tensors": [ | |
| { | |
| "name": "embed_tokens.weight", | |
| "quantized": false, | |
| "exact": true | |
| }, | |
| { | |
| "name": "layers.0.self_attn.to_q.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 3.631787930218696e-05 | |
| }, | |
| { | |
| "name": "layers.0.mlp.up_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 3.787928616525752e-05 | |
| }, | |
| { | |
| "name": "layers.21.self_attn.to_k.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 4.015870025460336e-05 | |
| }, | |
| { | |
| "name": "layers.21.mlp_moe_gen.down_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 4.617328465040616e-05 | |
| }, | |
| { | |
| "name": "layers.42.self_attn.add_v_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 4.422963696925805e-05 | |
| }, | |
| { | |
| "name": "layers.42.mlp.gate_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 3.8449613650290844e-05 | |
| }, | |
| { | |
| "name": "layers.63.self_attn.to_out.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 3.570206942213908e-05 | |
| }, | |
| { | |
| "name": "layers.63.mlp.down_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 4.066047565003439e-05 | |
| }, | |
| { | |
| "name": "norm.weight", | |
| "quantized": false, | |
| "exact": true | |
| }, | |
| { | |
| "name": "proj_in.weight", | |
| "quantized": false, | |
| "exact": true | |
| }, | |
| { | |
| "name": "proj_out.weight", | |
| "quantized": false, | |
| "exact": true | |
| } | |
| ] | |
| } | |