Text-to-Image
Diffusers
Safetensors
MLX
cosmos3
quantization
int4
w4a16
sdnq
apple-silicon
cuda
8-bit precision
Instructions to use JuliaML/Cosmos3-Super-Text2Image-4Step-INT4-G64-BF16 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Diffusers
How to use JuliaML/Cosmos3-Super-Text2Image-4Step-INT4-G64-BF16 with Diffusers:
pip install -U diffusers transformers accelerate
import torch from diffusers import DiffusionPipeline # switch to "mps" for apple devices pipe = DiffusionPipeline.from_pretrained("JuliaML/Cosmos3-Super-Text2Image-4Step-INT4-G64-BF16", dtype=torch.bfloat16, device_map="cuda") prompt = "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k" image = pipe(prompt).images[0] - MLX
How to use JuliaML/Cosmos3-Super-Text2Image-4Step-INT4-G64-BF16 with MLX:
# Download the model from the Hub pip install huggingface_hub[hf_xet] huggingface-cli download --local-dir Cosmos3-Super-Text2Image-4Step-INT4-G64-BF16 JuliaML/Cosmos3-Super-Text2Image-4Step-INT4-G64-BF16
- Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- LM Studio
- Draw Things
- DiffusionBee
- Atomic Chat
| { | |
| "format": "cosmos3-sdnq-int8-verification-v1", | |
| "source": "models/Cosmos3-Super-Text2Image-4Step/transformer", | |
| "checkpoint": "checkpoints/Cosmos3-Super-Text2Image-4Step-SDNQ-INT4-G64-BF16IO/transformer", | |
| "selection_report": "benchmarks/results/mac-grid-fix-priority/w8-common-selection-verification.json", | |
| "weights_dtype": "int4", | |
| "group_size": 64, | |
| "selected_tensors": [ | |
| "layers.0.mlp.down_proj.weight", | |
| "layers.10.self_attn.to_out.weight", | |
| "layers.13.self_attn.add_v_proj.weight", | |
| "layers.16.mlp_moe_gen.up_proj.weight", | |
| "layers.19.mlp.up_proj.weight", | |
| "layers.20.self_attn.to_v.weight", | |
| "layers.23.self_attn.to_add_out.weight", | |
| "layers.26.self_attn.add_k_proj.weight", | |
| "layers.29.mlp_moe_gen.down_proj.weight", | |
| "layers.31.mlp.down_proj.weight", | |
| "layers.33.self_attn.to_out.weight", | |
| "layers.36.self_attn.add_v_proj.weight", | |
| "layers.39.mlp_moe_gen.up_proj.weight", | |
| "layers.41.mlp.up_proj.weight", | |
| "layers.43.self_attn.to_v.weight", | |
| "layers.46.self_attn.to_k.weight", | |
| "layers.49.self_attn.add_q_proj.weight", | |
| "layers.51.mlp_moe_gen.gate_proj.weight", | |
| "layers.54.mlp.down_proj.weight", | |
| "layers.56.self_attn.to_out.weight", | |
| "layers.59.self_attn.add_v_proj.weight", | |
| "layers.61.mlp_moe_gen.up_proj.weight", | |
| "layers.7.mlp.up_proj.weight", | |
| "layers.9.self_attn.to_v.weight", | |
| "proj_in.weight", | |
| "proj_out.weight" | |
| ], | |
| "packing_exact": true, | |
| "global_relative_mse": 0.012738522815792236, | |
| "core_global_relative_mse": 0.012738522815792236, | |
| "elapsed_seconds": 2.4497190000001865, | |
| "tensors": [ | |
| { | |
| "name": "layers.0.mlp.down_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.013577689921649044 | |
| }, | |
| { | |
| "name": "layers.10.self_attn.to_out.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.011668780191616646 | |
| }, | |
| { | |
| "name": "layers.13.self_attn.add_v_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.02004613792282179 | |
| }, | |
| { | |
| "name": "layers.16.mlp_moe_gen.up_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.016857524505423833 | |
| }, | |
| { | |
| "name": "layers.19.mlp.up_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.011879418020906305 | |
| }, | |
| { | |
| "name": "layers.20.self_attn.to_v.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.012257318742073727 | |
| }, | |
| { | |
| "name": "layers.23.self_attn.to_add_out.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.011372944960590221 | |
| }, | |
| { | |
| "name": "layers.26.self_attn.add_k_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.013145807345461526 | |
| }, | |
| { | |
| "name": "layers.29.mlp_moe_gen.down_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.012573942032957688 | |
| }, | |
| { | |
| "name": "layers.31.mlp.down_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.012352129115070342 | |
| }, | |
| { | |
| "name": "layers.33.self_attn.to_out.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.011667493050606282 | |
| }, | |
| { | |
| "name": "layers.36.self_attn.add_v_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.012393261560939773 | |
| }, | |
| { | |
| "name": "layers.39.mlp_moe_gen.up_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.01264913131274574 | |
| }, | |
| { | |
| "name": "layers.41.mlp.up_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.012425400114189694 | |
| }, | |
| { | |
| "name": "layers.43.self_attn.to_v.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.012918060888532262 | |
| }, | |
| { | |
| "name": "layers.46.self_attn.to_k.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.013407680545844998 | |
| }, | |
| { | |
| "name": "layers.49.self_attn.add_q_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.013472581053409854 | |
| }, | |
| { | |
| "name": "layers.51.mlp_moe_gen.gate_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.01414886739670003 | |
| }, | |
| { | |
| "name": "layers.54.mlp.down_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.012075089532909209 | |
| }, | |
| { | |
| "name": "layers.56.self_attn.to_out.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.011702265359361176 | |
| }, | |
| { | |
| "name": "layers.59.self_attn.add_v_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.01238331423893221 | |
| }, | |
| { | |
| "name": "layers.61.mlp_moe_gen.up_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.012850118066620123 | |
| }, | |
| { | |
| "name": "layers.7.mlp.up_proj.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.012046420206636184 | |
| }, | |
| { | |
| "name": "layers.9.self_attn.to_v.weight", | |
| "quantized": true, | |
| "packing_exact": true, | |
| "relative_mse": 0.012194300025057668 | |
| }, | |
| { | |
| "name": "proj_in.weight", | |
| "quantized": false, | |
| "exact": true | |
| }, | |
| { | |
| "name": "proj_out.weight", | |
| "quantized": false, | |
| "exact": true | |
| } | |
| ] | |
| } | |