TokenFlow¶

Image generation model from ByteFlow AI. Generation only — no understanding or editing.

Original repository: https://github.com/ByteFlow-AI/TokenFlow
Backbone key: tokenflow
Capabilities: Generation ONLY

Dependencies¶

The model environment is managed via the tokenflow image defined in modal/images.py. For local setup, install the dependencies listed in model/TokenFlow/requirements.txt.

Inference¶

CLI¶

# Text-to-image generation
PYTHONPATH=src python -m umm.cli.main infer --config configs/inference/tokenflow_generation.yaml

# DPG Bench-style generation (dense prompt)
PYTHONPATH=src python -m umm.cli.main infer --config configs/inference/tokenflow_dpg_generation.yaml

Python API¶

from umm.inference.pipeline import InferencePipeline
from umm.inference.multimodal_inputs import InferenceRequest

pipeline = InferencePipeline(backbone_name="tokenflow", backbone_cfg={
    "model_path": "/path/to/tokenflow_weights",
    "tokenizer_path": "/path/to/tokenizer",
    "cfg": 7.5,
    "batch_size": 1,
})

# Generation
result = pipeline.run(InferenceRequest(
    backbone="tokenflow", task="generation",
    prompt="A cat sitting on a rainbow",
))

Note: TokenFlow supports generation only. Understanding and editing are not available.

Supported Benchmarks¶

Benchmark	Config
DPG Bench	`configs/eval/dpg_bench/dpg_bench_tokenflow.yaml`
GenEval	`configs/eval/geneval/geneval_tokenflow.yaml`
WISE	`configs/eval/wise/wise_tokenflow.yaml`
UEval	`configs/eval/ueval/ueval_tokenflow.yaml`

No understanding benchmarks are supported.

# Example: run GenEval
PYTHONPATH=src python -m umm.cli.main eval --config configs/eval/geneval/geneval_tokenflow.yaml

# Example: run DPG Bench
PYTHONPATH=src python -m umm.cli.main eval --config configs/eval/dpg_bench/dpg_bench_tokenflow.yaml

Key Configuration Parameters¶

Generation: cfg (guidance scale), loop, mixed_precision, batch_size