Florence-2-base-mic / integration.json
tarekziade's picture
tarekziade HF Staff
Publish MIC usage, slim fixture and reviewed Transformers expectations
da124c0 verified
Raw History Blame Contribute Delete
3.02 kB
{
"mic": "0.3",
"name": "Florence-2 image captioning",
"usage": "Florence2Processor + Florence2ForConditionalGeneration, batched images, encoder-decoder training with labels",
"architecture": "Florence2ForConditionalGeneration",
"checkpoints": {
"florence-community/Florence-2-base": {
"fixture": "huggingface-mic/Florence-2-base-mic"
}
},
"fixtures": {
"huggingface-mic/Florence-2-base-mic": {
"built_from": "florence-community/Florence-2-base",
"covers": [
"florence-community/Florence-2-base"
],
"build": {
"seed": 17,
"model_class": "Florence2ForConditionalGeneration",
"config_overrides": {
"fields": {
"text_config.d_model": 64,
"text_config.encoder_layers": 1,
"text_config.decoder_layers": 1,
"text_config.encoder_attention_heads": 2,
"text_config.decoder_attention_heads": 2,
"text_config.encoder_ffn_dim": 128,
"text_config.decoder_ffn_dim": 128,
"vision_config.embed_dim": [
32,
64,
128,
256
],
"vision_config.depths": [
1,
1,
2,
1
],
"vision_config.num_heads": [
2,
4,
8,
16
],
"vision_config.num_groups": [
2,
4,
8,
16
],
"vision_config.projection_dim": 64
},
"reason": "Shrink vision and text towers for CPU-runnable fixture; keep vocab_size intact for tokenizer compatibility"
}
}
}
},
"integrations": {
"transformers": {
"entry_point": "integrations/transformers/usage.md",
"format": "runnable-markdown",
"requires": [
"transformers>=5",
"torch",
"Pillow",
"numpy"
],
"targets": [
"fixtures",
"checkpoints"
],
"variants": {
"default": {},
"eager": {
"attn_implementation": "eager"
},
"bfloat16": {
"dtype": "bfloat16"
}
},
"training": true,
"resources": {
"fixture": {
"memory_gb": 2,
"download_mb": 100
},
"florence-community/Florence-2-base": {
"memory_gb": 8,
"download_mb": 1400,
"variants": {
"bfloat16": {
"memory_gb": 4
}
}
}
}
}
},
"outputs": {
"runtime": {
"compare": "exact"
},
"inference_logits": {
"compare": "allclose",
"atol": 0.0001,
"rtol": 0.0001
},
"training.loss": {
"compare": "allclose",
"atol": 0.0001,
"rtol": 0.0001
},
"training.gradients": {
"compare": "all_true"
},
"training.updated": {
"compare": "true"
}
}
}