pegasus-xsum-mic / integration.json
tarekziade's picture
tarekziade HF Staff
Publish MIC usage, slim fixture and reviewed Transformers expectations
285728d verified
Raw History Blame Contribute Delete
2.41 kB
{
"mic": "0.3",
"name": "Pegasus summarization",
"usage": "AutoTokenizer + AutoModelForSeq2SeqLM, batched, padded, generate + one training step",
"architecture": "PegasusForConditionalGeneration",
"checkpoints": {
"google/pegasus-xsum": {
"fixture": "huggingface-mic/pegasus-xsum-mic"
}
},
"fixtures": {
"huggingface-mic/pegasus-xsum-mic": {
"built_from": "google/pegasus-xsum",
"covers": [
"google/pegasus-xsum"
],
"why_separate": "The full checkpoint is 16 layers and 1024-d; the fixture is a slim 2-layer 64-d model for CPU tests.",
"build": {
"seed": 17,
"model_class": "PegasusForConditionalGeneration",
"config_overrides": {
"fields": {
"encoder_layers": 2,
"decoder_layers": 2,
"d_model": 64,
"encoder_attention_heads": 4,
"decoder_attention_heads": 4,
"encoder_ffn_dim": 128,
"decoder_ffn_dim": 128
},
"reason": "Shrink the 16-layer 1024-d model to a 2-layer 64-d fixture for CPU tests; keep vocab and token ids intact."
}
}
}
},
"integrations": {
"transformers": {
"entry_point": "integrations/transformers/usage.md",
"format": "runnable-markdown",
"requires": [
"transformers>=5",
"torch"
],
"targets": [
"fixtures",
"checkpoints"
],
"variants": {
"default": {},
"eager": {
"attn_implementation": "eager"
},
"bfloat16": {
"dtype": "bfloat16"
}
},
"training": true,
"resources": {
"fixture": {
"memory_gb": 1,
"download_mb": 20
},
"google/pegasus-xsum": {
"memory_gb": 8,
"download_mb": 2200,
"gpu_memory_gb": 12
}
}
}
},
"outputs": {
"runtime": {
"compare": "exact"
},
"summary_ids": {
"compare": "exact"
},
"summary_text": {
"compare": "exact"
},
"logits.probe": {
"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"
}
}
}