{ "mic": "0.3", "name": "SpeechT5-ASR", "usage": "AutoProcessor + AutoModelForSpeechSeq2Seq, batched synthetic audio, greedy decode, one SGD step", "architecture": "SpeechT5ForSpeechToText", "checkpoints": { "microsoft/speecht5_asr": { "fixture": "hf-internal-testing/mic-speech_t5-slim" } }, "fixtures": { "hf-internal-testing/mic-speech_t5-slim": { "built_from": "microsoft/speecht5_asr", "covers": [ "microsoft/speecht5_asr" ], "build": { "seed": 17, "accept_insensitive": { "input_values:halve": "The speech prenet applies per-channel GroupNorm after convolution, making positive amplitude scaling invariant up to epsilon. The measured half-amplitude delta is 2.09e-7; batch, reversal and padding-mask perturbations remain required and each exceeds 0.003." } } } }, "integrations": { "transformers": { "entry_point": "integrations/transformers/usage.md", "format": "runnable-markdown", "requires": [ "transformers>=4.30", "torch", "numpy" ], "targets": [ "fixtures" ], "variants": { "default": {}, "eager": { "attn_implementation": "eager" }, "bfloat16": { "dtype": "bfloat16" } }, "training": true, "resources": { "fixture": { "memory_gb": 2, "download_mb": 50 }, "microsoft/speecht5_asr": { "memory_gb": 4, "download_mb": 1100 } } } }, "outputs": { "runtime": { "compare": "exact" }, "predicted_tokens": { "compare": "exact" }, "predicted_text": { "compare": "exact" }, "selected_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" } } }