tarekziade HF Staff
Publish MIC usage, slim fixture and reviewed Transformers expectations
6676a46 verified Download build.json from huggingface-mic/language-perceiver-mic: direct link, hf CLI and curl.
- Browser
- Download file 4.33 kB
-
https://huggingface.co/huggingface-mic/language-perceiver-mic/resolve/main/build.json
- Command line
-
hf download hf://huggingface-mic/language-perceiver-mic/build.json
-
curl -L -o build.json https://huggingface.co/huggingface-mic/language-perceiver-mic/resolve/main/build.json
4.33 kB
| { | |
| "fixture": "huggingface-mic/language-perceiver-mic", | |
| "key": "bcce0e644a1445ca0298d69b426820f5a85bee4b6c29f251b1d624c350477f75", | |
| "key_fields": { | |
| "architectures": [ | |
| "PerceiverForMaskedLM" | |
| ], | |
| "output_hidden_states": false, | |
| "return_dict": true, | |
| "chunk_size_feed_forward": 0, | |
| "is_encoder_decoder": false, | |
| "id2label": { | |
| "0": "LABEL_0", | |
| "1": "LABEL_1" | |
| }, | |
| "label2id": { | |
| "LABEL_0": 0, | |
| "LABEL_1": 1 | |
| }, | |
| "problem_type": null, | |
| "num_latents": 256, | |
| "d_latents": 64, | |
| "d_model": 64, | |
| "num_blocks": 1, | |
| "num_self_attends_per_block": 2, | |
| "num_self_attention_heads": 1, | |
| "num_cross_attention_heads": 1, | |
| "qk_channels": 64, | |
| "v_channels": 64, | |
| "cross_attention_shape_for_attention": "kv", | |
| "self_attention_widening_factor": 1, | |
| "cross_attention_widening_factor": 1, | |
| "hidden_act": "gelu", | |
| "attention_probs_dropout_prob": 0.1, | |
| "initializer_range": 0.02, | |
| "layer_norm_eps": 1e-12, | |
| "use_query_residual": true, | |
| "vocab_size": 262, | |
| "max_position_embeddings": 2048, | |
| "image_size": 56, | |
| "train_size": [ | |
| 368, | |
| 496 | |
| ], | |
| "num_frames": 16, | |
| "audio_samples_per_frame": 1920, | |
| "samples_per_patch": 16, | |
| "output_shape": [ | |
| 1, | |
| 16, | |
| 224, | |
| 224 | |
| ], | |
| "output_num_channels": 512, | |
| "_label_trainable_num_channels": 1024, | |
| "model_type": "perceiver", | |
| "output_attentions": false | |
| }, | |
| "covers": { | |
| "deepmind/language-perceiver": "47a1ba43d8198ea10cf9a158fce114953e9441a1" | |
| }, | |
| "built_from": { | |
| "repo": "deepmind/language-perceiver", | |
| "revision": "47a1ba43d8198ea10cf9a158fce114953e9441a1" | |
| }, | |
| "seed": 17, | |
| "dtype": "float32", | |
| "builder": "example/build_fixtures.py", | |
| "builder_revision": "c28b75071336deade6570ed8f9f2ed05245b2210", | |
| "recipe": { | |
| "seed": 17, | |
| "config_overrides": { | |
| "fields": { | |
| "d_latents": 64, | |
| "v_channels": 64, | |
| "qk_channels": 64, | |
| "num_self_attends_per_block": 2 | |
| }, | |
| "reason": "The generic rule shrinks d_model but not the latent width: d_latents and v_channels stay 1280 and qk_channels 256 across 26 self-attention layers per block (97% of 194M parameters, a 777 MB fixture). Two self-attends keep the block's repeated-layer structure, as the Perceiver model tester does (d_latents 20, num_self_attends_per_block 2)." | |
| } | |
| }, | |
| "shrink": "rule", | |
| "model_class": "PerceiverForMaskedLM", | |
| "parameters": 421638, | |
| "missing_parameters": [], | |
| "serialization": { | |
| "file": "model.safetensors", | |
| "source_file": "model.safetensors", | |
| "reverted_conversions": [] | |
| }, | |
| "tester_sizes": { | |
| "source": "https://raw.githubusercontent.com/huggingface/transformers/main/tests/models/perceiver/test_modeling_perceiver.py", | |
| "fields": { | |
| "num_cross_attention_heads": 1, | |
| "num_self_attention_heads": 1 | |
| } | |
| }, | |
| "checkpoint_inspection": "parameter headers/indexes; bounded binary fallback", | |
| "numeric_health_scope": "CPU float32/bfloat16/float16; autocast for low-precision losses", | |
| "captured": { | |
| "call": "training step", | |
| "inputs": [ | |
| "attention_mask", | |
| "inputs", | |
| "labels" | |
| ], | |
| "class": "PerceiverForMaskedLM" | |
| }, | |
| "weights": "random", | |
| "gate": { | |
| "random": { | |
| "pass": true, | |
| "accepted_insensitive": {}, | |
| "threshold": 0.001, | |
| "deltas": { | |
| "attention_mask:all_ones": 0.012976217083632946 | |
| }, | |
| "weakest": "attention_mask:all_ones", | |
| "no_perturbation_defined": [ | |
| "inputs" | |
| ], | |
| "not_a_forward_parameter": [] | |
| } | |
| }, | |
| "numeric_health": { | |
| "float32": { | |
| "outputs_finite": true, | |
| "loss_finite": true, | |
| "gradients_finite": true, | |
| "finite": true | |
| }, | |
| "bfloat16": { | |
| "outputs_finite": true, | |
| "loss_finite": true, | |
| "gradients_finite": true, | |
| "finite": true | |
| }, | |
| "float16": { | |
| "outputs_finite": true, | |
| "loss_finite": true, | |
| "unscaled_gradients_finite": true, | |
| "finite": true | |
| } | |
| }, | |
| "copied_unchanged": [ | |
| "special_tokens_map.json", | |
| "tokenizer_config.json" | |
| ], | |
| "environment": { | |
| "python": "3.11.14", | |
| "machine": "arm64", | |
| "torch": "2.14.1", | |
| "transformers": "5.20.1", | |
| "tokenizers": "0.23.3" | |
| }, | |
| "sha256": { | |
| "config.json": "4f34e90632fe20eece553bc14a00de78c0db7a813466b292e677f63afae51fa1", | |
| "model.safetensors": "7d583587631bc35ebbf2d8270e0037ad3784e9c63c773114c271fbaa6b74cac9", | |
| "special_tokens_map.json": "08f4e103f6d851b4c14edd06d593dfb25db3339e05299393f08ad53e7733e454", | |
| "tokenizer_config.json": "1f818797228803a29d7889c3614edfa19d8b3a6742669825df20306ad8ea3abe" | |
| } | |
| } | |