tarekziade HF Staff commited on
Commit
6c70b7e
·
verified ·
1 Parent(s): 4ad5792

Publish MIC usage, slim fixture and reviewed Transformers expectations

Browse files
.mic-repo.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "format": "mic-umbrella-v1",
3
+ "contract_files": [
4
+ "integration.json",
5
+ "integrations/transformers/usage.md"
6
+ ]
7
+ }
README.md ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ tags:
3
+ - mic
4
+ - test-artifact
5
+ base_model: deepmind/language-perceiver
6
+ ---
7
+
8
+ # language-perceiver-mic
9
+
10
+ Slim model weights, executable usage and framework expectations in one repository.
11
+
12
+ - [Contract manifest](integration.json)
13
+ - [Transformers usage](integrations/transformers/usage.md)
14
+ - [Transformers expectations](integrations/transformers/expectations)
15
+ - [Fixture provenance](build.json)
16
+
17
+ ## Contract
18
+
19
+ ### Perceiver masked language modeling
20
+
21
+ Runnable usage and the framework expectations directory are linked above.
22
+
23
+ ## Slim model
24
+
25
+ ### Slim fixture: language-perceiver-mic
26
+
27
+ A 194.1M-parameter fixture for the [Perceiver MIC](https://huggingface.co/huggingface-mic/language-perceiver-mic), covering [`deepmind/language-perceiver`](https://huggingface.co/deepmind/language-perceiver). It is a test artifact, not a usable model, tagged `test-artifact` to keep it out of candidate-model search.
28
+
29
+ - **Fixture key `de355e4dc541`.** It covers every checkpoint whose processor files are byte-identical and whose config matches after shrinking.
30
+ - **Same architecture, smaller config:** `PerceiverForMaskedLM`, shrunk by the generic rule in [`sweep.py`](https://github.com/huggingface/transformers-mic/blob/main/experiments/sweep.py) ().
31
+ - **Processor files copied byte for byte** from `deepmind/language-perceiver`: `special_tokens_map.json`, `tokenizer_config.json`.
32
+ - **Random weights (seed 17).** They passed the input-sensitivity gate: every perturbation of the contract's inputs moved the logits by at least 0.001 (weakest: `attention_mask:all_ones`, 0.106), so no training was needed.
33
+ - **Same tensor names as the real checkpoint.** The real checkpoint has every parameter, and so does the fixture.
34
+ - **Saved in float32**, the dtype `deepmind/language-perceiver` loads in by default.
35
+ - **Finite in float32, bfloat16, and float16** (forward, loss, gradients).
36
+
37
+ Built by [`build_fixtures.py`](https://github.com/huggingface/transformers-mic/blob/main/example/build_fixtures.py); [`build.json`](build.json) records the pinned sources, gate results, and file hashes. This repository includes the slim weights, usage and framework expectations; rebuild the other files with `python example/build_fixtures.py --fixture language-perceiver-mic`.
build.json ADDED
@@ -0,0 +1,135 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "fixture": "huggingface-mic/language-perceiver-mic",
3
+ "key": "de355e4dc541831cc4c2086166054be8ceb0d3b120ebce38703e8abfb7f87829",
4
+ "key_fields": {
5
+ "architectures": [
6
+ "PerceiverForMaskedLM"
7
+ ],
8
+ "output_hidden_states": false,
9
+ "return_dict": true,
10
+ "chunk_size_feed_forward": 0,
11
+ "is_encoder_decoder": false,
12
+ "id2label": {
13
+ "0": "LABEL_0",
14
+ "1": "LABEL_1"
15
+ },
16
+ "label2id": {
17
+ "LABEL_0": 0,
18
+ "LABEL_1": 1
19
+ },
20
+ "problem_type": null,
21
+ "num_latents": 256,
22
+ "d_latents": 1280,
23
+ "d_model": 64,
24
+ "num_blocks": 1,
25
+ "num_self_attends_per_block": 26,
26
+ "num_self_attention_heads": 8,
27
+ "num_cross_attention_heads": 8,
28
+ "qk_channels": 256,
29
+ "v_channels": 1280,
30
+ "cross_attention_shape_for_attention": "kv",
31
+ "self_attention_widening_factor": 1,
32
+ "cross_attention_widening_factor": 1,
33
+ "hidden_act": "gelu",
34
+ "attention_probs_dropout_prob": 0.1,
35
+ "initializer_range": 0.02,
36
+ "layer_norm_eps": 1e-12,
37
+ "use_query_residual": true,
38
+ "vocab_size": 262,
39
+ "max_position_embeddings": 2048,
40
+ "image_size": 56,
41
+ "train_size": [
42
+ 368,
43
+ 496
44
+ ],
45
+ "num_frames": 16,
46
+ "audio_samples_per_frame": 1920,
47
+ "samples_per_patch": 16,
48
+ "output_shape": [
49
+ 1,
50
+ 16,
51
+ 224,
52
+ 224
53
+ ],
54
+ "output_num_channels": 512,
55
+ "_label_trainable_num_channels": 1024,
56
+ "model_type": "perceiver",
57
+ "output_attentions": false
58
+ },
59
+ "covers": {
60
+ "deepmind/language-perceiver": "47a1ba43d8198ea10cf9a158fce114953e9441a1"
61
+ },
62
+ "built_from": {
63
+ "repo": "deepmind/language-perceiver",
64
+ "revision": "47a1ba43d8198ea10cf9a158fce114953e9441a1"
65
+ },
66
+ "seed": 17,
67
+ "dtype": "float32",
68
+ "builder": "example/build_fixtures.py",
69
+ "model_class": "PerceiverForMaskedLM",
70
+ "parameters": 194113286,
71
+ "missing_parameters": [],
72
+ "checkpoint_inspection": "parameter headers/indexes; bounded binary fallback",
73
+ "numeric_health_scope": "CPU float32/bfloat16/float16; autocast for low-precision losses",
74
+ "captured": {
75
+ "call": "training step",
76
+ "inputs": [
77
+ "attention_mask",
78
+ "inputs",
79
+ "labels"
80
+ ],
81
+ "class": "PerceiverForMaskedLM"
82
+ },
83
+ "weights": "random",
84
+ "gate": {
85
+ "random": {
86
+ "pass": true,
87
+ "accepted_insensitive": {},
88
+ "threshold": 0.001,
89
+ "deltas": {
90
+ "attention_mask:all_ones": 0.10622552782297134
91
+ },
92
+ "weakest": "attention_mask:all_ones",
93
+ "no_perturbation_defined": [
94
+ "inputs"
95
+ ]
96
+ }
97
+ },
98
+ "numeric_health": {
99
+ "float32": {
100
+ "outputs_finite": true,
101
+ "loss_finite": true,
102
+ "gradients_finite": true,
103
+ "finite": true
104
+ },
105
+ "bfloat16": {
106
+ "outputs_finite": true,
107
+ "loss_finite": true,
108
+ "gradients_finite": true,
109
+ "finite": true
110
+ },
111
+ "float16": {
112
+ "outputs_finite": true,
113
+ "loss_finite": true,
114
+ "gradients_finite": true,
115
+ "finite": true
116
+ }
117
+ },
118
+ "copied_unchanged": [
119
+ "special_tokens_map.json",
120
+ "tokenizer_config.json"
121
+ ],
122
+ "environment": {
123
+ "python": "3.11.14",
124
+ "machine": "arm64",
125
+ "torch": "2.14.1",
126
+ "transformers": "5.20.1",
127
+ "tokenizers": "0.23.3"
128
+ },
129
+ "sha256": {
130
+ "config.json": "8a00596028a7d53fde5782061e185fdf6a5902d080e440b11e69d99593db06ed",
131
+ "model.safetensors": "da10fb9094185afcaba694447b7e60e0c84b60345c2c13bb477187abc0ec8338",
132
+ "special_tokens_map.json": "08f4e103f6d851b4c14edd06d593dfb25db3339e05299393f08ad53e7733e454",
133
+ "tokenizer_config.json": "1f818797228803a29d7889c3614edfa19d8b3a6742669825df20306ad8ea3abe"
134
+ }
135
+ }
config.json ADDED
@@ -0,0 +1,43 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_label_trainable_num_channels": 1024,
3
+ "architectures": [
4
+ "PerceiverForMaskedLM"
5
+ ],
6
+ "attention_probs_dropout_prob": 0.1,
7
+ "audio_samples_per_frame": 1920,
8
+ "cross_attention_shape_for_attention": "kv",
9
+ "cross_attention_widening_factor": 1,
10
+ "d_latents": 1280,
11
+ "d_model": 64,
12
+ "dtype": "float32",
13
+ "hidden_act": "gelu",
14
+ "image_size": 56,
15
+ "initializer_range": 0.02,
16
+ "layer_norm_eps": 1e-12,
17
+ "max_position_embeddings": 2048,
18
+ "model_type": "perceiver",
19
+ "num_blocks": 1,
20
+ "num_cross_attention_heads": 8,
21
+ "num_frames": 16,
22
+ "num_latents": 256,
23
+ "num_self_attends_per_block": 26,
24
+ "num_self_attention_heads": 8,
25
+ "output_num_channels": 512,
26
+ "output_shape": [
27
+ 1,
28
+ 16,
29
+ 224,
30
+ 224
31
+ ],
32
+ "qk_channels": 256,
33
+ "samples_per_patch": 16,
34
+ "self_attention_widening_factor": 1,
35
+ "train_size": [
36
+ 368,
37
+ 496
38
+ ],
39
+ "transformers_version": "5.20.1",
40
+ "use_query_residual": true,
41
+ "v_channels": 1280,
42
+ "vocab_size": 262
43
+ }
integration.json ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "mic": "0.3",
3
+ "name": "Perceiver masked language modeling",
4
+ "usage": "PerceiverTokenizer + PerceiverForMaskedLM, byte-level masked LM with two sentences",
5
+ "architecture": "PerceiverForMaskedLM",
6
+ "checkpoints": {
7
+ "deepmind/language-perceiver": {
8
+ "fixture": "huggingface-mic/language-perceiver-mic"
9
+ }
10
+ },
11
+ "fixtures": {
12
+ "huggingface-mic/language-perceiver-mic": {
13
+ "built_from": "deepmind/language-perceiver",
14
+ "covers": [
15
+ "deepmind/language-perceiver"
16
+ ],
17
+ "build": {
18
+ "seed": 17
19
+ }
20
+ }
21
+ },
22
+ "integrations": {
23
+ "transformers": {
24
+ "entry_point": "integrations/transformers/usage.md",
25
+ "format": "runnable-markdown",
26
+ "requires": [
27
+ "transformers>=5",
28
+ "torch"
29
+ ],
30
+ "targets": [
31
+ "fixtures",
32
+ "checkpoints"
33
+ ],
34
+ "variants": {
35
+ "default": {},
36
+ "eager": {
37
+ "attn_implementation": "eager"
38
+ },
39
+ "bfloat16": {
40
+ "dtype": "bfloat16"
41
+ }
42
+ },
43
+ "training": true,
44
+ "resources": {
45
+ "fixture": {
46
+ "memory_gb": 1,
47
+ "download_mb": 50
48
+ },
49
+ "deepmind/language-perceiver": {
50
+ "memory_gb": 3,
51
+ "download_mb": 1100,
52
+ "gpu_memory_gb": 2
53
+ }
54
+ }
55
+ }
56
+ },
57
+ "outputs": {
58
+ "runtime": {
59
+ "compare": "exact"
60
+ },
61
+ "masked_logits": {
62
+ "compare": "allclose",
63
+ "atol": 0.0001,
64
+ "rtol": 0.0001
65
+ },
66
+ "training.loss": {
67
+ "compare": "allclose",
68
+ "atol": 0.0001,
69
+ "rtol": 0.0001
70
+ },
71
+ "training.gradients": {
72
+ "compare": "all_true"
73
+ },
74
+ "training.updated": {
75
+ "compare": "true"
76
+ }
77
+ }
78
+ }
integrations/transformers/expectations/cpu-x86_64-bf16/language-perceiver-mic.bfloat16.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "recorded_with": {
3
+ "contract": "huggingface-mic/language-perceiver-mic",
4
+ "contract_name": "Perceiver",
5
+ "contract_revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
6
+ "framework": "transformers",
7
+ "variant": "bfloat16",
8
+ "target": "language-perceiver-mic",
9
+ "kind": "fixture",
10
+ "revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
11
+ "environment": {
12
+ "python": "3.11.17",
13
+ "transformers": "5.21.0.dev0",
14
+ "torch": "2.14.0+cpu",
15
+ "torchvision": "0.29.1+cpu",
16
+ "torchaudio": "2.11.0+cpu",
17
+ "tokenizers": "0.23.3",
18
+ "numpy": "2.3.5",
19
+ "pillow": "12.3.0",
20
+ "accelerate": "1.15.0",
21
+ "safetensors": "0.8.0",
22
+ "huggingface-hub": "2.3.0",
23
+ "sentencepiece": "0.2.2",
24
+ "protobuf": "7.36.2",
25
+ "librosa": "0.11.0",
26
+ "soundfile": "0.14.0",
27
+ "scipy": "1.17.1"
28
+ },
29
+ "seconds": 8.54,
30
+ "environment_differs": {
31
+ "torch": "2.14.0+cpu (pinned 2.11.0)",
32
+ "torchvision": "0.29.1+cpu (pinned 0.26.0)",
33
+ "tokenizers": "0.23.3 (pinned 0.23.1)",
34
+ "numpy": "2.3.5 (pinned 1.26.4)",
35
+ "pillow": "12.3.0 (pinned 12.2.0)",
36
+ "accelerate": "1.15.0 (pinned 1.13.0)",
37
+ "huggingface-hub": "2.3.0 (pinned 1.32.0)",
38
+ "sentencepiece": "0.2.2 (pinned 0.2.1)",
39
+ "protobuf": "7.36.2 (pinned 6.33.6)",
40
+ "soundfile": "0.14.0 (pinned 0.13.1)",
41
+ "scipy": "1.17.1 (pinned 1.15.3)"
42
+ },
43
+ "platform": "cpu-x86_64-bf16"
44
+ },
45
+ "outputs": {
46
+ "runtime": {
47
+ "device": "cpu",
48
+ "dtype": "bfloat16",
49
+ "attn_implementation": "eager"
50
+ },
51
+ "masked_logits": [
52
+ [
53
+ 0.01190185546875,
54
+ -0.01123046875,
55
+ 0.020263671875,
56
+ -0.0120849609375,
57
+ -0.0277099609375
58
+ ],
59
+ [
60
+ 0.0118408203125,
61
+ -0.01123046875,
62
+ 0.0201416015625,
63
+ -0.0120849609375,
64
+ -0.027587890625
65
+ ],
66
+ [
67
+ 0.01177978515625,
68
+ -0.01116943359375,
69
+ 0.0201416015625,
70
+ -0.0120849609375,
71
+ -0.027587890625
72
+ ]
73
+ ],
74
+ "training": {
75
+ "loss": 5.5,
76
+ "gradients": {
77
+ "perceiver.input_preprocessor.embeddings.weight": true,
78
+ "perceiver.encoder.self_attends.0.attention.self.query.weight": true
79
+ },
80
+ "updated": true
81
+ }
82
+ }
83
+ }
integrations/transformers/expectations/cpu-x86_64-bf16/language-perceiver.bfloat16.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "recorded_with": {
3
+ "contract": "huggingface-mic/language-perceiver-mic",
4
+ "contract_name": "Perceiver",
5
+ "contract_revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
6
+ "framework": "transformers",
7
+ "variant": "bfloat16",
8
+ "target": "language-perceiver",
9
+ "kind": "checkpoint",
10
+ "revision": "47a1ba43d8198ea10cf9a158fce114953e9441a1",
11
+ "environment": {
12
+ "python": "3.11.17",
13
+ "transformers": "5.21.0.dev0",
14
+ "torch": "2.14.0+cpu",
15
+ "torchvision": "0.29.1+cpu",
16
+ "torchaudio": "2.11.0+cpu",
17
+ "tokenizers": "0.23.3",
18
+ "numpy": "2.3.5",
19
+ "pillow": "12.3.0",
20
+ "accelerate": "1.15.0",
21
+ "safetensors": "0.8.0",
22
+ "huggingface-hub": "2.3.0",
23
+ "sentencepiece": "0.2.2",
24
+ "protobuf": "7.36.2",
25
+ "librosa": "0.11.0",
26
+ "soundfile": "0.14.0",
27
+ "scipy": "1.17.1"
28
+ },
29
+ "seconds": 8.91,
30
+ "environment_differs": {
31
+ "torch": "2.14.0+cpu (pinned 2.11.0)",
32
+ "torchvision": "0.29.1+cpu (pinned 0.26.0)",
33
+ "tokenizers": "0.23.3 (pinned 0.23.1)",
34
+ "numpy": "2.3.5 (pinned 1.26.4)",
35
+ "pillow": "12.3.0 (pinned 12.2.0)",
36
+ "accelerate": "1.15.0 (pinned 1.13.0)",
37
+ "huggingface-hub": "2.3.0 (pinned 1.32.0)",
38
+ "sentencepiece": "0.2.2 (pinned 0.2.1)",
39
+ "protobuf": "7.36.2 (pinned 6.33.6)",
40
+ "soundfile": "0.14.0 (pinned 0.13.1)",
41
+ "scipy": "1.17.1 (pinned 1.15.3)"
42
+ },
43
+ "platform": "cpu-x86_64-bf16"
44
+ },
45
+ "outputs": {
46
+ "runtime": {
47
+ "device": "cpu",
48
+ "dtype": "bfloat16",
49
+ "attn_implementation": "eager"
50
+ },
51
+ "masked_logits": [
52
+ [
53
+ -16.75,
54
+ -16.5,
55
+ -16.375,
56
+ -11.8125,
57
+ -16.875
58
+ ],
59
+ [
60
+ -12.1875,
61
+ -12.4375,
62
+ -12.25,
63
+ -15.3125,
64
+ -12.25
65
+ ],
66
+ [
67
+ -12.75,
68
+ -13.0,
69
+ -12.8125,
70
+ -16.25,
71
+ -12.75
72
+ ]
73
+ ],
74
+ "training": {
75
+ "loss": 6.6875,
76
+ "gradients": {
77
+ "perceiver.input_preprocessor.embeddings.weight": true,
78
+ "perceiver.encoder.self_attends.0.attention.self.query.weight": true
79
+ },
80
+ "updated": true
81
+ }
82
+ }
83
+ }
integrations/transformers/expectations/cpu-x86_64/language-perceiver-mic.eager.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "recorded_with": {
3
+ "contract": "huggingface-mic/language-perceiver-mic",
4
+ "contract_name": "Perceiver",
5
+ "contract_revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
6
+ "framework": "transformers",
7
+ "variant": "eager",
8
+ "target": "language-perceiver-mic",
9
+ "kind": "fixture",
10
+ "revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
11
+ "environment": {
12
+ "python": "3.11.17",
13
+ "transformers": "5.21.0.dev0",
14
+ "torch": "2.14.0+cpu",
15
+ "torchvision": "0.29.1+cpu",
16
+ "torchaudio": "2.11.0+cpu",
17
+ "tokenizers": "0.23.3",
18
+ "numpy": "2.3.5",
19
+ "pillow": "12.3.0",
20
+ "accelerate": "1.15.0",
21
+ "safetensors": "0.8.0",
22
+ "huggingface-hub": "2.3.0",
23
+ "sentencepiece": "0.2.2",
24
+ "protobuf": "7.36.2",
25
+ "librosa": "0.11.0",
26
+ "soundfile": "0.14.0",
27
+ "scipy": "1.17.1"
28
+ },
29
+ "seconds": 10.64,
30
+ "environment_differs": {
31
+ "torch": "2.14.0+cpu (pinned 2.11.0)",
32
+ "torchvision": "0.29.1+cpu (pinned 0.26.0)",
33
+ "tokenizers": "0.23.3 (pinned 0.23.1)",
34
+ "numpy": "2.3.5 (pinned 1.26.4)",
35
+ "pillow": "12.3.0 (pinned 12.2.0)",
36
+ "accelerate": "1.15.0 (pinned 1.13.0)",
37
+ "huggingface-hub": "2.3.0 (pinned 1.32.0)",
38
+ "sentencepiece": "0.2.2 (pinned 0.2.1)",
39
+ "protobuf": "7.36.2 (pinned 6.33.6)",
40
+ "soundfile": "0.14.0 (pinned 0.13.1)",
41
+ "scipy": "1.17.1 (pinned 1.15.3)"
42
+ },
43
+ "platform": "cpu-x86_64"
44
+ },
45
+ "outputs": {
46
+ "runtime": {
47
+ "device": "cpu",
48
+ "dtype": "float32",
49
+ "attn_implementation": "eager"
50
+ },
51
+ "masked_logits": [
52
+ [
53
+ 0.011857529170811176,
54
+ -0.011009469628334045,
55
+ 0.020300013944506645,
56
+ -0.012026689946651459,
57
+ -0.027553845196962357
58
+ ],
59
+ [
60
+ 0.011856728233397007,
61
+ -0.011009469628334045,
62
+ 0.02030036970973015,
63
+ -0.012030145153403282,
64
+ -0.027552612125873566
65
+ ],
66
+ [
67
+ 0.011856435798108578,
68
+ -0.011008954606950283,
69
+ 0.020306482911109924,
70
+ -0.01202866155654192,
71
+ -0.02755468711256981
72
+ ]
73
+ ],
74
+ "training": {
75
+ "loss": 5.569273471832275,
76
+ "gradients": {
77
+ "perceiver.input_preprocessor.embeddings.weight": true,
78
+ "perceiver.encoder.self_attends.0.attention.self.query.weight": true
79
+ },
80
+ "updated": true
81
+ }
82
+ }
83
+ }
integrations/transformers/expectations/cpu-x86_64/language-perceiver-mic.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "recorded_with": {
3
+ "contract": "huggingface-mic/language-perceiver-mic",
4
+ "contract_name": "Perceiver",
5
+ "contract_revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
6
+ "framework": "transformers",
7
+ "variant": "default",
8
+ "target": "language-perceiver-mic",
9
+ "kind": "fixture",
10
+ "revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
11
+ "environment": {
12
+ "python": "3.11.17",
13
+ "transformers": "5.21.0.dev0",
14
+ "torch": "2.14.0+cpu",
15
+ "torchvision": "0.29.1+cpu",
16
+ "torchaudio": "2.11.0+cpu",
17
+ "tokenizers": "0.23.3",
18
+ "numpy": "2.3.5",
19
+ "pillow": "12.3.0",
20
+ "accelerate": "1.15.0",
21
+ "safetensors": "0.8.0",
22
+ "huggingface-hub": "2.3.0",
23
+ "sentencepiece": "0.2.2",
24
+ "protobuf": "7.36.2",
25
+ "librosa": "0.11.0",
26
+ "soundfile": "0.14.0",
27
+ "scipy": "1.17.1"
28
+ },
29
+ "seconds": 10.52,
30
+ "environment_differs": {
31
+ "torch": "2.14.0+cpu (pinned 2.11.0)",
32
+ "torchvision": "0.29.1+cpu (pinned 0.26.0)",
33
+ "tokenizers": "0.23.3 (pinned 0.23.1)",
34
+ "numpy": "2.3.5 (pinned 1.26.4)",
35
+ "pillow": "12.3.0 (pinned 12.2.0)",
36
+ "accelerate": "1.15.0 (pinned 1.13.0)",
37
+ "huggingface-hub": "2.3.0 (pinned 1.32.0)",
38
+ "sentencepiece": "0.2.2 (pinned 0.2.1)",
39
+ "protobuf": "7.36.2 (pinned 6.33.6)",
40
+ "soundfile": "0.14.0 (pinned 0.13.1)",
41
+ "scipy": "1.17.1 (pinned 1.15.3)"
42
+ },
43
+ "platform": "cpu-x86_64"
44
+ },
45
+ "outputs": {
46
+ "runtime": {
47
+ "device": "cpu",
48
+ "dtype": "float32",
49
+ "attn_implementation": "eager"
50
+ },
51
+ "masked_logits": [
52
+ [
53
+ 0.011857529170811176,
54
+ -0.011009469628334045,
55
+ 0.020300013944506645,
56
+ -0.012026689946651459,
57
+ -0.027553845196962357
58
+ ],
59
+ [
60
+ 0.011856728233397007,
61
+ -0.011009469628334045,
62
+ 0.02030036970973015,
63
+ -0.012030145153403282,
64
+ -0.027552612125873566
65
+ ],
66
+ [
67
+ 0.011856435798108578,
68
+ -0.011008954606950283,
69
+ 0.020306482911109924,
70
+ -0.01202866155654192,
71
+ -0.02755468711256981
72
+ ]
73
+ ],
74
+ "training": {
75
+ "loss": 5.569273471832275,
76
+ "gradients": {
77
+ "perceiver.input_preprocessor.embeddings.weight": true,
78
+ "perceiver.encoder.self_attends.0.attention.self.query.weight": true
79
+ },
80
+ "updated": true
81
+ }
82
+ }
83
+ }
integrations/transformers/expectations/cpu-x86_64/language-perceiver.eager.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "recorded_with": {
3
+ "contract": "huggingface-mic/language-perceiver-mic",
4
+ "contract_name": "Perceiver",
5
+ "contract_revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
6
+ "framework": "transformers",
7
+ "variant": "eager",
8
+ "target": "language-perceiver",
9
+ "kind": "checkpoint",
10
+ "revision": "47a1ba43d8198ea10cf9a158fce114953e9441a1",
11
+ "environment": {
12
+ "python": "3.11.17",
13
+ "transformers": "5.21.0.dev0",
14
+ "torch": "2.14.0+cpu",
15
+ "torchvision": "0.29.1+cpu",
16
+ "torchaudio": "2.11.0+cpu",
17
+ "tokenizers": "0.23.3",
18
+ "numpy": "2.3.5",
19
+ "pillow": "12.3.0",
20
+ "accelerate": "1.15.0",
21
+ "safetensors": "0.8.0",
22
+ "huggingface-hub": "2.3.0",
23
+ "sentencepiece": "0.2.2",
24
+ "protobuf": "7.36.2",
25
+ "librosa": "0.11.0",
26
+ "soundfile": "0.14.0",
27
+ "scipy": "1.17.1"
28
+ },
29
+ "seconds": 10.64,
30
+ "environment_differs": {
31
+ "torch": "2.14.0+cpu (pinned 2.11.0)",
32
+ "torchvision": "0.29.1+cpu (pinned 0.26.0)",
33
+ "tokenizers": "0.23.3 (pinned 0.23.1)",
34
+ "numpy": "2.3.5 (pinned 1.26.4)",
35
+ "pillow": "12.3.0 (pinned 12.2.0)",
36
+ "accelerate": "1.15.0 (pinned 1.13.0)",
37
+ "huggingface-hub": "2.3.0 (pinned 1.32.0)",
38
+ "sentencepiece": "0.2.2 (pinned 0.2.1)",
39
+ "protobuf": "7.36.2 (pinned 6.33.6)",
40
+ "soundfile": "0.14.0 (pinned 0.13.1)",
41
+ "scipy": "1.17.1 (pinned 1.15.3)"
42
+ },
43
+ "platform": "cpu-x86_64"
44
+ },
45
+ "outputs": {
46
+ "runtime": {
47
+ "device": "cpu",
48
+ "dtype": "float32",
49
+ "attn_implementation": "eager"
50
+ },
51
+ "masked_logits": [
52
+ [
53
+ -16.573013305664062,
54
+ -16.39983367919922,
55
+ -16.173620223999023,
56
+ -11.96962833404541,
57
+ -16.72991943359375
58
+ ],
59
+ [
60
+ -12.123590469360352,
61
+ -12.390275001525879,
62
+ -12.191612243652344,
63
+ -15.236956596374512,
64
+ -12.192231178283691
65
+ ],
66
+ [
67
+ -12.757885932922363,
68
+ -12.936495780944824,
69
+ -12.775068283081055,
70
+ -16.288785934448242,
71
+ -12.742189407348633
72
+ ]
73
+ ],
74
+ "training": {
75
+ "loss": 6.6892828941345215,
76
+ "gradients": {
77
+ "perceiver.input_preprocessor.embeddings.weight": true,
78
+ "perceiver.encoder.self_attends.0.attention.self.query.weight": true
79
+ },
80
+ "updated": true
81
+ }
82
+ }
83
+ }
integrations/transformers/expectations/cpu-x86_64/language-perceiver.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "recorded_with": {
3
+ "contract": "huggingface-mic/language-perceiver-mic",
4
+ "contract_name": "Perceiver",
5
+ "contract_revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
6
+ "framework": "transformers",
7
+ "variant": "default",
8
+ "target": "language-perceiver",
9
+ "kind": "checkpoint",
10
+ "revision": "47a1ba43d8198ea10cf9a158fce114953e9441a1",
11
+ "environment": {
12
+ "python": "3.11.17",
13
+ "transformers": "5.21.0.dev0",
14
+ "torch": "2.14.0+cpu",
15
+ "torchvision": "0.29.1+cpu",
16
+ "torchaudio": "2.11.0+cpu",
17
+ "tokenizers": "0.23.3",
18
+ "numpy": "2.3.5",
19
+ "pillow": "12.3.0",
20
+ "accelerate": "1.15.0",
21
+ "safetensors": "0.8.0",
22
+ "huggingface-hub": "2.3.0",
23
+ "sentencepiece": "0.2.2",
24
+ "protobuf": "7.36.2",
25
+ "librosa": "0.11.0",
26
+ "soundfile": "0.14.0",
27
+ "scipy": "1.17.1"
28
+ },
29
+ "seconds": 10.96,
30
+ "environment_differs": {
31
+ "torch": "2.14.0+cpu (pinned 2.11.0)",
32
+ "torchvision": "0.29.1+cpu (pinned 0.26.0)",
33
+ "tokenizers": "0.23.3 (pinned 0.23.1)",
34
+ "numpy": "2.3.5 (pinned 1.26.4)",
35
+ "pillow": "12.3.0 (pinned 12.2.0)",
36
+ "accelerate": "1.15.0 (pinned 1.13.0)",
37
+ "huggingface-hub": "2.3.0 (pinned 1.32.0)",
38
+ "sentencepiece": "0.2.2 (pinned 0.2.1)",
39
+ "protobuf": "7.36.2 (pinned 6.33.6)",
40
+ "soundfile": "0.14.0 (pinned 0.13.1)",
41
+ "scipy": "1.17.1 (pinned 1.15.3)"
42
+ },
43
+ "platform": "cpu-x86_64"
44
+ },
45
+ "outputs": {
46
+ "runtime": {
47
+ "device": "cpu",
48
+ "dtype": "float32",
49
+ "attn_implementation": "eager"
50
+ },
51
+ "masked_logits": [
52
+ [
53
+ -16.573013305664062,
54
+ -16.39983367919922,
55
+ -16.173620223999023,
56
+ -11.96962833404541,
57
+ -16.72991943359375
58
+ ],
59
+ [
60
+ -12.123590469360352,
61
+ -12.390275001525879,
62
+ -12.191612243652344,
63
+ -15.236956596374512,
64
+ -12.192231178283691
65
+ ],
66
+ [
67
+ -12.757885932922363,
68
+ -12.936495780944824,
69
+ -12.775068283081055,
70
+ -16.288785934448242,
71
+ -12.742189407348633
72
+ ]
73
+ ],
74
+ "training": {
75
+ "loss": 6.6892828941345215,
76
+ "gradients": {
77
+ "perceiver.input_preprocessor.embeddings.weight": true,
78
+ "perceiver.encoder.self_attends.0.attention.self.query.weight": true
79
+ },
80
+ "updated": true
81
+ }
82
+ }
83
+ }
integrations/transformers/expectations/cuda/language-perceiver-mic.bfloat16.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "recorded_with": {
3
+ "contract": "huggingface-mic/language-perceiver-mic",
4
+ "contract_name": "Perceiver",
5
+ "contract_revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
6
+ "framework": "transformers",
7
+ "variant": "bfloat16",
8
+ "target": "language-perceiver-mic",
9
+ "kind": "fixture",
10
+ "revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
11
+ "environment": {
12
+ "python": "3.11.18",
13
+ "transformers": "5.21.0.dev0",
14
+ "torch": "2.14.0+cu130",
15
+ "torchvision": "0.29.1+cu130",
16
+ "torchaudio": "2.11.0+cu130",
17
+ "tokenizers": "0.23.3",
18
+ "numpy": "2.4.6",
19
+ "pillow": "13.0.0",
20
+ "accelerate": "1.16.0.dev0",
21
+ "safetensors": "0.8.0",
22
+ "huggingface-hub": "2.3.0",
23
+ "sentencepiece": "0.2.2",
24
+ "protobuf": "7.36.2",
25
+ "librosa": "0.11.0",
26
+ "soundfile": "0.14.0",
27
+ "scipy": "1.17.1"
28
+ },
29
+ "seconds": 7.39,
30
+ "environment_differs": {
31
+ "torch": "2.14.0+cu130 (pinned 2.11.0)",
32
+ "torchvision": "0.29.1+cu130 (pinned 0.26.0)",
33
+ "tokenizers": "0.23.3 (pinned 0.23.1)",
34
+ "numpy": "2.4.6 (pinned 1.26.4)",
35
+ "pillow": "13.0.0 (pinned 12.2.0)",
36
+ "accelerate": "1.16.0.dev0 (pinned 1.13.0)",
37
+ "huggingface-hub": "2.3.0 (pinned 1.32.0)",
38
+ "sentencepiece": "0.2.2 (pinned 0.2.1)",
39
+ "protobuf": "7.36.2 (pinned 6.33.6)",
40
+ "soundfile": "0.14.0 (pinned 0.13.1)",
41
+ "scipy": "1.17.1 (pinned 1.15.3)"
42
+ },
43
+ "platform": "cuda"
44
+ },
45
+ "outputs": {
46
+ "runtime": {
47
+ "device": "cuda",
48
+ "dtype": "bfloat16",
49
+ "attn_implementation": "eager"
50
+ },
51
+ "masked_logits": [
52
+ [
53
+ 0.0118408203125,
54
+ -0.01129150390625,
55
+ 0.020263671875,
56
+ -0.01202392578125,
57
+ -0.0277099609375
58
+ ],
59
+ [
60
+ 0.0118408203125,
61
+ -0.01129150390625,
62
+ 0.020263671875,
63
+ -0.01202392578125,
64
+ -0.0277099609375
65
+ ],
66
+ [
67
+ 0.0118408203125,
68
+ -0.0113525390625,
69
+ 0.0203857421875,
70
+ -0.011962890625,
71
+ -0.0277099609375
72
+ ]
73
+ ],
74
+ "training": {
75
+ "loss": 5.5625,
76
+ "gradients": {
77
+ "perceiver.input_preprocessor.embeddings.weight": true,
78
+ "perceiver.encoder.self_attends.0.attention.self.query.weight": true
79
+ },
80
+ "updated": true
81
+ }
82
+ }
83
+ }
integrations/transformers/expectations/cuda/language-perceiver-mic.eager.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "recorded_with": {
3
+ "contract": "huggingface-mic/language-perceiver-mic",
4
+ "contract_name": "Perceiver",
5
+ "contract_revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
6
+ "framework": "transformers",
7
+ "variant": "eager",
8
+ "target": "language-perceiver-mic",
9
+ "kind": "fixture",
10
+ "revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
11
+ "environment": {
12
+ "python": "3.11.18",
13
+ "transformers": "5.21.0.dev0",
14
+ "torch": "2.14.0+cu130",
15
+ "torchvision": "0.29.1+cu130",
16
+ "torchaudio": "2.11.0+cu130",
17
+ "tokenizers": "0.23.3",
18
+ "numpy": "2.4.6",
19
+ "pillow": "13.0.0",
20
+ "accelerate": "1.16.0.dev0",
21
+ "safetensors": "0.8.0",
22
+ "huggingface-hub": "2.3.0",
23
+ "sentencepiece": "0.2.2",
24
+ "protobuf": "7.36.2",
25
+ "librosa": "0.11.0",
26
+ "soundfile": "0.14.0",
27
+ "scipy": "1.17.1"
28
+ },
29
+ "seconds": 7.62,
30
+ "environment_differs": {
31
+ "torch": "2.14.0+cu130 (pinned 2.11.0)",
32
+ "torchvision": "0.29.1+cu130 (pinned 0.26.0)",
33
+ "tokenizers": "0.23.3 (pinned 0.23.1)",
34
+ "numpy": "2.4.6 (pinned 1.26.4)",
35
+ "pillow": "13.0.0 (pinned 12.2.0)",
36
+ "accelerate": "1.16.0.dev0 (pinned 1.13.0)",
37
+ "huggingface-hub": "2.3.0 (pinned 1.32.0)",
38
+ "sentencepiece": "0.2.2 (pinned 0.2.1)",
39
+ "protobuf": "7.36.2 (pinned 6.33.6)",
40
+ "soundfile": "0.14.0 (pinned 0.13.1)",
41
+ "scipy": "1.17.1 (pinned 1.15.3)"
42
+ },
43
+ "platform": "cuda"
44
+ },
45
+ "outputs": {
46
+ "runtime": {
47
+ "device": "cuda",
48
+ "dtype": "float32",
49
+ "attn_implementation": "eager"
50
+ },
51
+ "masked_logits": [
52
+ [
53
+ 0.011857532896101475,
54
+ -0.01100946869701147,
55
+ 0.02030000649392605,
56
+ -0.012026693671941757,
57
+ -0.027553843334317207
58
+ ],
59
+ [
60
+ 0.011856730096042156,
61
+ -0.01100947242230177,
62
+ 0.02030036970973015,
63
+ -0.012030155397951603,
64
+ -0.027552619576454163
65
+ ],
66
+ [
67
+ 0.011856437660753727,
68
+ -0.011008958332240582,
69
+ 0.02030649036169052,
70
+ -0.012028663419187069,
71
+ -0.027554688975214958
72
+ ]
73
+ ],
74
+ "training": {
75
+ "loss": 5.569441318511963,
76
+ "gradients": {
77
+ "perceiver.input_preprocessor.embeddings.weight": true,
78
+ "perceiver.encoder.self_attends.0.attention.self.query.weight": true
79
+ },
80
+ "updated": true
81
+ }
82
+ }
83
+ }
integrations/transformers/expectations/cuda/language-perceiver-mic.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "recorded_with": {
3
+ "contract": "huggingface-mic/language-perceiver-mic",
4
+ "contract_name": "Perceiver",
5
+ "contract_revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
6
+ "framework": "transformers",
7
+ "variant": "default",
8
+ "target": "language-perceiver-mic",
9
+ "kind": "fixture",
10
+ "revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
11
+ "environment": {
12
+ "python": "3.11.18",
13
+ "transformers": "5.21.0.dev0",
14
+ "torch": "2.14.0+cu130",
15
+ "torchvision": "0.29.1+cu130",
16
+ "torchaudio": "2.11.0+cu130",
17
+ "tokenizers": "0.23.3",
18
+ "numpy": "2.4.6",
19
+ "pillow": "13.0.0",
20
+ "accelerate": "1.16.0.dev0",
21
+ "safetensors": "0.8.0",
22
+ "huggingface-hub": "2.3.0",
23
+ "sentencepiece": "0.2.2",
24
+ "protobuf": "7.36.2",
25
+ "librosa": "0.11.0",
26
+ "soundfile": "0.14.0",
27
+ "scipy": "1.17.1"
28
+ },
29
+ "seconds": 7.51,
30
+ "environment_differs": {
31
+ "torch": "2.14.0+cu130 (pinned 2.11.0)",
32
+ "torchvision": "0.29.1+cu130 (pinned 0.26.0)",
33
+ "tokenizers": "0.23.3 (pinned 0.23.1)",
34
+ "numpy": "2.4.6 (pinned 1.26.4)",
35
+ "pillow": "13.0.0 (pinned 12.2.0)",
36
+ "accelerate": "1.16.0.dev0 (pinned 1.13.0)",
37
+ "huggingface-hub": "2.3.0 (pinned 1.32.0)",
38
+ "sentencepiece": "0.2.2 (pinned 0.2.1)",
39
+ "protobuf": "7.36.2 (pinned 6.33.6)",
40
+ "soundfile": "0.14.0 (pinned 0.13.1)",
41
+ "scipy": "1.17.1 (pinned 1.15.3)"
42
+ },
43
+ "platform": "cuda"
44
+ },
45
+ "outputs": {
46
+ "runtime": {
47
+ "device": "cuda",
48
+ "dtype": "float32",
49
+ "attn_implementation": "eager"
50
+ },
51
+ "masked_logits": [
52
+ [
53
+ 0.011857532896101475,
54
+ -0.01100946869701147,
55
+ 0.02030000649392605,
56
+ -0.012026693671941757,
57
+ -0.027553843334317207
58
+ ],
59
+ [
60
+ 0.011856730096042156,
61
+ -0.01100947242230177,
62
+ 0.02030036970973015,
63
+ -0.012030155397951603,
64
+ -0.027552619576454163
65
+ ],
66
+ [
67
+ 0.011856437660753727,
68
+ -0.011008958332240582,
69
+ 0.02030649036169052,
70
+ -0.012028663419187069,
71
+ -0.027554688975214958
72
+ ]
73
+ ],
74
+ "training": {
75
+ "loss": 5.569441318511963,
76
+ "gradients": {
77
+ "perceiver.input_preprocessor.embeddings.weight": true,
78
+ "perceiver.encoder.self_attends.0.attention.self.query.weight": true
79
+ },
80
+ "updated": true
81
+ }
82
+ }
83
+ }
integrations/transformers/expectations/cuda/language-perceiver.bfloat16.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "recorded_with": {
3
+ "contract": "huggingface-mic/language-perceiver-mic",
4
+ "contract_name": "Perceiver",
5
+ "contract_revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
6
+ "framework": "transformers",
7
+ "variant": "bfloat16",
8
+ "target": "language-perceiver",
9
+ "kind": "checkpoint",
10
+ "revision": "47a1ba43d8198ea10cf9a158fce114953e9441a1",
11
+ "environment": {
12
+ "python": "3.11.18",
13
+ "transformers": "5.21.0.dev0",
14
+ "torch": "2.14.0+cu130",
15
+ "torchvision": "0.29.1+cu130",
16
+ "torchaudio": "2.11.0+cu130",
17
+ "tokenizers": "0.23.3",
18
+ "numpy": "2.4.6",
19
+ "pillow": "13.0.0",
20
+ "accelerate": "1.16.0.dev0",
21
+ "safetensors": "0.8.0",
22
+ "huggingface-hub": "2.3.0",
23
+ "sentencepiece": "0.2.2",
24
+ "protobuf": "7.36.2",
25
+ "librosa": "0.11.0",
26
+ "soundfile": "0.14.0",
27
+ "scipy": "1.17.1"
28
+ },
29
+ "seconds": 7.27,
30
+ "environment_differs": {
31
+ "torch": "2.14.0+cu130 (pinned 2.11.0)",
32
+ "torchvision": "0.29.1+cu130 (pinned 0.26.0)",
33
+ "tokenizers": "0.23.3 (pinned 0.23.1)",
34
+ "numpy": "2.4.6 (pinned 1.26.4)",
35
+ "pillow": "13.0.0 (pinned 12.2.0)",
36
+ "accelerate": "1.16.0.dev0 (pinned 1.13.0)",
37
+ "huggingface-hub": "2.3.0 (pinned 1.32.0)",
38
+ "sentencepiece": "0.2.2 (pinned 0.2.1)",
39
+ "protobuf": "7.36.2 (pinned 6.33.6)",
40
+ "soundfile": "0.14.0 (pinned 0.13.1)",
41
+ "scipy": "1.17.1 (pinned 1.15.3)"
42
+ },
43
+ "platform": "cuda"
44
+ },
45
+ "outputs": {
46
+ "runtime": {
47
+ "device": "cuda",
48
+ "dtype": "bfloat16",
49
+ "attn_implementation": "eager"
50
+ },
51
+ "masked_logits": [
52
+ [
53
+ -16.75,
54
+ -16.5,
55
+ -16.25,
56
+ -11.8125,
57
+ -16.875
58
+ ],
59
+ [
60
+ -12.1875,
61
+ -12.4375,
62
+ -12.25,
63
+ -15.25,
64
+ -12.25
65
+ ],
66
+ [
67
+ -12.75,
68
+ -12.9375,
69
+ -12.8125,
70
+ -16.375,
71
+ -12.75
72
+ ]
73
+ ],
74
+ "training": {
75
+ "loss": 6.8125,
76
+ "gradients": {
77
+ "perceiver.input_preprocessor.embeddings.weight": true,
78
+ "perceiver.encoder.self_attends.0.attention.self.query.weight": true
79
+ },
80
+ "updated": true
81
+ }
82
+ }
83
+ }
integrations/transformers/expectations/cuda/language-perceiver.eager.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "recorded_with": {
3
+ "contract": "huggingface-mic/language-perceiver-mic",
4
+ "contract_name": "Perceiver",
5
+ "contract_revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
6
+ "framework": "transformers",
7
+ "variant": "eager",
8
+ "target": "language-perceiver",
9
+ "kind": "checkpoint",
10
+ "revision": "47a1ba43d8198ea10cf9a158fce114953e9441a1",
11
+ "environment": {
12
+ "python": "3.11.18",
13
+ "transformers": "5.21.0.dev0",
14
+ "torch": "2.14.0+cu130",
15
+ "torchvision": "0.29.1+cu130",
16
+ "torchaudio": "2.11.0+cu130",
17
+ "tokenizers": "0.23.3",
18
+ "numpy": "2.4.6",
19
+ "pillow": "13.0.0",
20
+ "accelerate": "1.16.0.dev0",
21
+ "safetensors": "0.8.0",
22
+ "huggingface-hub": "2.3.0",
23
+ "sentencepiece": "0.2.2",
24
+ "protobuf": "7.36.2",
25
+ "librosa": "0.11.0",
26
+ "soundfile": "0.14.0",
27
+ "scipy": "1.17.1"
28
+ },
29
+ "seconds": 7.29,
30
+ "environment_differs": {
31
+ "torch": "2.14.0+cu130 (pinned 2.11.0)",
32
+ "torchvision": "0.29.1+cu130 (pinned 0.26.0)",
33
+ "tokenizers": "0.23.3 (pinned 0.23.1)",
34
+ "numpy": "2.4.6 (pinned 1.26.4)",
35
+ "pillow": "13.0.0 (pinned 12.2.0)",
36
+ "accelerate": "1.16.0.dev0 (pinned 1.13.0)",
37
+ "huggingface-hub": "2.3.0 (pinned 1.32.0)",
38
+ "sentencepiece": "0.2.2 (pinned 0.2.1)",
39
+ "protobuf": "7.36.2 (pinned 6.33.6)",
40
+ "soundfile": "0.14.0 (pinned 0.13.1)",
41
+ "scipy": "1.17.1 (pinned 1.15.3)"
42
+ },
43
+ "platform": "cuda"
44
+ },
45
+ "outputs": {
46
+ "runtime": {
47
+ "device": "cuda",
48
+ "dtype": "float32",
49
+ "attn_implementation": "eager"
50
+ },
51
+ "masked_logits": [
52
+ [
53
+ -16.573028564453125,
54
+ -16.399845123291016,
55
+ -16.173633575439453,
56
+ -11.969634056091309,
57
+ -16.729930877685547
58
+ ],
59
+ [
60
+ -12.123595237731934,
61
+ -12.390280723571777,
62
+ -12.191617012023926,
63
+ -15.236940383911133,
64
+ -12.19223690032959
65
+ ],
66
+ [
67
+ -12.757892608642578,
68
+ -12.936500549316406,
69
+ -12.775073051452637,
70
+ -16.288787841796875,
71
+ -12.742195129394531
72
+ ]
73
+ ],
74
+ "training": {
75
+ "loss": 6.059415817260742,
76
+ "gradients": {
77
+ "perceiver.input_preprocessor.embeddings.weight": true,
78
+ "perceiver.encoder.self_attends.0.attention.self.query.weight": true
79
+ },
80
+ "updated": true
81
+ }
82
+ }
83
+ }
integrations/transformers/expectations/cuda/language-perceiver.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "recorded_with": {
3
+ "contract": "huggingface-mic/language-perceiver-mic",
4
+ "contract_name": "Perceiver",
5
+ "contract_revision": "103ee7b7c5faff0bb516dc2550021d3e5453241a",
6
+ "framework": "transformers",
7
+ "variant": "default",
8
+ "target": "language-perceiver",
9
+ "kind": "checkpoint",
10
+ "revision": "47a1ba43d8198ea10cf9a158fce114953e9441a1",
11
+ "environment": {
12
+ "python": "3.11.18",
13
+ "transformers": "5.21.0.dev0",
14
+ "torch": "2.14.0+cu130",
15
+ "torchvision": "0.29.1+cu130",
16
+ "torchaudio": "2.11.0+cu130",
17
+ "tokenizers": "0.23.3",
18
+ "numpy": "2.4.6",
19
+ "pillow": "13.0.0",
20
+ "accelerate": "1.16.0.dev0",
21
+ "safetensors": "0.8.0",
22
+ "huggingface-hub": "2.3.0",
23
+ "sentencepiece": "0.2.2",
24
+ "protobuf": "7.36.2",
25
+ "librosa": "0.11.0",
26
+ "soundfile": "0.14.0",
27
+ "scipy": "1.17.1"
28
+ },
29
+ "seconds": 7.32,
30
+ "environment_differs": {
31
+ "torch": "2.14.0+cu130 (pinned 2.11.0)",
32
+ "torchvision": "0.29.1+cu130 (pinned 0.26.0)",
33
+ "tokenizers": "0.23.3 (pinned 0.23.1)",
34
+ "numpy": "2.4.6 (pinned 1.26.4)",
35
+ "pillow": "13.0.0 (pinned 12.2.0)",
36
+ "accelerate": "1.16.0.dev0 (pinned 1.13.0)",
37
+ "huggingface-hub": "2.3.0 (pinned 1.32.0)",
38
+ "sentencepiece": "0.2.2 (pinned 0.2.1)",
39
+ "protobuf": "7.36.2 (pinned 6.33.6)",
40
+ "soundfile": "0.14.0 (pinned 0.13.1)",
41
+ "scipy": "1.17.1 (pinned 1.15.3)"
42
+ },
43
+ "platform": "cuda"
44
+ },
45
+ "outputs": {
46
+ "runtime": {
47
+ "device": "cuda",
48
+ "dtype": "float32",
49
+ "attn_implementation": "eager"
50
+ },
51
+ "masked_logits": [
52
+ [
53
+ -16.573028564453125,
54
+ -16.399845123291016,
55
+ -16.173633575439453,
56
+ -11.969634056091309,
57
+ -16.729930877685547
58
+ ],
59
+ [
60
+ -12.123595237731934,
61
+ -12.390280723571777,
62
+ -12.191617012023926,
63
+ -15.236940383911133,
64
+ -12.19223690032959
65
+ ],
66
+ [
67
+ -12.757892608642578,
68
+ -12.936500549316406,
69
+ -12.775073051452637,
70
+ -16.288787841796875,
71
+ -12.742195129394531
72
+ ]
73
+ ],
74
+ "training": {
75
+ "loss": 6.059415817260742,
76
+ "gradients": {
77
+ "perceiver.input_preprocessor.embeddings.weight": true,
78
+ "perceiver.encoder.self_attends.0.attention.self.query.weight": true
79
+ },
80
+ "updated": true
81
+ }
82
+ }
83
+ }
integrations/transformers/usage.md ADDED
@@ -0,0 +1,80 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ # Perceiver: masked language modeling with Transformers
2
+
3
+ This page is the Transformers contract of the Perceiver MIC and also a usage example. [doc-builder](https://github.com/huggingface/doc-builder/blob/main/docs/runnable-code-blocks.md) renders the `runnable` blocks below as documentation, and its pytest plugin runs them as written against `deepmind/language-perceiver`. The MIC runner runs the same blocks against each slim fixture and each checkpoint in [`integration.json`](../../integration.json). Lines marked `# doc-builder: hide` only matter there: they swap in the pinned checkpoint or variant and print the result record.
4
+
5
+ ## Predict the masked bytes
6
+
7
+ ```py runnable:masked_lm
8
+ import torch
9
+ from transformers import PerceiverTokenizer, PerceiverForMaskedLM, set_seed
10
+ import json # doc-builder: hide
11
+ import os # doc-builder: hide
12
+
13
+ model_id = "deepmind/language-perceiver"
14
+ model_id = os.environ.get("MIC_MODEL", model_id) # doc-builder: hide
15
+ variant = json.loads(os.environ.get("MIC_MODEL_KWARGS", "{}")) # doc-builder: hide
16
+ device = "cuda" if torch.cuda.is_available() else "cpu"
17
+ set_seed(0)
18
+
19
+ tokenizer = PerceiverTokenizer.from_pretrained(model_id)
20
+ model = PerceiverForMaskedLM.from_pretrained(model_id).to(device)
21
+ if variant: # doc-builder: hide
22
+ del model # doc-builder: hide
23
+ model = PerceiverForMaskedLM.from_pretrained(model_id, **variant).to(device) # doc-builder: hide
24
+
25
+ # Two short sentences; the tokenizer maps each character to its byte value.
26
+ # The decoder predicts a fixed number of positions, so pad to that length.
27
+ texts = ["The cat sat on the mat.", "A dog ran in the park."]
28
+ encoding = tokenizer(texts, padding="max_length", max_length=model.config.max_position_embeddings, return_tensors="pt")
29
+ input_ids = encoding["input_ids"].to(device)
30
+ attention_mask = encoding["attention_mask"].to(device)
31
+
32
+ original_ids = input_ids.clone()
33
+
34
+ # Mask "mat" in the first sentence: characters 19-21, shifted by one for the leading [CLS] token
35
+ start = texts[0].index("mat") + 1
36
+ input_ids[0, start : start + 3] = tokenizer.mask_token_id
37
+
38
+ with torch.no_grad():
39
+ outputs = model(inputs=input_ids, attention_mask=attention_mask)
40
+ logits = outputs.logits
41
+
42
+ # Compact probe: first 5 logits at each masked position
43
+ masked = input_ids == tokenizer.mask_token_id
44
+ masked_logits = logits[masked]
45
+ print(masked_logits[:, :5].tolist())
46
+ ```
47
+
48
+ ## One training step
49
+
50
+ ```py runnable:masked_lm:2
51
+ # Labels: the original bytes; ignore pad positions with -100
52
+ labels = original_ids.clone()
53
+ labels[attention_mask == 0] = -100
54
+
55
+ model.train()
56
+ optimizer = torch.optim.SGD(model.parameters(), lr=0.01)
57
+ loss = model(inputs=input_ids, attention_mask=attention_mask, labels=labels).loss
58
+ loss.backward()
59
+ before = {n: p.detach().cpu().clone() for n, p in model.named_parameters() if p.requires_grad} # doc-builder: hide
60
+ optimizer.step()
61
+ print(f"loss: {loss.item():.4f}")
62
+
63
+ if os.environ.get("MIC_RUNNER"): # doc-builder: hide
64
+ named = dict(model.named_parameters())
65
+ checked = ("perceiver.input_preprocessor.embeddings.weight", "perceiver.encoder.self_attends.0.attention.self.query.weight")
66
+ result = {
67
+ "runtime": {
68
+ "device": model.device.type,
69
+ "dtype": str(model.dtype).removeprefix("torch."),
70
+ "attn_implementation": model.config._attn_implementation,
71
+ },
72
+ "masked_logits": masked_logits[:, :5].float().tolist(),
73
+ "training": {
74
+ "loss": loss.item(),
75
+ "gradients": {n: bool(named[n].grad is not None and torch.count_nonzero(named[n].grad) > 0 and torch.isfinite(named[n].grad).all()) for n in checked},
76
+ "updated": any(not torch.equal(p.detach().cpu(), before[n]) for n, p in model.named_parameters() if n in before),
77
+ },
78
+ }
79
+ print(json.dumps(result))
80
+ ```
model.safetensors ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:da10fb9094185afcaba694447b7e60e0c84b60345c2c13bb477187abc0ec8338
3
+ size 776511992
special_tokens_map.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"bos_token": {"content": "[BOS]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "eos_token": {"content": "[EOS]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "sep_token": {"content": "[SEP]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "pad_token": {"content": "[PAD]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "cls_token": {"content": "[CLS]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}, "mask_token": {"content": "[MASK]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true}}
tokenizer_config.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"pad_token": {"content": "[PAD]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "bos_token": {"content": "[BOS]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "eos_token": {"content": "[EOS]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "mask_token": {"content": "[MASK]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "cls_token": {"content": "[CLS]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "sep_token": {"content": "[SEP]", "single_word": false, "lstrip": false, "rstrip": false, "normalized": true, "__type": "AddedToken"}, "model_max_length": 2048, "tokenizer_class": "PerceiverTokenizer"}