Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
 Community
dacorvo HF Staff commited on
Commit
41e8c23
·
verified ·
1 Parent(s): 4bbee00

Synchronizing local compiler cache.

Browse files
Files changed (12) hide show
  1. neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev3/dbd3bd79e50c8457d4e4475c9dcbeb0f9af6a8f49b1819904fe560bfed266d11/22445ab3999f2d0e855b.json +142 -0
  2. neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev3/smollm3/HuggingFaceTB/SmolLM3-3B-Base/22445ab3999f2d0e855b.json +142 -0
  3. neuronxcc-2.21.33363.0+82129205/MODULE_1759fced9dd9a678973e+24129607/model.hlo_module.pb +1 -1
  4. neuronxcc-2.21.33363.0+82129205/MODULE_1759fced9dd9a678973e+24129607/model.neff +1 -1
  5. neuronxcc-2.21.33363.0+82129205/MODULE_4ff804c2533da46e613e+24129607/model.hlo_module.pb +1 -1
  6. neuronxcc-2.21.33363.0+82129205/MODULE_4ff804c2533da46e613e+24129607/model.neff +1 -1
  7. neuronxcc-2.21.33363.0+82129205/MODULE_5742ef8f5910f4e17377+a02c3a36/model.hlo_module.pb +1 -1
  8. neuronxcc-2.21.33363.0+82129205/MODULE_5742ef8f5910f4e17377+a02c3a36/model.neff +1 -1
  9. neuronxcc-2.21.33363.0+82129205/MODULE_5742ef8f5910f4e17377+a02c3a36/wrapped_neff.hlo +1 -1
  10. neuronxcc-2.21.33363.0+82129205/MODULE_88833531cb6c582d4830+a02c3a36/model.hlo_module.pb +1 -1
  11. neuronxcc-2.21.33363.0+82129205/MODULE_88833531cb6c582d4830+a02c3a36/model.neff +1 -1
  12. neuronxcc-2.21.33363.0+82129205/MODULE_88833531cb6c582d4830+a02c3a36/wrapped_neff.hlo +1 -1
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev3/dbd3bd79e50c8457d4e4475c9dcbeb0f9af6a8f49b1819904fe560bfed266d11/22445ab3999f2d0e855b.json ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "HuggingFaceTB/SmolLM3-3B-Base",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "SmolLM3ForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "dtype": "bfloat16",
11
+ "hidden_act": "silu",
12
+ "hidden_size": 2048,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 11008,
15
+ "layer_types": [
16
+ "full_attention",
17
+ "full_attention",
18
+ "full_attention",
19
+ "full_attention",
20
+ "full_attention",
21
+ "full_attention",
22
+ "full_attention",
23
+ "full_attention",
24
+ "full_attention",
25
+ "full_attention",
26
+ "full_attention",
27
+ "full_attention",
28
+ "full_attention",
29
+ "full_attention",
30
+ "full_attention",
31
+ "full_attention",
32
+ "full_attention",
33
+ "full_attention",
34
+ "full_attention",
35
+ "full_attention",
36
+ "full_attention",
37
+ "full_attention",
38
+ "full_attention",
39
+ "full_attention",
40
+ "full_attention",
41
+ "full_attention",
42
+ "full_attention",
43
+ "full_attention",
44
+ "full_attention",
45
+ "full_attention",
46
+ "full_attention",
47
+ "full_attention",
48
+ "full_attention",
49
+ "full_attention",
50
+ "full_attention",
51
+ "full_attention"
52
+ ],
53
+ "max_position_embeddings": 65536,
54
+ "mlp_bias": false,
55
+ "model_type": "smollm3",
56
+ "neuron": {
57
+ "_serialized_key": "NxDNeuronConfig",
58
+ "batch_size": 4,
59
+ "capacity_factor": null,
60
+ "checkpoint_id": "HuggingFaceTB/SmolLM3-3B-Base",
61
+ "checkpoint_revision": "d78a42f79198603e614095753484a04c10c2b940",
62
+ "continuous_batching": true,
63
+ "ep_degree": 1,
64
+ "fused_qkv": true,
65
+ "glu_mlp": true,
66
+ "local_ranks_size": 2,
67
+ "max_batch_size": 4,
68
+ "max_context_length": 1024,
69
+ "max_topk": 256,
70
+ "n_active_tokens": 1024,
71
+ "neuronxcc_version": "2.21.33363.0+82129205",
72
+ "on_device_sampling": true,
73
+ "optimum_neuron_version": "0.4.6.dev3",
74
+ "output_logits": false,
75
+ "pp_degree": 1,
76
+ "prefill_chunk_size": 0,
77
+ "sequence_length": 1024,
78
+ "speculation_length": 0,
79
+ "start_rank_id": 0,
80
+ "target": "trn1",
81
+ "torch_dtype": "bfloat16",
82
+ "tp_degree": 2
83
+ },
84
+ "no_rope_layer_interval": 4,
85
+ "no_rope_layers": [
86
+ 1,
87
+ 1,
88
+ 1,
89
+ 0,
90
+ 1,
91
+ 1,
92
+ 1,
93
+ 0,
94
+ 1,
95
+ 1,
96
+ 1,
97
+ 0,
98
+ 1,
99
+ 1,
100
+ 1,
101
+ 0,
102
+ 1,
103
+ 1,
104
+ 1,
105
+ 0,
106
+ 1,
107
+ 1,
108
+ 1,
109
+ 0,
110
+ 1,
111
+ 1,
112
+ 1,
113
+ 0,
114
+ 1,
115
+ 1,
116
+ 1,
117
+ 0,
118
+ 1,
119
+ 1,
120
+ 1,
121
+ 0
122
+ ],
123
+ "num_attention_heads": 16,
124
+ "num_hidden_layers": 36,
125
+ "num_key_value_heads": 4,
126
+ "pretraining_tp": 2,
127
+ "rms_norm_eps": 1e-06,
128
+ "rope_scaling": null,
129
+ "rope_theta": 5000000.0,
130
+ "sliding_window": null,
131
+ "transformers.js_config": {
132
+ "dtype": "q4",
133
+ "kv_cache_dtype": {
134
+ "fp16": "float16",
135
+ "q4f16": "float16"
136
+ },
137
+ "use_external_data_format": true
138
+ },
139
+ "use_cache": true,
140
+ "use_sliding_window": false,
141
+ "vocab_size": 128256
142
+ }
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev3/smollm3/HuggingFaceTB/SmolLM3-3B-Base/22445ab3999f2d0e855b.json ADDED
@@ -0,0 +1,142 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "HuggingFaceTB/SmolLM3-3B-Base",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "SmolLM3ForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "dtype": "bfloat16",
11
+ "hidden_act": "silu",
12
+ "hidden_size": 2048,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 11008,
15
+ "layer_types": [
16
+ "full_attention",
17
+ "full_attention",
18
+ "full_attention",
19
+ "full_attention",
20
+ "full_attention",
21
+ "full_attention",
22
+ "full_attention",
23
+ "full_attention",
24
+ "full_attention",
25
+ "full_attention",
26
+ "full_attention",
27
+ "full_attention",
28
+ "full_attention",
29
+ "full_attention",
30
+ "full_attention",
31
+ "full_attention",
32
+ "full_attention",
33
+ "full_attention",
34
+ "full_attention",
35
+ "full_attention",
36
+ "full_attention",
37
+ "full_attention",
38
+ "full_attention",
39
+ "full_attention",
40
+ "full_attention",
41
+ "full_attention",
42
+ "full_attention",
43
+ "full_attention",
44
+ "full_attention",
45
+ "full_attention",
46
+ "full_attention",
47
+ "full_attention",
48
+ "full_attention",
49
+ "full_attention",
50
+ "full_attention",
51
+ "full_attention"
52
+ ],
53
+ "max_position_embeddings": 65536,
54
+ "mlp_bias": false,
55
+ "model_type": "smollm3",
56
+ "neuron": {
57
+ "_serialized_key": "NxDNeuronConfig",
58
+ "batch_size": 4,
59
+ "capacity_factor": null,
60
+ "checkpoint_id": "HuggingFaceTB/SmolLM3-3B-Base",
61
+ "checkpoint_revision": "d78a42f79198603e614095753484a04c10c2b940",
62
+ "continuous_batching": true,
63
+ "ep_degree": 1,
64
+ "fused_qkv": true,
65
+ "glu_mlp": true,
66
+ "local_ranks_size": 2,
67
+ "max_batch_size": 4,
68
+ "max_context_length": 1024,
69
+ "max_topk": 256,
70
+ "n_active_tokens": 1024,
71
+ "neuronxcc_version": "2.21.33363.0+82129205",
72
+ "on_device_sampling": true,
73
+ "optimum_neuron_version": "0.4.6.dev3",
74
+ "output_logits": false,
75
+ "pp_degree": 1,
76
+ "prefill_chunk_size": 0,
77
+ "sequence_length": 1024,
78
+ "speculation_length": 0,
79
+ "start_rank_id": 0,
80
+ "target": "trn1",
81
+ "torch_dtype": "bfloat16",
82
+ "tp_degree": 2
83
+ },
84
+ "no_rope_layer_interval": 4,
85
+ "no_rope_layers": [
86
+ 1,
87
+ 1,
88
+ 1,
89
+ 0,
90
+ 1,
91
+ 1,
92
+ 1,
93
+ 0,
94
+ 1,
95
+ 1,
96
+ 1,
97
+ 0,
98
+ 1,
99
+ 1,
100
+ 1,
101
+ 0,
102
+ 1,
103
+ 1,
104
+ 1,
105
+ 0,
106
+ 1,
107
+ 1,
108
+ 1,
109
+ 0,
110
+ 1,
111
+ 1,
112
+ 1,
113
+ 0,
114
+ 1,
115
+ 1,
116
+ 1,
117
+ 0,
118
+ 1,
119
+ 1,
120
+ 1,
121
+ 0
122
+ ],
123
+ "num_attention_heads": 16,
124
+ "num_hidden_layers": 36,
125
+ "num_key_value_heads": 4,
126
+ "pretraining_tp": 2,
127
+ "rms_norm_eps": 1e-06,
128
+ "rope_scaling": null,
129
+ "rope_theta": 5000000.0,
130
+ "sliding_window": null,
131
+ "transformers.js_config": {
132
+ "dtype": "q4",
133
+ "kv_cache_dtype": {
134
+ "fp16": "float16",
135
+ "q4f16": "float16"
136
+ },
137
+ "use_external_data_format": true
138
+ },
139
+ "use_cache": true,
140
+ "use_sliding_window": false,
141
+ "vocab_size": 128256
142
+ }
neuronxcc-2.21.33363.0+82129205/MODULE_1759fced9dd9a678973e+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8223a0880f04437df80f89de1868b68be53988eb9a109c31fbf2423e0595ab55
3
  size 684041
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9b12e4bc9a9646f83987e5eb925b237f68f4d7a5569ef4a101e91e22748bfcb6
3
  size 684041
neuronxcc-2.21.33363.0+82129205/MODULE_1759fced9dd9a678973e+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:716d2d5bdbafb617fe64ce7e95b970d04f7da3600f9ea8defbaecd511d634961
3
  size 7732224
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a849a1d1b254aef58958b0a532465d7fd2588408c54ce33e87d87c11fb2871cc
3
  size 7732224
neuronxcc-2.21.33363.0+82129205/MODULE_4ff804c2533da46e613e+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:77aa6e56466f63a602dedff46d9989619913f612654147b3c637f7d3de61f7bb
3
  size 1050679
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:868cca689a34e99963ba4838928c3d05b3b7ff0c504770b6e3b057f912c499df
3
  size 1050679
neuronxcc-2.21.33363.0+82129205/MODULE_4ff804c2533da46e613e+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:34a2c883e2aa715e151451bee73497a2f06cc46696d7df7deaa3002177def80f
3
  size 14050304
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7092c426f768f52ed7be2a5ae675118ff348d21dc203e5a969ad6da48a833fb0
3
  size 14050304
neuronxcc-2.21.33363.0+82129205/MODULE_5742ef8f5910f4e17377+a02c3a36/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7c2a0e27aea45eec39afceeccd973f758e74d7f2b39164586a4bd50817a82017
3
  size 1117688
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b5e7fcdf9f65af1b99f353b41bfb54288571325ca47b98d54e3e989a576ff880
3
  size 1117688
neuronxcc-2.21.33363.0+82129205/MODULE_5742ef8f5910f4e17377+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d1a407e255505988a25235719e0362bc0b25a57529345a8e1a677f7e3958cd16
3
  size 5387264
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:29f3a621f39fe257c700d6bfc0b4b4ab251957d63970b10bb8ecf19501d9f36a
3
  size 5387264
neuronxcc-2.21.33363.0+82129205/MODULE_5742ef8f5910f4e17377+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:134d868a2ff9c32bba7e5cee35716ccddbe5d981756fb94db39373e8bbae9aa9
3
  size 5571457
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61c0c9873fdc5c05d11f5911176dcef5b6049e8a5c619cd4e51ed757c641e98e
3
  size 5571457
neuronxcc-2.21.33363.0+82129205/MODULE_88833531cb6c582d4830+a02c3a36/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a344a8bfb0bd70d150a25d375f0a7ad1432403ec9c9a9455db46debc4a8e84eb
3
  size 727217
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:97aecb8c044acc0ebbfea336087df614b77fd28df2dfa780403927456239a624
3
  size 727217
neuronxcc-2.21.33363.0+82129205/MODULE_88833531cb6c582d4830+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:25bc5e849b16d05bd4e3955ff08c5681b576bef793e02c4e34a4578f4a537922
3
  size 3472384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e8768b115e2249ec808a67cd0136f0169269ce5e8186bbae6a46c21db2c0594
3
  size 3472384
neuronxcc-2.21.33363.0+82129205/MODULE_88833531cb6c582d4830+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a8145a60228c00cae89502e7d9814a447a7f0719dc42e45698b88165002e717a
3
  size 3610166
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7bf0374f79f6879ee32742c88e40cd6578de4b103c9b299baea56cb034b5e7ab
3
  size 3610166