Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
Community
optimum-internal-testing-user commited on
Commit
dff8bf4
·
verified ·
1 Parent(s): 775d8f3

Synchronizing local compiler cache.

Browse files
Files changed (9) hide show
  1. neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.1.dev0/smollm3/HuggingFaceTB/SmolLM3-3B/9caa0945eebe2d5114ca.json +134 -0
  2. neuronxcc-2.21.18209.0+043b1bf7/MODULE_4be33b117e68e2de3326+a9d440f5/model.neff +1 -1
  3. neuronxcc-2.21.18209.0+043b1bf7/MODULE_4be33b117e68e2de3326+a9d440f5/wrapped_neff.hlo +1 -1
  4. neuronxcc-2.21.18209.0+043b1bf7/MODULE_50d30ce495feb34fe5c1+a9d440f5/model.neff +1 -1
  5. neuronxcc-2.21.18209.0+043b1bf7/MODULE_50d30ce495feb34fe5c1+a9d440f5/wrapped_neff.hlo +1 -1
  6. neuronxcc-2.21.18209.0+043b1bf7/MODULE_73b50e8e61409e1f8fe1+ed72d204/model.hlo_module.pb +1 -1
  7. neuronxcc-2.21.18209.0+043b1bf7/MODULE_73b50e8e61409e1f8fe1+ed72d204/model.neff +1 -1
  8. neuronxcc-2.21.18209.0+043b1bf7/MODULE_cae14ab78f54c8b4e73f+ed72d204/model.hlo_module.pb +1 -1
  9. neuronxcc-2.21.18209.0+043b1bf7/MODULE_cae14ab78f54c8b4e73f+ed72d204/model.neff +1 -1
neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.1.dev0/smollm3/HuggingFaceTB/SmolLM3-3B/9caa0945eebe2d5114ca.json ADDED
@@ -0,0 +1,134 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "HuggingFaceTB/SmolLM3-3B",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "SmolLM3ForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "hidden_act": "silu",
11
+ "hidden_size": 2048,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 11008,
14
+ "layer_types": [
15
+ "full_attention",
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
+ ],
52
+ "max_position_embeddings": 65536,
53
+ "max_window_layers": 28,
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",
61
+ "checkpoint_revision": "a07cc9a04f16550a088caea529712d1d335b0ac1",
62
+ "continuous_batching": true,
63
+ "enable_bucketing": false,
64
+ "ep_degree": 1,
65
+ "fused_qkv": true,
66
+ "glu_mlp": true,
67
+ "local_ranks_size": 2,
68
+ "max_batch_size": 4,
69
+ "max_context_length": 4096,
70
+ "max_topk": 256,
71
+ "n_active_tokens": 4096,
72
+ "neuronxcc_version": "2.21.18209.0+043b1bf7",
73
+ "on_device_sampling": true,
74
+ "optimum_neuron_version": "0.4.1.dev0",
75
+ "output_logits": false,
76
+ "pp_degree": 1,
77
+ "sequence_length": 4096,
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
+ "use_cache": false,
132
+ "use_sliding_window": false,
133
+ "vocab_size": 128256
134
+ }
neuronxcc-2.21.18209.0+043b1bf7/MODULE_4be33b117e68e2de3326+a9d440f5/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0cd0dd1029ad343d311182a5e633595c05978671b33b56acf65e65368824d43b
3
  size 3093504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d70d650de28d7b13defbb5f9d78b06f99ec63ba26cda803845b298cbca3d0581
3
  size 3093504
neuronxcc-2.21.18209.0+043b1bf7/MODULE_4be33b117e68e2de3326+a9d440f5/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:35ad902abd90ec49120c5aafd4769fc1a23a9c186b65dfbd97c48053401daf0d
3
  size 3167748
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d0d1b1e40a092a3eec560354aeb1191196e8b39c2312249a4d6f358c7ef2a51
3
  size 3167748
neuronxcc-2.21.18209.0+043b1bf7/MODULE_50d30ce495feb34fe5c1+a9d440f5/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ac9f3679002992b1656af6c058f7ac32d474f4e8d1eb85ff012f3fe9a13f4b61
3
  size 1926144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:af93b3d02b40084bdec001a9496f6b98115e3bfebad2f126075400a4c60fbd87
3
  size 1926144
neuronxcc-2.21.18209.0+043b1bf7/MODULE_50d30ce495feb34fe5c1+a9d440f5/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3bcff4445fc89f39652209d48d0c50067ba2c86b62b3a8d57388c2a222b3fba1
3
  size 2082478
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6951f83519dce668f7723834939eacc71fdb113475631340f061d1db9cd51852
3
  size 2082478
neuronxcc-2.21.18209.0+043b1bf7/MODULE_73b50e8e61409e1f8fe1+ed72d204/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:637e1374e15bf184dd2bb4e3d96b85bd9be0e22c5fd09c32e2c8de0cafcf7e80
3
  size 509252
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:441fac3e96e30c894231d7f6dd20d4e89e6b4cf1295ed681a96fb69af4ea7582
3
  size 509252
neuronxcc-2.21.18209.0+043b1bf7/MODULE_73b50e8e61409e1f8fe1+ed72d204/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ecc8cd2933d03bd72923e70db0219d9f5eb1266f43f6f91fc9ca077ef6802850
3
  size 41585664
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ccfc71e4eaf13a28e93ac88d5f63f2d222b919a9db2e9f4a966422a8f787f591
3
  size 41585664
neuronxcc-2.21.18209.0+043b1bf7/MODULE_cae14ab78f54c8b4e73f+ed72d204/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:78c8d3d532959cd2c094d5636aea4eb94bfeb288aa58e9c67b64bab124d47d0d
3
  size 739366
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b8511b22bc58244c0086580ec7d72532553a8f93a374268bdb1fbb737acfc3a4
3
  size 739366
neuronxcc-2.21.18209.0+043b1bf7/MODULE_cae14ab78f54c8b4e73f+ed72d204/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a9b55de97b0d5ccfea4ce04bc3f98a36b49c0a1d4d0bfbf598f4570587737a2f
3
  size 26133504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3ffb3d7bc9903f8346667ae3bc452aa306ed217fe3fc73cd25365d476c98eac8
3
  size 26133504