Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
 Community
optimum-internal-testing-user commited on
Commit
f6a91e1
·
verified ·
1 Parent(s): 96b426b

Synchronizing local compiler cache.

Browse files
Files changed (6) hide show
  1. neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7.dev0/smollm3/HuggingFaceTB/SmolLM3-3B/2754c7a28eac1a918f8c.json +138 -0
  2. neuronxcc-2.26.6360.0+6f180f47/MODULE_14eb69eb2a634838a288+519f203d/model.hlo_module.pb +1 -1
  3. neuronxcc-2.26.6360.0+6f180f47/MODULE_14eb69eb2a634838a288+519f203d/model.neff +1 -1
  4. neuronxcc-2.26.6360.0+6f180f47/MODULE_78eee8e882426dc56f47+84337dd9/model.hlo_module.pb +1 -1
  5. neuronxcc-2.26.6360.0+6f180f47/MODULE_78eee8e882426dc56f47+84337dd9/model.neff +2 -2
  6. neuronxcc-2.26.6360.0+6f180f47/MODULE_78eee8e882426dc56f47+84337dd9/wrapped_neff.hlo +2 -2
neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7.dev0/smollm3/HuggingFaceTB/SmolLM3-3B/2754c7a28eac1a918f8c.json ADDED
@@ -0,0 +1,138 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "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
+ "max_window_layers": 28,
55
+ "mlp_bias": false,
56
+ "model_type": "smollm3",
57
+ "neuron": {
58
+ "_serialized_key": "NxDNeuronConfig",
59
+ "batch_size": 4,
60
+ "capacity_factor": null,
61
+ "checkpoint_id": "HuggingFaceTB/SmolLM3-3B",
62
+ "checkpoint_revision": "a07cc9a04f16550a088caea529712d1d335b0ac1",
63
+ "continuous_batching": true,
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": 1024,
70
+ "max_topk": 256,
71
+ "n_active_tokens": 1024,
72
+ "neuronxcc_version": "2.26.6360.0+6f180f47",
73
+ "on_device_sampling": true,
74
+ "optimum_neuron_version": "0.4.7.dev0",
75
+ "output_logits": false,
76
+ "pp_degree": 1,
77
+ "prefill_chunk_size": 0,
78
+ "sequence_length": 1024,
79
+ "speculation_length": 0,
80
+ "start_rank_id": 0,
81
+ "target": "trn1",
82
+ "torch_dtype": "bfloat16",
83
+ "tp_degree": 2
84
+ },
85
+ "no_rope_layer_interval": 4,
86
+ "no_rope_layers": [
87
+ 1,
88
+ 1,
89
+ 1,
90
+ 0,
91
+ 1,
92
+ 1,
93
+ 1,
94
+ 0,
95
+ 1,
96
+ 1,
97
+ 1,
98
+ 0,
99
+ 1,
100
+ 1,
101
+ 1,
102
+ 0,
103
+ 1,
104
+ 1,
105
+ 1,
106
+ 0,
107
+ 1,
108
+ 1,
109
+ 1,
110
+ 0,
111
+ 1,
112
+ 1,
113
+ 1,
114
+ 0,
115
+ 1,
116
+ 1,
117
+ 1,
118
+ 0,
119
+ 1,
120
+ 1,
121
+ 1,
122
+ 0
123
+ ],
124
+ "num_attention_heads": 16,
125
+ "num_hidden_layers": 36,
126
+ "num_key_value_heads": 4,
127
+ "pretraining_tp": 2,
128
+ "rms_norm_eps": 1e-06,
129
+ "rope_parameters": {
130
+ "rope_theta": 5000000.0,
131
+ "rope_type": "default"
132
+ },
133
+ "sliding_window": null,
134
+ "tie_word_embeddings": true,
135
+ "use_cache": false,
136
+ "use_sliding_window": false,
137
+ "vocab_size": 128256
138
+ }
neuronxcc-2.26.6360.0+6f180f47/MODULE_14eb69eb2a634838a288+519f203d/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:185aa220d6b1d830603dcbb187074dc1e617ade83ef15966a2843858e879f8b2
3
  size 810052
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b7d79938f385bab8e01df87c737ad2ea761a2e0c3158ee4b8cf48e6b61e92468
3
  size 810052
neuronxcc-2.26.6360.0+6f180f47/MODULE_14eb69eb2a634838a288+519f203d/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:64b2e67c779ff0ec1f7e7b9e063bfc2407e5f8bf8717e9d1348a73b801c8f265
3
  size 2653184
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5fa7480efe3c7982cad60b7f191beca0a0e11ced2030ba499af10b5f3bf7c3d9
3
  size 2653184
neuronxcc-2.26.6360.0+6f180f47/MODULE_78eee8e882426dc56f47+84337dd9/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3cae6d5ffcf3e30f267613da5a3486f8f53b2a2342d0a0b2870ff4480342c832
3
  size 847851
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1daba7941b6dd4f2e18a181986485eae247cd7e5aad9012851398caa19964b73
3
  size 847851
neuronxcc-2.26.6360.0+6f180f47/MODULE_78eee8e882426dc56f47+84337dd9/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b6f2404f5c119fbd6ea3cbfec3808c2a25f13c190736c64d17bf2e19709f6c58
3
- size 5018624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5e6597c4a5b3ae1efdfa4efa3ffae2aab18a27dfc343adbd5ba4cb992960cd5b
3
+ size 5008384
neuronxcc-2.26.6360.0+6f180f47/MODULE_78eee8e882426dc56f47+84337dd9/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:20286c34ec76e2dcb03dd42fc78d1b8aa74a467a56a7638753993b23696a4888
3
- size 5185148
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:33be5126a35de6b43078d747ecaa02adfccba1c6f332b365985a54325e8a7e21
3
+ size 5174908