Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
 Community
tengomucho HF Staff commited on
Commit
c7c42bb
·
verified ·
1 Parent(s): 6ee35bc

Synchronizing local compiler cache.

Browse files
Files changed (6) hide show
  1. neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/smollm3/HuggingFaceTB/SmolLM3-3B/c100f806a6006feef39b.json +134 -0
  2. neuronxcc-2.21.33363.0+82129205/MODULE_26a5b03803fa3c302f48+ac10809c/model.hlo_module.pb +1 -1
  3. neuronxcc-2.21.33363.0+82129205/MODULE_26a5b03803fa3c302f48+ac10809c/model.neff +1 -1
  4. neuronxcc-2.21.33363.0+82129205/MODULE_26a5b03803fa3c302f48+ac10809c/wrapped_neff.hlo +1 -1
  5. neuronxcc-2.21.33363.0+82129205/MODULE_bb55057d1263c7ecc9f1+a32116a7/model.hlo_module.pb +1 -1
  6. neuronxcc-2.21.33363.0+82129205/MODULE_bb55057d1263c7ecc9f1+a32116a7/model.neff +1 -1
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/smollm3/HuggingFaceTB/SmolLM3-3B/c100f806a6006feef39b.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
+ "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": 4,
68
+ "max_batch_size": 4,
69
+ "max_context_length": 4096,
70
+ "max_topk": 256,
71
+ "n_active_tokens": 4096,
72
+ "neuronxcc_version": "2.21.33363.0+82129205",
73
+ "on_device_sampling": true,
74
+ "optimum_neuron_version": "0.4.5.dev2",
75
+ "output_logits": false,
76
+ "pp_degree": 1,
77
+ "sequence_length": 4096,
78
+ "speculation_length": 0,
79
+ "start_rank_id": 0,
80
+ "target": "trn2",
81
+ "torch_dtype": "bfloat16",
82
+ "tp_degree": 4
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-2.21.33363.0+82129205/MODULE_26a5b03803fa3c302f48+ac10809c/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7f54cddffcbeb5a11cd23081e943f267ba5eb801d100d078e14551dfa11e5740
3
  size 2471814
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:515923c3030d2428381c1e339c579aa8c1bca8c13c4126d52b584b403ea7f239
3
  size 2471814
neuronxcc-2.21.33363.0+82129205/MODULE_26a5b03803fa3c302f48+ac10809c/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e9a06677efada90633726543132fcc258b66e3766a4af867b1ebaa4f146e64b5
3
  size 4465664
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:09f88c02ac05549bd61ac3eaadac45d919c5a2503d49bf8f9ce2db019a290b98
3
  size 4465664
neuronxcc-2.21.33363.0+82129205/MODULE_26a5b03803fa3c302f48+ac10809c/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:23aef34dc31d1b6150d943fa0e702f501cb7d45d9291ecdd44255e0882ffa426
3
  size 4633208
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:5f1fe13c618c3c040c37a95b308340e5edee22cd8697b6e9cf5edcd89721a3eb
3
  size 4633208
neuronxcc-2.21.33363.0+82129205/MODULE_bb55057d1263c7ecc9f1+a32116a7/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c452936fbd1bb0d11fe021e472f5e67a606740d403eabed84a377220da560734
3
  size 2269712
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3a597ba392f32fa74c6dd19878ca60b30e654816519f1f54a05fa02bad0bdd39
3
  size 2269712
neuronxcc-2.21.33363.0+82129205/MODULE_bb55057d1263c7ecc9f1+a32116a7/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d64b9ba0b7cf4df0c6b8662d611aaa976162ca3596b13e10dbe3631d2976f512
3
  size 34028544
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fe2bbe7cfbaff44a575a1986cd4085a36d70c72b9740e8666c06bb7a7230b6cd
3
  size 34028544