Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
 Community
optimum-internal-testing-user commited on
Commit
8ffdffa
·
verified ·
1 Parent(s): ed2fc23

Synchronizing local compiler cache.

Browse files
Files changed (6) hide show
  1. neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev4/qwen3/Qwen/Qwen3-0.6B/1c88c352d1e2acf2c0db.json +88 -0
  2. neuronxcc-2.21.33363.0+82129205/MODULE_1759fced9dd9a678973e+24129607/model.hlo_module.pb +1 -1
  3. neuronxcc-2.21.33363.0+82129205/MODULE_1759fced9dd9a678973e+24129607/model.neff +1 -1
  4. neuronxcc-2.21.33363.0+82129205/MODULE_88833531cb6c582d4830+a02c3a36/model.hlo_module.pb +1 -1
  5. neuronxcc-2.21.33363.0+82129205/MODULE_88833531cb6c582d4830+a02c3a36/model.neff +1 -1
  6. 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.dev4/qwen3/Qwen/Qwen3-0.6B/1c88c352d1e2acf2c0db.json ADDED
@@ -0,0 +1,88 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "Qwen/Qwen3-0.6B",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Qwen3ForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "dtype": "bfloat16",
11
+ "head_dim": 128,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 1024,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 3072,
16
+ "layer_types": [
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
+ ],
46
+ "max_position_embeddings": 40960,
47
+ "max_window_layers": 28,
48
+ "model_type": "qwen3",
49
+ "neuron": {
50
+ "_serialized_key": "NxDNeuronConfig",
51
+ "batch_size": 4,
52
+ "capacity_factor": null,
53
+ "checkpoint_id": "Qwen/Qwen3-0.6B",
54
+ "checkpoint_revision": "c1899de289a04d12100db370d81485cdf75e47ca",
55
+ "continuous_batching": true,
56
+ "ep_degree": 1,
57
+ "fused_qkv": true,
58
+ "glu_mlp": true,
59
+ "local_ranks_size": 1,
60
+ "max_batch_size": 4,
61
+ "max_context_length": 1024,
62
+ "max_topk": 256,
63
+ "n_active_tokens": 1024,
64
+ "neuronxcc_version": "2.21.33363.0+82129205",
65
+ "on_device_sampling": true,
66
+ "optimum_neuron_version": "0.4.6.dev4",
67
+ "output_logits": false,
68
+ "pp_degree": 1,
69
+ "prefill_chunk_size": 0,
70
+ "sequence_length": 1024,
71
+ "speculation_length": 0,
72
+ "start_rank_id": 0,
73
+ "target": "trn1",
74
+ "torch_dtype": "bfloat16",
75
+ "tp_degree": 1
76
+ },
77
+ "num_attention_heads": 16,
78
+ "num_hidden_layers": 28,
79
+ "num_key_value_heads": 8,
80
+ "rms_norm_eps": 1e-06,
81
+ "rope_scaling": null,
82
+ "rope_theta": 1000000,
83
+ "sliding_window": null,
84
+ "tie_word_embeddings": true,
85
+ "use_cache": true,
86
+ "use_sliding_window": false,
87
+ "vocab_size": 151936
88
+ }
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:c4c451ee5e3d3817baf2e50dd1de77c54b7586e76eff272fba45c965ec641462
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 sha256:11a252aa1c29150cd778db1e50c46950f1d7d0cc3cd2ad07d1799ff53fb6c2a1
3
  size 7732224
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a1b06a9acdf1690c293052a89209e977a553b6773528b661895c595ae18ccd3d
3
  size 7732224
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 sha256:18adbe50e9216cd7336e7f765110680cd947b35bb2712491666057767936edf9
3
  size 3472384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:de34c20badc27e31138c278612cebcbeaea2e4e3473638ef00db2295004ab83e
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 sha256:5699fab1e345884f2b200655f838d49fb7e0a98cbbc7700392d3daa5d3ee3205
3
  size 3610166
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d85f9318e09d9314d0c26b0eb8a222aa72d35b3755f8eea7e0b8d85eef774320
3
  size 3610166