Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
 Community
optimum-internal-testing-user commited on
Commit
874816b
·
verified ·
1 Parent(s): 39830d9

Synchronizing local compiler cache.

Browse files
Files changed (6) hide show
  1. neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7.dev0/qwen2/Qwen/Qwen2.5-0.5B/0c27e9e199d54ea0efcd.json +85 -0
  2. neuronxcc-2.26.6360.0+6f180f47/MODULE_51bdae28051bc87badd5+84337dd9/model.hlo_module.pb +1 -1
  3. neuronxcc-2.26.6360.0+6f180f47/MODULE_51bdae28051bc87badd5+84337dd9/model.neff +1 -1
  4. neuronxcc-2.26.6360.0+6f180f47/MODULE_51bdae28051bc87badd5+84337dd9/wrapped_neff.hlo +1 -1
  5. neuronxcc-2.26.6360.0+6f180f47/MODULE_f9dd548904a4ef181037+12dd9119/model.hlo_module.pb +1 -1
  6. neuronxcc-2.26.6360.0+6f180f47/MODULE_f9dd548904a4ef181037+12dd9119/model.neff +1 -1
neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7.dev0/qwen2/Qwen/Qwen2.5-0.5B/0c27e9e199d54ea0efcd.json ADDED
@@ -0,0 +1,85 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "Qwen/Qwen2.5-0.5B",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Qwen2ForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "dtype": "bfloat16",
10
+ "hidden_act": "silu",
11
+ "hidden_size": 896,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 4864,
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
+ ],
40
+ "max_position_embeddings": 32768,
41
+ "max_window_layers": 24,
42
+ "model_type": "qwen2",
43
+ "neuron": {
44
+ "_serialized_key": "NxDNeuronConfig",
45
+ "batch_size": 1,
46
+ "capacity_factor": null,
47
+ "checkpoint_id": "Qwen/Qwen2.5-0.5B",
48
+ "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987",
49
+ "continuous_batching": false,
50
+ "ep_degree": 1,
51
+ "fused_qkv": false,
52
+ "glu_mlp": true,
53
+ "local_ranks_size": 2,
54
+ "max_batch_size": 1,
55
+ "max_context_length": 8192,
56
+ "max_topk": 256,
57
+ "n_active_tokens": 8192,
58
+ "neuronxcc_version": "2.26.6360.0+6f180f47",
59
+ "on_device_sampling": true,
60
+ "optimum_neuron_version": "0.4.7.dev0",
61
+ "output_logits": false,
62
+ "pp_degree": 1,
63
+ "prefill_chunk_size": 1024,
64
+ "sequence_length": 8192,
65
+ "speculation_length": 0,
66
+ "start_rank_id": 0,
67
+ "target": "trn1",
68
+ "torch_dtype": "bfloat16",
69
+ "tp_degree": 2
70
+ },
71
+ "num_attention_heads": 14,
72
+ "num_hidden_layers": 24,
73
+ "num_key_value_heads": 2,
74
+ "rms_norm_eps": 1e-06,
75
+ "rope_parameters": {
76
+ "rope_theta": 1000000.0,
77
+ "rope_type": "default"
78
+ },
79
+ "sliding_window": null,
80
+ "tie_word_embeddings": true,
81
+ "use_cache": true,
82
+ "use_mrope": false,
83
+ "use_sliding_window": false,
84
+ "vocab_size": 151936
85
+ }
neuronxcc-2.26.6360.0+6f180f47/MODULE_51bdae28051bc87badd5+84337dd9/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d0186113c22e9d6fad8f144034063d6a80721da85aa5b4a384c99996af9cba49
3
  size 632134
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9c23890e1c39e36f6681bebabfc4e56bbe4eec23d5ca496ce2cdfb7b9f42cf4b
3
  size 632134
neuronxcc-2.26.6360.0+6f180f47/MODULE_51bdae28051bc87badd5+84337dd9/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4c4f84889038f80840947255fbfa94ae68599091c39d39113a9b11b1990b9dea
3
  size 1588224
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:78793d19e7b659480220507f11e2535f877b7acba1242b9e267e09dc934aa64d
3
  size 1588224
neuronxcc-2.26.6360.0+6f180f47/MODULE_51bdae28051bc87badd5+84337dd9/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:259742ca93d4c9eb3cab92c0bcabd1ec103cf6c8cbac09e090ae01e603ac2134
3
  size 1744530
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:14cbd4d41d3a589d9b574361d14de016a9ec78a364ac7126db8e7013dd26f30f
3
  size 1744530
neuronxcc-2.26.6360.0+6f180f47/MODULE_f9dd548904a4ef181037+12dd9119/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:88986767c66dbe283b3b6c664d0d5f14f66598ab0a23e7cb1258d5e5aedb92c8
3
  size 827306
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2b4842cf3fd345e5425b57ed6d9874cc0ff04f9ed7edf535fe6f19a3a2d9a4bc
3
  size 827306
neuronxcc-2.26.6360.0+6f180f47/MODULE_f9dd548904a4ef181037+12dd9119/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8476a457f8e32d7a0a1c8ba2a54e7894746273d170b40030ab0c172ffec829e5
3
  size 14593024
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6bdc6f098fc0e418aa2091c96655d7df4c2b91ffd1d17b1a563d6d41e95b871c
3
  size 14593024