optimum-internal-testing-user commited on
Commit
a39c82e
·
verified ·
1 Parent(s): a635a90

Synchronizing local compiler cache.

Browse files
neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev0/qwen2/Qwen/Qwen2.5-0.5B/843dbfd7ba684a222d28.json ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "hidden_act": "silu",
10
+ "hidden_size": 896,
11
+ "initializer_range": 0.02,
12
+ "intermediate_size": 4864,
13
+ "layer_types": [
14
+ "full_attention",
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
+ ],
39
+ "max_position_embeddings": 32768,
40
+ "max_window_layers": 24,
41
+ "model_type": "qwen2",
42
+ "neuron": {
43
+ "_serialized_key": "NxDNeuronConfig",
44
+ "batch_size": 4,
45
+ "capacity_factor": null,
46
+ "checkpoint_id": "Qwen/Qwen2.5-0.5B",
47
+ "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987",
48
+ "continuous_batching": true,
49
+ "dtype": "bfloat16",
50
+ "ep_degree": 1,
51
+ "fused_qkv": false,
52
+ "glu_mlp": true,
53
+ "local_ranks_size": 2,
54
+ "max_batch_size": 4,
55
+ "max_context_length": 4096,
56
+ "max_topk": 256,
57
+ "n_active_tokens": 4096,
58
+ "neuronxcc_version": "2.21.18209.0+043b1bf7",
59
+ "on_device_sampling": false,
60
+ "optimum_neuron_version": "0.4.2.dev0",
61
+ "output_logits": false,
62
+ "pp_degree": 1,
63
+ "sequence_length": 4096,
64
+ "speculation_length": 0,
65
+ "start_rank_id": 0,
66
+ "target": "trn1",
67
+ "tp_degree": 2
68
+ },
69
+ "num_attention_heads": 14,
70
+ "num_hidden_layers": 24,
71
+ "num_key_value_heads": 2,
72
+ "rms_norm_eps": 1e-06,
73
+ "rope_scaling": null,
74
+ "rope_theta": 1000000.0,
75
+ "sliding_window": null,
76
+ "tie_word_embeddings": true,
77
+ "use_cache": true,
78
+ "use_mrope": false,
79
+ "use_sliding_window": false,
80
+ "vocab_size": 151936
81
+ }
neuronxcc-2.21.18209.0+043b1bf7/MODULE_9d35248a93d4142a3cf1+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:89d95332373996418dd36c71e92d055f25151179af769bdcbf34f30aa3bcacb6
3
  size 739558
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a9b4cf2ae15a0658e20dd364719045c89e2c0417ef74d2b393941ef24ef20d75
3
  size 739558
neuronxcc-2.21.18209.0+043b1bf7/MODULE_9d35248a93d4142a3cf1+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:dee47034ab783eac59be02bccb1aa5ab2b746cf1bd350578fff367704fec7c2c
3
  size 26133504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:67159e9bc3c51de0c853eaa5308dfe06fd1f674343ac521d56d239ff78101408
3
  size 26133504
neuronxcc-2.21.18209.0+043b1bf7/MODULE_c2b3afa51c57d431a332+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:78410670de86fb85b626bbcd51a5cb4825e7176c757a61fdf64a03dafb9915fd
3
  size 1926144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:af4dcbc6c6d0e9fe14ff22b539e7d0e3dfbdf398b95dbdd156d5cc4b1693108d
3
  size 1926144
neuronxcc-2.21.18209.0+043b1bf7/MODULE_c2b3afa51c57d431a332+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3146e765c0a8e68312e1844f7d08a458fe6e0e46ad72afb51d3cade89de7d9e9
3
  size 2082478
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ac368621f57bc5e821bd0b980f8d0438420cee53c29ee4039492c2fd6b4829a1
3
  size 2082478