Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
Community
optimum-internal-testing-user commited on
Commit
a4250ac
·
verified ·
1 Parent(s): d9c70f8

Synchronizing local compiler cache.

Browse files
Files changed (9) hide show
  1. neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.1.dev0/qwen2/Qwen/Qwen2.5-0.5B/a35f43930569fd7bd14c.json +82 -0
  2. neuronxcc-2.21.18209.0+043b1bf7/MODULE_4be33b117e68e2de3326+a9d440f5/model.neff +1 -1
  3. neuronxcc-2.21.18209.0+043b1bf7/MODULE_4be33b117e68e2de3326+a9d440f5/wrapped_neff.hlo +1 -1
  4. neuronxcc-2.21.18209.0+043b1bf7/MODULE_50d30ce495feb34fe5c1+a9d440f5/model.neff +1 -1
  5. neuronxcc-2.21.18209.0+043b1bf7/MODULE_50d30ce495feb34fe5c1+a9d440f5/wrapped_neff.hlo +1 -1
  6. neuronxcc-2.21.18209.0+043b1bf7/MODULE_73b50e8e61409e1f8fe1+ed72d204/model.hlo_module.pb +1 -1
  7. neuronxcc-2.21.18209.0+043b1bf7/MODULE_73b50e8e61409e1f8fe1+ed72d204/model.neff +1 -1
  8. neuronxcc-2.21.18209.0+043b1bf7/MODULE_cae14ab78f54c8b4e73f+ed72d204/model.hlo_module.pb +1 -1
  9. neuronxcc-2.21.18209.0+043b1bf7/MODULE_cae14ab78f54c8b4e73f+ed72d204/model.neff +1 -1
neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.1.dev0/qwen2/Qwen/Qwen2.5-0.5B/a35f43930569fd7bd14c.json ADDED
@@ -0,0 +1,82 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
+ "enable_bucketing": false,
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": "0.4.1.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
+ "torch_dtype": "float16",
68
+ "tp_degree": 2
69
+ },
70
+ "num_attention_heads": 14,
71
+ "num_hidden_layers": 24,
72
+ "num_key_value_heads": 2,
73
+ "rms_norm_eps": 1e-06,
74
+ "rope_scaling": null,
75
+ "rope_theta": 1000000.0,
76
+ "sliding_window": null,
77
+ "tie_word_embeddings": true,
78
+ "use_cache": true,
79
+ "use_mrope": false,
80
+ "use_sliding_window": false,
81
+ "vocab_size": 151936
82
+ }
neuronxcc-2.21.18209.0+043b1bf7/MODULE_4be33b117e68e2de3326+a9d440f5/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0cd0dd1029ad343d311182a5e633595c05978671b33b56acf65e65368824d43b
3
  size 3093504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d70d650de28d7b13defbb5f9d78b06f99ec63ba26cda803845b298cbca3d0581
3
  size 3093504
neuronxcc-2.21.18209.0+043b1bf7/MODULE_4be33b117e68e2de3326+a9d440f5/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:35ad902abd90ec49120c5aafd4769fc1a23a9c186b65dfbd97c48053401daf0d
3
  size 3167748
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8d0d1b1e40a092a3eec560354aeb1191196e8b39c2312249a4d6f358c7ef2a51
3
  size 3167748
neuronxcc-2.21.18209.0+043b1bf7/MODULE_50d30ce495feb34fe5c1+a9d440f5/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f80595b7e847fd3a15e9ed6861cf1f3ca3b0f167aaea438bd1a75cfe5c87a0ba
3
  size 1926144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:af93b3d02b40084bdec001a9496f6b98115e3bfebad2f126075400a4c60fbd87
3
  size 1926144
neuronxcc-2.21.18209.0+043b1bf7/MODULE_50d30ce495feb34fe5c1+a9d440f5/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2109e78743e0b1641ef88efc71caf3d2fe73f52a2b13422e3d5ee38f1485bba1
3
  size 2082478
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6951f83519dce668f7723834939eacc71fdb113475631340f061d1db9cd51852
3
  size 2082478
neuronxcc-2.21.18209.0+043b1bf7/MODULE_73b50e8e61409e1f8fe1+ed72d204/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:637e1374e15bf184dd2bb4e3d96b85bd9be0e22c5fd09c32e2c8de0cafcf7e80
3
  size 509252
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:441fac3e96e30c894231d7f6dd20d4e89e6b4cf1295ed681a96fb69af4ea7582
3
  size 509252
neuronxcc-2.21.18209.0+043b1bf7/MODULE_73b50e8e61409e1f8fe1+ed72d204/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ecc8cd2933d03bd72923e70db0219d9f5eb1266f43f6f91fc9ca077ef6802850
3
  size 41585664
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ccfc71e4eaf13a28e93ac88d5f63f2d222b919a9db2e9f4a966422a8f787f591
3
  size 41585664
neuronxcc-2.21.18209.0+043b1bf7/MODULE_cae14ab78f54c8b4e73f+ed72d204/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d9ce15e9ef444f76e6084ba9f4cf55300119d7da731a84d4e90d0e7019b30aa7
3
  size 739366
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b8511b22bc58244c0086580ec7d72532553a8f93a374268bdb1fbb737acfc3a4
3
  size 739366
neuronxcc-2.21.18209.0+043b1bf7/MODULE_cae14ab78f54c8b4e73f+ed72d204/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e8fe600bff94def682d6aec75bbf39a4644962af78d254d493b67c900df7fe4b
3
  size 26133504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3ffb3d7bc9903f8346667ae3bc452aa306ed217fe3fc73cd25365d476c98eac8
3
  size 26133504