Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
 Community
optimum-internal-testing-user commited on
Commit
2e6ee2f
·
verified ·
1 Parent(s): b0a39ea

Synchronizing local compiler cache.

Browse files
Files changed (12) hide show
  1. neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/929b02754a13cbfdf657d863c3fc6f3bce672879bc6ae48ab45be21e881e9ec2/e4a2dbc4e1473e9dd12b.json +87 -0
  2. neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/qwen3/Qwen/Qwen3-0.6B/e4a2dbc4e1473e9dd12b.json +87 -0
  3. neuronxcc-2.21.33363.0+82129205/MODULE_6a21c8e3eeb9125f1a9c+a02c3a36/model.hlo_module.pb +1 -1
  4. neuronxcc-2.21.33363.0+82129205/MODULE_6a21c8e3eeb9125f1a9c+a02c3a36/model.neff +1 -1
  5. neuronxcc-2.21.33363.0+82129205/MODULE_6a21c8e3eeb9125f1a9c+a02c3a36/wrapped_neff.hlo +1 -1
  6. neuronxcc-2.21.33363.0+82129205/MODULE_958d41f716f615f7fb01+24129607/model.hlo_module.pb +1 -1
  7. neuronxcc-2.21.33363.0+82129205/MODULE_958d41f716f615f7fb01+24129607/model.neff +1 -1
  8. neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/model.hlo_module.pb +1 -1
  9. neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/model.neff +1 -1
  10. neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/wrapped_neff.hlo +1 -1
  11. neuronxcc-2.21.33363.0+82129205/MODULE_e45e2c512692918b8ac7+24129607/model.hlo_module.pb +1 -1
  12. neuronxcc-2.21.33363.0+82129205/MODULE_e45e2c512692918b8ac7+24129607/model.neff +1 -1
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/929b02754a13cbfdf657d863c3fc6f3bce672879bc6ae48ab45be21e881e9ec2/e4a2dbc4e1473e9dd12b.json ADDED
@@ -0,0 +1,87 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 2,
60
+ "max_batch_size": 4,
61
+ "max_context_length": 4096,
62
+ "max_topk": 256,
63
+ "n_active_tokens": 4096,
64
+ "neuronxcc_version": "2.21.33363.0+82129205",
65
+ "on_device_sampling": false,
66
+ "optimum_neuron_version": "0.4.6.dev0",
67
+ "output_logits": false,
68
+ "pp_degree": 1,
69
+ "sequence_length": 4096,
70
+ "speculation_length": 0,
71
+ "start_rank_id": 0,
72
+ "target": "trn1",
73
+ "torch_dtype": "bfloat16",
74
+ "tp_degree": 2
75
+ },
76
+ "num_attention_heads": 16,
77
+ "num_hidden_layers": 28,
78
+ "num_key_value_heads": 8,
79
+ "rms_norm_eps": 1e-06,
80
+ "rope_scaling": null,
81
+ "rope_theta": 1000000,
82
+ "sliding_window": null,
83
+ "tie_word_embeddings": true,
84
+ "use_cache": true,
85
+ "use_sliding_window": false,
86
+ "vocab_size": 151936
87
+ }
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/qwen3/Qwen/Qwen3-0.6B/e4a2dbc4e1473e9dd12b.json ADDED
@@ -0,0 +1,87 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 2,
60
+ "max_batch_size": 4,
61
+ "max_context_length": 4096,
62
+ "max_topk": 256,
63
+ "n_active_tokens": 4096,
64
+ "neuronxcc_version": "2.21.33363.0+82129205",
65
+ "on_device_sampling": false,
66
+ "optimum_neuron_version": "0.4.6.dev0",
67
+ "output_logits": false,
68
+ "pp_degree": 1,
69
+ "sequence_length": 4096,
70
+ "speculation_length": 0,
71
+ "start_rank_id": 0,
72
+ "target": "trn1",
73
+ "torch_dtype": "bfloat16",
74
+ "tp_degree": 2
75
+ },
76
+ "num_attention_heads": 16,
77
+ "num_hidden_layers": 28,
78
+ "num_key_value_heads": 8,
79
+ "rms_norm_eps": 1e-06,
80
+ "rope_scaling": null,
81
+ "rope_theta": 1000000,
82
+ "sliding_window": null,
83
+ "tie_word_embeddings": true,
84
+ "use_cache": true,
85
+ "use_sliding_window": false,
86
+ "vocab_size": 151936
87
+ }
neuronxcc-2.21.33363.0+82129205/MODULE_6a21c8e3eeb9125f1a9c+a02c3a36/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e4fe1fe7fbe05c357ec6ce7787ba7569de3551fd1b854171acd8c6526152830e
3
  size 569536
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:526e8478dff68672c6091200f03d94fa784d760b6af9495a5b5bc52bf238d9bb
3
  size 569536
neuronxcc-2.21.33363.0+82129205/MODULE_6a21c8e3eeb9125f1a9c+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:fd3b6c11b9fedd9b55e5f3ea8c0819381e224337b1da6ba12066ed97de9925ef
3
  size 1373184
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1cdc4e858377a179e32b2566e433818bdf91ddda5ddfe5151dc0e78101c9066b
3
  size 1373184
neuronxcc-2.21.33363.0+82129205/MODULE_6a21c8e3eeb9125f1a9c+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:26ad5befcb8454d72ca0270d00a14a2e838f1830613f5a504b9b4be66ea75b98
3
  size 1495573
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2ee85311c9e3891aa28d60adf06ff5fbc80d7036133daffe4922823d320286cb
3
  size 1495573
neuronxcc-2.21.33363.0+82129205/MODULE_958d41f716f615f7fb01+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:87f55c3f481969e9042baa2996a436c447062c57a41a9bf5f6043febbde67363
3
  size 532500
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d87aedd12dd5989e8d84f53a3aecb58df8c31e5adc70a3682b0c33ffc3799e7e
3
  size 532500
neuronxcc-2.21.33363.0+82129205/MODULE_958d41f716f615f7fb01+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0b294c0f43a710998b20991ce8dbe6dc73073d74f0a0232e83fcad31f51b959a
3
  size 29942784
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fa81b8f7bcb32c5911e3bd7fe5883d54ba8e6da6cf13e02c30b845d91f9d37de
3
  size 29942784
neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:66f6450490b10388dafb6e4fb250f49f37b11224dd2e5c49f481a0976f729106
3
  size 562167
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cedde36e9447943a3efdf4ed22ba8e8377c18dc164c05dc9ec3b2404fd285d6c
3
  size 562167
neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ac673781bbf3e2dbc475872bf420164d20d0771203247969452879b64ad49b1f
3
  size 1793024
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:26b19d16e87edce0eb519472854464ea428c8c6621ce680463c7a0576186d64b
3
  size 1793024
neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:794b24bed8dc827a4275a8e1acbadd1ec8b9d25d8643b83854e9dfaa4b94f296
3
  size 1915413
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4364f3e30295be20e1b36a9f8b3877323fbdd44fe0154ef2aa2e7e559a991578
3
  size 1915413
neuronxcc-2.21.33363.0+82129205/MODULE_e45e2c512692918b8ac7+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:709aef187eaab30b99967d754304fe6bf08161b2e4820c8d294053c66f68c8ae
3
  size 571822
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:09002100cbee99f3cf9d355464c3b373418336fa7f26fce3379b075a8ff4dd8b
3
  size 571822
neuronxcc-2.21.33363.0+82129205/MODULE_e45e2c512692918b8ac7+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a0946eccda0ccaa11fd83c9e6b1926be6d653f863251c91dca40ee872de00538
3
  size 9503744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b94048b1e583bea84cb9e311cfe1f6298e201e617dedcd340fbc12ea64ab735d
3
  size 9503744