Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
 Community
optimum-internal-testing-user commited on
Commit
09782b6
·
verified ·
1 Parent(s): 25b9205

Synchronizing local compiler cache.

Browse files
Files changed (12) hide show
  1. neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7/4cb7aff9e2a15c151396f2b684013e39d6739f0dec83e5c9dabbfe9d5fcf77b7/e7b113210101250c9dc0.json +83 -0
  2. neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7/qwen2/Qwen/Qwen2.5-0.5B/e7b113210101250c9dc0.json +83 -0
  3. neuronxcc-2.26.6360.0+6f180f47/MODULE_31f800285afef122700a+519f203d/model.hlo_module.pb +1 -1
  4. neuronxcc-2.26.6360.0+6f180f47/MODULE_31f800285afef122700a+519f203d/model.neff +1 -1
  5. neuronxcc-2.26.6360.0+6f180f47/MODULE_5459e3e60c7ca34be52b+84337dd9/model.neff +1 -1
  6. neuronxcc-2.26.6360.0+6f180f47/MODULE_5459e3e60c7ca34be52b+84337dd9/wrapped_neff.hlo +1 -1
  7. neuronxcc-2.26.6360.0+6f180f47/MODULE_813c4591c1c397fa42bc+84337dd9/model.neff +1 -1
  8. neuronxcc-2.26.6360.0+6f180f47/MODULE_813c4591c1c397fa42bc+84337dd9/wrapped_neff.hlo +1 -1
  9. neuronxcc-2.26.6360.0+6f180f47/MODULE_ca940b5929c96ea4ebad+12dd9119/model.neff +1 -1
  10. neuronxcc-2.26.6360.0+6f180f47/MODULE_cef50dd7a142db2bfa48+84337dd9/model.neff +1 -1
  11. neuronxcc-2.26.6360.0+6f180f47/MODULE_cef50dd7a142db2bfa48+84337dd9/wrapped_neff.hlo +1 -1
  12. neuronxcc-2.26.6360.0+6f180f47/MODULE_dc04527b827075028b3a+519f203d/model.neff +1 -1
neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7/4cb7aff9e2a15c151396f2b684013e39d6739f0dec83e5c9dabbfe9d5fcf77b7/e7b113210101250c9dc0.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 4,
46
+ "capacity_factor": null,
47
+ "checkpoint_id": "Qwen/Qwen2.5-0.5B",
48
+ "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987",
49
+ "continuous_batching": true,
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": 1024,
56
+ "max_topk": 256,
57
+ "n_active_tokens": 1024,
58
+ "neuronxcc_version": "2.26.6360.0+6f180f47",
59
+ "on_device_sampling": false,
60
+ "optimum_neuron_version": "0.4.7",
61
+ "output_logits": false,
62
+ "pp_degree": 1,
63
+ "prefill_chunk_size": 0,
64
+ "sequence_length": 1024,
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_scaling": null,
76
+ "rope_theta": 1000000.0,
77
+ "sliding_window": null,
78
+ "tie_word_embeddings": true,
79
+ "use_cache": true,
80
+ "use_mrope": false,
81
+ "use_sliding_window": false,
82
+ "vocab_size": 151936
83
+ }
neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7/qwen2/Qwen/Qwen2.5-0.5B/e7b113210101250c9dc0.json ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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": 4,
46
+ "capacity_factor": null,
47
+ "checkpoint_id": "Qwen/Qwen2.5-0.5B",
48
+ "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987",
49
+ "continuous_batching": true,
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": 1024,
56
+ "max_topk": 256,
57
+ "n_active_tokens": 1024,
58
+ "neuronxcc_version": "2.26.6360.0+6f180f47",
59
+ "on_device_sampling": false,
60
+ "optimum_neuron_version": "0.4.7",
61
+ "output_logits": false,
62
+ "pp_degree": 1,
63
+ "prefill_chunk_size": 0,
64
+ "sequence_length": 1024,
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_scaling": null,
76
+ "rope_theta": 1000000.0,
77
+ "sliding_window": null,
78
+ "tie_word_embeddings": true,
79
+ "use_cache": true,
80
+ "use_mrope": false,
81
+ "use_sliding_window": false,
82
+ "vocab_size": 151936
83
+ }
neuronxcc-2.26.6360.0+6f180f47/MODULE_31f800285afef122700a+519f203d/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c8454d98412c6125cf2c85a924935363d03cabaae0a58f1998fa24081e8a2f81
3
  size 573928
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d42b644b3eeef1a5595cccdbd7bbd9d9305f77c48a3b5aa7fd889a75fa7c4536
3
  size 573928
neuronxcc-2.26.6360.0+6f180f47/MODULE_31f800285afef122700a+519f203d/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:455bad02115c6303cd64814373764c6cd81b2233065334546a1b4631a05333a6
3
  size 3216384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30dbdddc510010c02caa7e4d8005fb3125d7fe04338c7cffcf81d15e1226e042
3
  size 3216384
neuronxcc-2.26.6360.0+6f180f47/MODULE_5459e3e60c7ca34be52b+84337dd9/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c5443c91b284bc69daaf0f24598e9bbdbfb9a054b4ba75c15ab7e1b29533aa48
3
  size 1547264
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fc7f1f16e738148da0ad40ffd6e8e9ce4a3eb8303549968ddaaaa7e4ee2921d4
3
  size 1547264
neuronxcc-2.26.6360.0+6f180f47/MODULE_5459e3e60c7ca34be52b+84337dd9/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:33a77979c908bac7b0ee49d1e2613d8f674ac15f2f0c3f72739fa70cd2a432bb
3
  size 1703455
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:475a88e21367427ad9b691c04ad2913424083143959ce2642d852c03eb0dfa31
3
  size 1703455
neuronxcc-2.26.6360.0+6f180f47/MODULE_813c4591c1c397fa42bc+84337dd9/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:eb5d7da597bf1a2ed25b08b6edb93c6dc52246a7327f823a939add4d28ddd32c
3
  size 2161664
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1b5b0ad23250795560eb46b65f81ea4d40e7605ec3757ac36042abfe20f718a3
3
  size 2161664
neuronxcc-2.26.6360.0+6f180f47/MODULE_813c4591c1c397fa42bc+84337dd9/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f2a4a1417e67cf31483eac836b72045c82bac36dda994e71e8c53d15e7cf2968
3
  size 2235751
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:048569c61bdc1d0d139e2840859e928d6c0e6ebe17793cf719079ef2cd1ab930
3
  size 2235751
neuronxcc-2.26.6360.0+6f180f47/MODULE_ca940b5929c96ea4ebad+12dd9119/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:37ede2094f37e6d40e959ae49f099ee7c9473baeae5ffe10ce03155103efd67f
3
  size 10261504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6550756bc53826f9f3918843f7f8b7e1b19e0bdbe3d8c145500060c44d7287da
3
  size 10261504
neuronxcc-2.26.6360.0+6f180f47/MODULE_cef50dd7a142db2bfa48+84337dd9/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8f5b3616b07fdaa4792a3fcc3030f5c7adae04435842132949bd507d66a5bc59
3
  size 2110464
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:60fc849ebed8542f68c2bc0fb3b4e7af21d22b7ff9e0af22d4c86977c2cd6be5
3
  size 2110464
neuronxcc-2.26.6360.0+6f180f47/MODULE_cef50dd7a142db2bfa48+84337dd9/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:18dba82a9f15fd64d2bd6681cf7ddb5a7b94f8136c51663a5dfa44044f4a5f49
3
  size 2184663
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c775d6f64d7854903d53d94a043ca2ae523192e7d45118b956935d48ce058197
3
  size 2184663
neuronxcc-2.26.6360.0+6f180f47/MODULE_dc04527b827075028b3a+519f203d/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:138e001d5cc1d491942248e60dc74e632b7ba75c27991db48046728a539e2e43
3
  size 3062784
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:48b34f34b9e356e9450669c356d13f8c58dd2e9a1a5eb634b28367c2c4624b96
3
  size 3062784