Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
Community
optimum-internal-testing-user commited on
Commit
8829ac1
·
verified ·
1 Parent(s): 657afd3

Synchronizing local compiler cache.

Browse files
Files changed (41) hide show
  1. neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/granite/hf-internal-testing/tiny-random-GraniteForCausalLM/c7cee7420399240ed7cb.json +58 -0
  2. neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/llama/llamafactory/tiny-random-Llama-3/590d09e219b67db1ce85.json +62 -0
  3. neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/llama/unsloth/Llama-3.2-1B-Instruct/3c9cadb9b044a3b82ed4.json +63 -0
  4. neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/llama4_text/tiny-random/llama-4/a883ca3cda054b4cc268.json +81 -0
  5. neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/mixtral/dacorvo/Mixtral-tiny/b6ee8412b5ba78767bd4.json +58 -0
  6. neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/phi3/yujiepan/phi-4-tiny-random/4da10399ec36ed9644a8.json +59 -0
  7. neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/qwen2/yujiepan/qwen2.5-128k-tiny-random/06d241212f96b3260dd4.json +64 -0
  8. neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/qwen3_moe/optimum-internal-testing/tiny-random-qwen3_moe/c429c7346a7794899b4e.json +65 -0
  9. neuronxcc-2.21.18209.0+043b1bf7/MODULE_0cdc152aff6a7cbf36f6+a02c3a36/model.neff +1 -1
  10. neuronxcc-2.21.18209.0+043b1bf7/MODULE_0cdc152aff6a7cbf36f6+a02c3a36/wrapped_neff.hlo +1 -1
  11. neuronxcc-2.21.18209.0+043b1bf7/MODULE_3e60ddce17a65ac4ee68+24129607/model.hlo_module.pb +1 -1
  12. neuronxcc-2.21.18209.0+043b1bf7/MODULE_3e60ddce17a65ac4ee68+24129607/model.neff +1 -1
  13. neuronxcc-2.21.18209.0+043b1bf7/MODULE_410ef55a3aca5cd80fc8+24129607/model.hlo_module.pb +1 -1
  14. neuronxcc-2.21.18209.0+043b1bf7/MODULE_410ef55a3aca5cd80fc8+24129607/model.neff +1 -1
  15. neuronxcc-2.21.18209.0+043b1bf7/MODULE_62abb2454c6da8d57ad5+a02c3a36/model.neff +1 -1
  16. neuronxcc-2.21.18209.0+043b1bf7/MODULE_62abb2454c6da8d57ad5+a02c3a36/wrapped_neff.hlo +1 -1
  17. neuronxcc-2.21.18209.0+043b1bf7/MODULE_73587bc7122ba5a81758+80d05c3f/model.neff +1 -1
  18. neuronxcc-2.21.18209.0+043b1bf7/MODULE_73587bc7122ba5a81758+80d05c3f/wrapped_neff.hlo +1 -1
  19. neuronxcc-2.21.18209.0+043b1bf7/MODULE_75329916c4da1120954c+6e4949b4/model.hlo_module.pb +1 -1
  20. neuronxcc-2.21.18209.0+043b1bf7/MODULE_75329916c4da1120954c+6e4949b4/model.neff +1 -1
  21. neuronxcc-2.21.18209.0+043b1bf7/MODULE_7c590ab98f45f8a8dd15+283df001/model.hlo_module.pb +1 -1
  22. neuronxcc-2.21.18209.0+043b1bf7/MODULE_7c590ab98f45f8a8dd15+283df001/model.neff +1 -1
  23. neuronxcc-2.21.18209.0+043b1bf7/MODULE_944763b36d810f4af582+24129607/model.hlo_module.pb +1 -1
  24. neuronxcc-2.21.18209.0+043b1bf7/MODULE_944763b36d810f4af582+24129607/model.neff +1 -1
  25. neuronxcc-2.21.18209.0+043b1bf7/MODULE_b23271513f9095bde4db+2dde74c7/model.neff +1 -1
  26. neuronxcc-2.21.18209.0+043b1bf7/MODULE_b23271513f9095bde4db+2dde74c7/wrapped_neff.hlo +1 -1
  27. neuronxcc-2.21.18209.0+043b1bf7/MODULE_b3cfe28275e1a02a015c+24129607/model.hlo_module.pb +1 -1
  28. neuronxcc-2.21.18209.0+043b1bf7/MODULE_b3cfe28275e1a02a015c+24129607/model.neff +1 -1
  29. neuronxcc-2.21.18209.0+043b1bf7/MODULE_bdf25cde1a48d420319c+a02c3a36/model.neff +1 -1
  30. neuronxcc-2.21.18209.0+043b1bf7/MODULE_bdf25cde1a48d420319c+a02c3a36/wrapped_neff.hlo +1 -1
  31. neuronxcc-2.21.18209.0+043b1bf7/MODULE_c4737893663f5ff66b53+24129607/model.hlo_module.pb +1 -1
  32. neuronxcc-2.21.18209.0+043b1bf7/MODULE_c4737893663f5ff66b53+24129607/model.neff +1 -1
  33. neuronxcc-2.21.18209.0+043b1bf7/MODULE_cde3417e6918fabff4a7+a02c3a36/model.neff +1 -1
  34. neuronxcc-2.21.18209.0+043b1bf7/MODULE_cde3417e6918fabff4a7+a02c3a36/wrapped_neff.hlo +1 -1
  35. neuronxcc-2.21.18209.0+043b1bf7/MODULE_d4df6648fd68fe444e68+a02c3a36/model.hlo_module.pb +1 -1
  36. neuronxcc-2.21.18209.0+043b1bf7/MODULE_d4df6648fd68fe444e68+a02c3a36/model.neff +1 -1
  37. neuronxcc-2.21.18209.0+043b1bf7/MODULE_d4df6648fd68fe444e68+a02c3a36/wrapped_neff.hlo +1 -1
  38. neuronxcc-2.21.18209.0+043b1bf7/MODULE_f52fe8e88daf6dc9619a+24129607/model.hlo_module.pb +1 -1
  39. neuronxcc-2.21.18209.0+043b1bf7/MODULE_f52fe8e88daf6dc9619a+24129607/model.neff +1 -1
  40. neuronxcc-2.21.18209.0+043b1bf7/MODULE_fef9240b8fbfaca895ba+a02c3a36/model.neff +1 -1
  41. neuronxcc-2.21.18209.0+043b1bf7/MODULE_fef9240b8fbfaca895ba+a02c3a36/wrapped_neff.hlo +1 -1
neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/granite/hf-internal-testing/tiny-random-GraniteForCausalLM/c7cee7420399240ed7cb.json ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "hf-internal-testing/tiny-random-GraniteForCausalLM",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "GraniteForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "attention_multiplier": 1.0,
11
+ "dtype": "float32",
12
+ "embedding_multiplier": 1.0,
13
+ "hidden_act": "silu",
14
+ "hidden_size": 32,
15
+ "initializer_range": 0.02,
16
+ "intermediate_size": 64,
17
+ "logits_scaling": 1.0,
18
+ "max_position_embeddings": 2048,
19
+ "mlp_bias": false,
20
+ "model_type": "granite",
21
+ "neuron": {
22
+ "_serialized_key": "NxDNeuronConfig",
23
+ "batch_size": 1,
24
+ "capacity_factor": null,
25
+ "checkpoint_id": "hf-internal-testing/tiny-random-GraniteForCausalLM",
26
+ "checkpoint_revision": "c3074ebc0ac2fe545305f5e5f6cce2cc9b2aa0c5",
27
+ "continuous_batching": false,
28
+ "ep_degree": 1,
29
+ "fused_qkv": true,
30
+ "glu_mlp": true,
31
+ "local_ranks_size": 2,
32
+ "max_batch_size": 1,
33
+ "max_context_length": 1024,
34
+ "max_topk": 256,
35
+ "n_active_tokens": 1024,
36
+ "neuronxcc_version": "2.21.18209.0+043b1bf7",
37
+ "on_device_sampling": true,
38
+ "optimum_neuron_version": "0.4.2.dev1",
39
+ "output_logits": false,
40
+ "pp_degree": 1,
41
+ "sequence_length": 1024,
42
+ "speculation_length": 0,
43
+ "start_rank_id": 0,
44
+ "target": "trn1",
45
+ "torch_dtype": "float32",
46
+ "tp_degree": 2
47
+ },
48
+ "num_attention_heads": 4,
49
+ "num_hidden_layers": 2,
50
+ "num_key_value_heads": 4,
51
+ "residual_multiplier": 1.0,
52
+ "rms_norm_eps": 1e-06,
53
+ "rope_scaling": null,
54
+ "rope_theta": 10000.0,
55
+ "tie_word_embeddings": false,
56
+ "use_cache": true,
57
+ "vocab_size": 49152
58
+ }
neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/llama/llamafactory/tiny-random-Llama-3/590d09e219b67db1ce85.json ADDED
@@ -0,0 +1,62 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "llamafactory/tiny-random-Llama-3",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "LlamaForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "dtype": "float16",
11
+ "head_dim": 4,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 16,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 64,
16
+ "max_position_embeddings": 131072,
17
+ "mlp_bias": false,
18
+ "model_type": "llama",
19
+ "neuron": {
20
+ "_serialized_key": "NxDNeuronConfig",
21
+ "batch_size": 1,
22
+ "capacity_factor": null,
23
+ "checkpoint_id": "llamafactory/tiny-random-Llama-3",
24
+ "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8",
25
+ "continuous_batching": false,
26
+ "ep_degree": 1,
27
+ "fused_qkv": true,
28
+ "glu_mlp": true,
29
+ "local_ranks_size": 2,
30
+ "max_batch_size": 1,
31
+ "max_context_length": 1024,
32
+ "max_topk": 256,
33
+ "n_active_tokens": 1024,
34
+ "neuronxcc_version": "2.21.18209.0+043b1bf7",
35
+ "on_device_sampling": true,
36
+ "optimum_neuron_version": "0.4.2.dev1",
37
+ "output_logits": false,
38
+ "pp_degree": 1,
39
+ "sequence_length": 1024,
40
+ "speculation_length": 0,
41
+ "start_rank_id": 0,
42
+ "target": "trn1",
43
+ "torch_dtype": "float16",
44
+ "tp_degree": 2
45
+ },
46
+ "num_attention_heads": 4,
47
+ "num_hidden_layers": 2,
48
+ "num_key_value_heads": 4,
49
+ "pretraining_tp": 1,
50
+ "rms_norm_eps": 1e-05,
51
+ "rope_scaling": {
52
+ "factor": 8.0,
53
+ "high_freq_factor": 4.0,
54
+ "low_freq_factor": 1.0,
55
+ "original_max_position_embeddings": 8192,
56
+ "rope_type": "llama3"
57
+ },
58
+ "rope_theta": 500000.0,
59
+ "tie_word_embeddings": false,
60
+ "use_cache": true,
61
+ "vocab_size": 128256
62
+ }
neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/llama/unsloth/Llama-3.2-1B-Instruct/3c9cadb9b044a3b82ed4.json ADDED
@@ -0,0 +1,63 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "unsloth/Llama-3.2-1B-Instruct",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "LlamaForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "dtype": "bfloat16",
11
+ "head_dim": 64,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 2048,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 8192,
16
+ "max_position_embeddings": 131072,
17
+ "mlp_bias": false,
18
+ "model_type": "llama",
19
+ "neuron": {
20
+ "_serialized_key": "NxDNeuronConfig",
21
+ "batch_size": 1,
22
+ "capacity_factor": null,
23
+ "checkpoint_id": "unsloth/Llama-3.2-1B-Instruct",
24
+ "checkpoint_revision": null,
25
+ "continuous_batching": false,
26
+ "ep_degree": 1,
27
+ "fused_qkv": false,
28
+ "glu_mlp": true,
29
+ "local_ranks_size": 2,
30
+ "max_batch_size": 1,
31
+ "max_context_length": 4096,
32
+ "max_topk": 256,
33
+ "n_active_tokens": 4096,
34
+ "neuronxcc_version": "2.21.18209.0+043b1bf7",
35
+ "on_device_sampling": false,
36
+ "optimum_neuron_version": "0.4.2.dev1",
37
+ "output_logits": false,
38
+ "pp_degree": 1,
39
+ "sequence_length": 4096,
40
+ "speculation_length": 5,
41
+ "start_rank_id": 0,
42
+ "target": "trn1",
43
+ "torch_dtype": "bfloat16",
44
+ "tp_degree": 2
45
+ },
46
+ "num_attention_heads": 32,
47
+ "num_hidden_layers": 16,
48
+ "num_key_value_heads": 8,
49
+ "pretraining_tp": 1,
50
+ "rms_norm_eps": 1e-05,
51
+ "rope_scaling": {
52
+ "factor": 32.0,
53
+ "high_freq_factor": 4.0,
54
+ "low_freq_factor": 1.0,
55
+ "original_max_position_embeddings": 8192,
56
+ "rope_type": "llama3"
57
+ },
58
+ "rope_theta": 500000.0,
59
+ "tie_word_embeddings": true,
60
+ "unsloth_fixed": true,
61
+ "use_cache": true,
62
+ "vocab_size": 128256
63
+ }
neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/llama4_text/tiny-random/llama-4/a883ca3cda054b4cc268.json ADDED
@@ -0,0 +1,81 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "tiny-random/llama-4",
4
+ "_task": "text-generation",
5
+ "attention_bias": false,
6
+ "attention_chunk_size": 128,
7
+ "attention_dropout": 0.0,
8
+ "attn_scale": 0.1,
9
+ "attn_temperature_tuning": 4,
10
+ "cache_implementation": "hybrid",
11
+ "dtype": "bfloat16",
12
+ "floor_scale": 8192,
13
+ "for_llm_compressor": false,
14
+ "head_dim": 32,
15
+ "hidden_act": "silu",
16
+ "hidden_size": 32,
17
+ "initializer_range": 0.02,
18
+ "interleave_moe_layer_step": 2,
19
+ "intermediate_size": 64,
20
+ "intermediate_size_mlp": 128,
21
+ "layer_types": [
22
+ "chunked_attention",
23
+ "chunked_attention",
24
+ "chunked_attention",
25
+ "full_attention"
26
+ ],
27
+ "max_position_embeddings": 1048576,
28
+ "model_type": "llama4_text",
29
+ "moe_layers": [
30
+ 1,
31
+ 3
32
+ ],
33
+ "neuron": {
34
+ "_serialized_key": "NxDNeuronConfig",
35
+ "batch_size": 1,
36
+ "capacity_factor": null,
37
+ "checkpoint_id": "tiny-random/llama-4",
38
+ "checkpoint_revision": "9e716f5d4d1ffe0a44a15f46f4a12b840439aba4",
39
+ "continuous_batching": false,
40
+ "ep_degree": 1,
41
+ "fused_qkv": false,
42
+ "glu_mlp": true,
43
+ "local_ranks_size": 2,
44
+ "max_batch_size": 1,
45
+ "max_context_length": 1024,
46
+ "max_topk": 256,
47
+ "n_active_tokens": 1024,
48
+ "neuronxcc_version": "2.21.18209.0+043b1bf7",
49
+ "on_device_sampling": true,
50
+ "optimum_neuron_version": "0.4.2.dev1",
51
+ "output_logits": false,
52
+ "pp_degree": 1,
53
+ "sequence_length": 1024,
54
+ "speculation_length": 0,
55
+ "start_rank_id": 0,
56
+ "target": "trn1",
57
+ "torch_dtype": "bfloat16",
58
+ "tp_degree": 2
59
+ },
60
+ "no_rope_layers": [
61
+ 1,
62
+ 1,
63
+ 1,
64
+ 0
65
+ ],
66
+ "num_attention_heads": 1,
67
+ "num_experts_per_tok": 1,
68
+ "num_hidden_layers": 4,
69
+ "num_key_value_heads": 1,
70
+ "num_local_experts": 8,
71
+ "output_router_logits": false,
72
+ "rms_norm_eps": 1e-05,
73
+ "rope_scaling": null,
74
+ "rope_theta": 500000.0,
75
+ "router_aux_loss_coef": 0.001,
76
+ "router_jitter_noise": 0.0,
77
+ "tie_word_embeddings": true,
78
+ "use_cache": true,
79
+ "use_qk_norm": true,
80
+ "vocab_size": 202048
81
+ }
neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/mixtral/dacorvo/Mixtral-tiny/b6ee8412b5ba78767bd4.json ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "dacorvo/Mixtral-tiny",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "MixtralForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "dtype": "float16",
10
+ "head_dim": 32,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 1024,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 3584,
15
+ "max_position_embeddings": 1024,
16
+ "model_type": "mixtral",
17
+ "neuron": {
18
+ "_serialized_key": "NxDNeuronConfig",
19
+ "batch_size": 1,
20
+ "capacity_factor": null,
21
+ "checkpoint_id": "dacorvo/Mixtral-tiny",
22
+ "checkpoint_revision": "c557ba205ddff6ea911f4719e0d543d6c08356b6",
23
+ "continuous_batching": false,
24
+ "ep_degree": 1,
25
+ "fused_qkv": false,
26
+ "glu_mlp": true,
27
+ "local_ranks_size": 2,
28
+ "max_batch_size": 1,
29
+ "max_context_length": 1024,
30
+ "max_topk": 256,
31
+ "n_active_tokens": 1024,
32
+ "neuronxcc_version": "2.21.18209.0+043b1bf7",
33
+ "on_device_sampling": false,
34
+ "optimum_neuron_version": "0.4.2.dev1",
35
+ "output_logits": false,
36
+ "pp_degree": 1,
37
+ "sequence_length": 1024,
38
+ "speculation_length": 0,
39
+ "start_rank_id": 0,
40
+ "target": "trn1",
41
+ "torch_dtype": "float16",
42
+ "tp_degree": 2
43
+ },
44
+ "num_attention_heads": 32,
45
+ "num_experts_per_tok": 2,
46
+ "num_hidden_layers": 2,
47
+ "num_key_value_heads": 8,
48
+ "num_local_experts": 8,
49
+ "output_router_logits": false,
50
+ "rms_norm_eps": 1e-05,
51
+ "rope_theta": 10000.0,
52
+ "router_aux_loss_coef": 0.001,
53
+ "router_jitter_noise": 0.0,
54
+ "sliding_window": 4096,
55
+ "tie_word_embeddings": false,
56
+ "use_cache": true,
57
+ "vocab_size": 32000
58
+ }
neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/phi3/yujiepan/phi-4-tiny-random/4da10399ec36ed9644a8.json ADDED
@@ -0,0 +1,59 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "yujiepan/phi-4-tiny-random",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Phi3ForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "auto_map": {},
11
+ "dtype": "bfloat16",
12
+ "embd_pdrop": 0.0,
13
+ "hidden_act": "silu",
14
+ "hidden_size": 16,
15
+ "initializer_range": 0.02,
16
+ "intermediate_size": 32,
17
+ "max_position_embeddings": 16384,
18
+ "model_type": "phi3",
19
+ "neuron": {
20
+ "_serialized_key": "NxDNeuronConfig",
21
+ "batch_size": 1,
22
+ "capacity_factor": null,
23
+ "checkpoint_id": "yujiepan/phi-4-tiny-random",
24
+ "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a",
25
+ "continuous_batching": false,
26
+ "ep_degree": 1,
27
+ "fused_qkv": true,
28
+ "glu_mlp": true,
29
+ "local_ranks_size": 2,
30
+ "max_batch_size": 1,
31
+ "max_context_length": 1024,
32
+ "max_topk": 256,
33
+ "n_active_tokens": 1024,
34
+ "neuronxcc_version": "2.21.18209.0+043b1bf7",
35
+ "on_device_sampling": true,
36
+ "optimum_neuron_version": "0.4.2.dev1",
37
+ "output_logits": false,
38
+ "pp_degree": 1,
39
+ "sequence_length": 1024,
40
+ "speculation_length": 0,
41
+ "start_rank_id": 0,
42
+ "target": "trn1",
43
+ "torch_dtype": "bfloat16",
44
+ "tp_degree": 2
45
+ },
46
+ "num_attention_heads": 2,
47
+ "num_hidden_layers": 2,
48
+ "num_key_value_heads": 1,
49
+ "original_max_position_embeddings": 16384,
50
+ "partial_rotary_factor": 1.0,
51
+ "resid_pdrop": 0.0,
52
+ "rms_norm_eps": 1e-05,
53
+ "rope_scaling": null,
54
+ "rope_theta": 250000,
55
+ "sliding_window": null,
56
+ "tie_word_embeddings": false,
57
+ "use_cache": true,
58
+ "vocab_size": 100352
59
+ }
neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/qwen2/yujiepan/qwen2.5-128k-tiny-random/06d241212f96b3260dd4.json ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "yujiepan/qwen2.5-128k-tiny-random",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Qwen2ForCausalLM"
7
+ ],
8
+ "attention_dropout": 0.0,
9
+ "dtype": "bfloat16",
10
+ "hidden_act": "silu",
11
+ "hidden_size": 8,
12
+ "initializer_range": 0.02,
13
+ "intermediate_size": 16,
14
+ "layer_types": [
15
+ "full_attention",
16
+ "full_attention"
17
+ ],
18
+ "max_position_embeddings": 32768,
19
+ "max_window_layers": 1,
20
+ "model_type": "qwen2",
21
+ "neuron": {
22
+ "_serialized_key": "NxDNeuronConfig",
23
+ "batch_size": 1,
24
+ "capacity_factor": null,
25
+ "checkpoint_id": "yujiepan/qwen2.5-128k-tiny-random",
26
+ "checkpoint_revision": "c8296d4ca3f87782876d2382fbb6481d1beb8ef0",
27
+ "continuous_batching": false,
28
+ "ep_degree": 1,
29
+ "fused_qkv": false,
30
+ "glu_mlp": true,
31
+ "local_ranks_size": 2,
32
+ "max_batch_size": 1,
33
+ "max_context_length": 1024,
34
+ "max_topk": 256,
35
+ "n_active_tokens": 1024,
36
+ "neuronxcc_version": "2.21.18209.0+043b1bf7",
37
+ "on_device_sampling": true,
38
+ "optimum_neuron_version": "0.4.2.dev1",
39
+ "output_logits": false,
40
+ "pp_degree": 1,
41
+ "sequence_length": 1024,
42
+ "speculation_length": 0,
43
+ "start_rank_id": 0,
44
+ "target": "trn1",
45
+ "torch_dtype": "bfloat16",
46
+ "tp_degree": 2
47
+ },
48
+ "num_attention_heads": 4,
49
+ "num_hidden_layers": 2,
50
+ "num_key_value_heads": 2,
51
+ "rms_norm_eps": 1e-06,
52
+ "rope_scaling": {
53
+ "factor": 4.0,
54
+ "original_max_position_embeddings": 32768,
55
+ "rope_type": "yarn",
56
+ "type": "yarn"
57
+ },
58
+ "rope_theta": 1000000.0,
59
+ "sliding_window": null,
60
+ "tie_word_embeddings": false,
61
+ "use_cache": true,
62
+ "use_sliding_window": false,
63
+ "vocab_size": 152064
64
+ }
neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev1/qwen3_moe/optimum-internal-testing/tiny-random-qwen3_moe/c429c7346a7794899b4e.json ADDED
@@ -0,0 +1,65 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "optimum-internal-testing/tiny-random-qwen3_moe",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Qwen3MoeForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "decoder_sparse_step": 2,
11
+ "dtype": "float32",
12
+ "head_dim": 32,
13
+ "hidden_act": "silu",
14
+ "hidden_size": 64,
15
+ "initializer_range": 0.02,
16
+ "intermediate_size": 128,
17
+ "max_position_embeddings": 40960,
18
+ "max_window_layers": 1,
19
+ "mlp_only_layers": [],
20
+ "model_type": "qwen3_moe",
21
+ "moe_intermediate_size": 128,
22
+ "neuron": {
23
+ "_serialized_key": "NxDNeuronConfig",
24
+ "batch_size": 1,
25
+ "capacity_factor": null,
26
+ "checkpoint_id": "optimum-internal-testing/tiny-random-qwen3_moe",
27
+ "checkpoint_revision": "e0230be2839556b44b7400a233c73c74b4abb7af",
28
+ "continuous_batching": false,
29
+ "ep_degree": 1,
30
+ "fused_qkv": false,
31
+ "glu_mlp": true,
32
+ "local_ranks_size": 2,
33
+ "max_batch_size": 1,
34
+ "max_context_length": 1024,
35
+ "max_topk": 256,
36
+ "n_active_tokens": 1024,
37
+ "neuronxcc_version": "2.21.18209.0+043b1bf7",
38
+ "on_device_sampling": true,
39
+ "optimum_neuron_version": "0.4.2.dev1",
40
+ "output_logits": false,
41
+ "pp_degree": 1,
42
+ "sequence_length": 1024,
43
+ "speculation_length": 0,
44
+ "start_rank_id": 0,
45
+ "target": "trn1",
46
+ "torch_dtype": "float32",
47
+ "tp_degree": 2
48
+ },
49
+ "norm_topk_prob": true,
50
+ "num_attention_heads": 2,
51
+ "num_experts": 8,
52
+ "num_experts_per_tok": 2,
53
+ "num_hidden_layers": 2,
54
+ "num_key_value_heads": 1,
55
+ "output_router_logits": false,
56
+ "rms_norm_eps": 1e-06,
57
+ "rope_scaling": null,
58
+ "rope_theta": 1000000.0,
59
+ "router_aux_loss_coef": 0.001,
60
+ "sliding_window": null,
61
+ "tie_word_embeddings": true,
62
+ "use_cache": true,
63
+ "use_sliding_window": false,
64
+ "vocab_size": 151936
65
+ }
neuronxcc-2.21.18209.0+043b1bf7/MODULE_0cdc152aff6a7cbf36f6+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f124ea2d6597bf655f6fb7446768e218c01ebab966fc81379240036eda12bcbc
3
  size 277504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d30da794f77ba9e95125bacb1fd6fdb8e5a76f89a9ca736b29d165eade109eff
3
  size 277504
neuronxcc-2.21.18209.0+043b1bf7/MODULE_0cdc152aff6a7cbf36f6+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:374bb2537191af773cb7ca91a417f431638a6d7caddabd2d2615e8cad441728e
3
  size 289031
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e96654ad374c613af0b526bcd47b4d76bdc9899db2dd0f82384987b61514617e
3
  size 289031
neuronxcc-2.21.18209.0+043b1bf7/MODULE_3e60ddce17a65ac4ee68+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:aaf68a7d1f22a4d042c18355e15d17dda246df63cd9c6096c304ef5df5caae98
3
  size 82772
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:daa56f4ac44a9f5ef0c2bea00b070484c64f869b2f270f64ad0216a1b10a8a45
3
  size 82772
neuronxcc-2.21.18209.0+043b1bf7/MODULE_3e60ddce17a65ac4ee68+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:00df735b20d5b7f6fa72847fe6a047f65a354ae6bcab40552c948bfce425d15f
3
  size 267264
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8793d01ad310aa958419f492fdbf979c1ee4903ad3b51e378c2c8bf166025e1b
3
  size 267264
neuronxcc-2.21.18209.0+043b1bf7/MODULE_410ef55a3aca5cd80fc8+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9c971171ef784d5e3c310619622af4468efbf24641e81d99f59e84eb89f8f425
3
  size 81516
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0c3a49dd72fdb5d2a07cca8e30e0036959275a194da4ffc5cad20247ce22ac7d
3
  size 81516
neuronxcc-2.21.18209.0+043b1bf7/MODULE_410ef55a3aca5cd80fc8+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:08725a1377c63337d4a9d035036da7a0af327454873892075c3723b7b6d23590
3
  size 267264
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a6ced9d0e1a179e5d7a1d5492c54f3bf9932d149a9c636ebd3d4df281f41320b
3
  size 267264
neuronxcc-2.21.18209.0+043b1bf7/MODULE_62abb2454c6da8d57ad5+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a9a5e6b624cb279e4143f78f2d786e182f7b7d04335e33d9a3435f1f5664835b
3
  size 543744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ed638326e44dcfbc04afae254b316f1b49686eee58d1bcc20375569d8e111789
3
  size 543744
neuronxcc-2.21.18209.0+043b1bf7/MODULE_62abb2454c6da8d57ad5+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:292dc6689e1ee54833a185ff676e6dc7a57aed8c2af983d1dafdbb9a290c6c6f
3
  size 563380
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:822f20169e3132b8cfd09a5825449557afa3ef22b5a31fca16d9eba5d1f43c21
3
  size 563380
neuronxcc-2.21.18209.0+043b1bf7/MODULE_73587bc7122ba5a81758+80d05c3f/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:285434825fb69508ada73558f8ece260eb88fe7d77b1306b722590582a63ce19
3
  size 277504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1b05ecb14e7ee94ab397282e9a4424a795152ca6f6c5e9a39a528c92de7b7a39
3
  size 277504
neuronxcc-2.21.18209.0+043b1bf7/MODULE_73587bc7122ba5a81758+80d05c3f/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:aca5938acfdd3cc620ef4cda3119f55484caa55642bd1424e58473aeadc46e82
3
  size 289571
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d53eda7ac30f732b2726ca6042091b1b56df903b7fbbdc0d3d2cdbee86fef5d6
3
  size 289571
neuronxcc-2.21.18209.0+043b1bf7/MODULE_75329916c4da1120954c+6e4949b4/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:523bf908d44337b2e698c2ef6650bd21ab8f8a637ee86c68021985ffd904a38c
3
  size 84807
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6126e0b2e546e6ea64244384b3f3fc8553a6e38d382d288bc90c252603b8574a
3
  size 84807
neuronxcc-2.21.18209.0+043b1bf7/MODULE_75329916c4da1120954c+6e4949b4/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:90380916d4c3092a55c5d60df5d5633bb19768c38bb0365a37f5aa123e44d7be
3
  size 646144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f8a6d65c8fa171c3a335a30e1481b8afc5002be4ae1ed9d5c667c4a28f290f71
3
  size 646144
neuronxcc-2.21.18209.0+043b1bf7/MODULE_7c590ab98f45f8a8dd15+283df001/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:de110bdb5133cc01f89f207d9983dca72a91e0e7b908cd7431102ac4cd244a30
3
  size 97794
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:054c5055b65405714e51c64f753a419433ec87fb494bf2753f8927c2cda60cdc
3
  size 97794
neuronxcc-2.21.18209.0+043b1bf7/MODULE_7c590ab98f45f8a8dd15+283df001/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7bf4d0fc17fb679bfd503fffd031937feb0e0fcc5e7e9ba299553710ea959920
3
  size 410624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b686d0dda024c1a1c026fbe02982d23ce6054166dfd1721389598bc71e3d0ead
3
  size 410624
neuronxcc-2.21.18209.0+043b1bf7/MODULE_944763b36d810f4af582+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e45e267eba9429aaf1e543c5d123b3a4eb8d92bf34c940b1cf60bdf35b86979d
3
  size 83504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ba30bac1303e4d846c40c65f8e4d8a644794b6964038372ff73e155714762bdc
3
  size 83504
neuronxcc-2.21.18209.0+043b1bf7/MODULE_944763b36d810f4af582+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6e6d95f711cf07ae2ffe64489acb453c1417037381a9701e60a53d42c64eb06a
3
  size 328704
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:34131fa4eac476f701cc997fa1d814a098348140e269823119bb04d3739331c1
3
  size 328704
neuronxcc-2.21.18209.0+043b1bf7/MODULE_b23271513f9095bde4db+2dde74c7/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6462b7a92c24f2ced3a3389b12a06fdc46f192fcb35116eba3e74a0a6bbc4956
3
  size 369664
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9cdaf94187b39213cb662c8a0fc4b0d357a9a0c78d907af73d7d131f727408bb
3
  size 369664
neuronxcc-2.21.18209.0+043b1bf7/MODULE_b23271513f9095bde4db+2dde74c7/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7a769f2d05a085fc2b475d67384ce9da70922fc35af0de537582cf36565444fa
3
  size 379362
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:803736b931481737b965f71e7af419bc8c99724c6830791f71062a587c6eca8f
3
  size 379362
neuronxcc-2.21.18209.0+043b1bf7/MODULE_b3cfe28275e1a02a015c+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7d98bb4abcef46bc4dfcb539ece861af724c555279e78c4ba6ef27c16f25cecc
3
  size 739558
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:649ff83e0cbee13abda84e02c1a114e8b60d780456161e42f68c99c57744691c
3
  size 739558
neuronxcc-2.21.18209.0+043b1bf7/MODULE_b3cfe28275e1a02a015c+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:eb447dfb9b9c358a7db742dd7704be431966fd924b365845869e8939a2abd571
3
  size 26133504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:732c12b8c4fb3d0c32afec54fb17a9f3c93592cd1a6790abf69c4d765137bbe1
3
  size 26133504
neuronxcc-2.21.18209.0+043b1bf7/MODULE_bdf25cde1a48d420319c+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d95604f17aa9930977e79c554033f90df9a4964fb65038ef69ebffab0131fd99
3
  size 246784
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b8c94b67bfac622c1ba2e3674d69b4a7807b95593fae1d583e31f7752bdc8c14
3
  size 246784
neuronxcc-2.21.18209.0+043b1bf7/MODULE_bdf25cde1a48d420319c+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:08f6168b8b47679cc5078be53555afefa23abfe58c9bf0660994a7754d823045
3
  size 255104
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8302079bfc0b0e738353e44dbc22fff76780094c04219d1912745b787f0db837
3
  size 255104
neuronxcc-2.21.18209.0+043b1bf7/MODULE_c4737893663f5ff66b53+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:147792162da53dacec876ecd4752bc3390dfb4b29155d5ae61081ac93fbfb045
3
  size 90382
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:e68a1baa2fe3d4b52585e883dcdadfdf184aa6279668ea4e5c8ebe6f8a4a42d2
3
  size 90382
neuronxcc-2.21.18209.0+043b1bf7/MODULE_c4737893663f5ff66b53+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:46c97acff0dac80a69ccf045ded2ed748f9cb353bd108a80e0664832c0f4e5de
3
  size 359424
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:77422caa4ac266c76feea26ed4446ed080a35691518e58d79b8e9c842346703c
3
  size 359424
neuronxcc-2.21.18209.0+043b1bf7/MODULE_cde3417e6918fabff4a7+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7a64dad05247beb56f26df49c7e8ad59c787a244fcb76a5f72db30b6b3d7e9a2
3
  size 277504
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:51d75148836e75cc300541f47a4f2c644fe52908390c7c45a8a561f0676c4847
3
  size 277504
neuronxcc-2.21.18209.0+043b1bf7/MODULE_cde3417e6918fabff4a7+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b561430da4b81c91cc16c87193c32869573bd2d128e4f505e89a2d11384d17d0
3
  size 285854
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:06a711369153370466b976d6a02b796eace22a4c61525753b0081a6b6e6009a3
3
  size 285854
neuronxcc-2.21.18209.0+043b1bf7/MODULE_d4df6648fd68fe444e68+a02c3a36/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:4b570926ce0fcfd782ef61d818285f77f7df61321cfef5c4f1085b51b81d2a1a
3
  size 588406
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2c2920e5fde77a80d2d66c9ba5addf46e53d032f533c7097cba34e31243f84ed
3
  size 588406
neuronxcc-2.21.18209.0+043b1bf7/MODULE_d4df6648fd68fe444e68+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:87e67a770b83d195d58763fe262a8184b4a620ca0f88603cdab2094f008c895d
3
  size 1926144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89a337760756ccdce6db50969700ad4d617c6a37811665ea5c63a8315f366f5d
3
  size 1926144
neuronxcc-2.21.18209.0+043b1bf7/MODULE_d4df6648fd68fe444e68+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c8a8fcff7d6a50c1c86d5df788ea7f8988d1078854dcf3d8b163b5c0203e7fa8
3
  size 2082478
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:287829616709f5e4c53d80b02ed1f385b3f6a0c9c90544f00297d8ff70849ffb
3
  size 2082478
neuronxcc-2.21.18209.0+043b1bf7/MODULE_f52fe8e88daf6dc9619a+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0937859dbae30fe8eb955de7385ac31d4bcebd8a482e63a3f8f587d10426cee1
3
  size 694128
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6fe1533083be685e04908eaafd5dd3e002ff344b3624f4b42176f7d99ff3d634
3
  size 694128
neuronxcc-2.21.18209.0+043b1bf7/MODULE_f52fe8e88daf6dc9619a+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0e0d39fe6c9d23ec2f100aa35bac1cc0afd8a4339a35483f1ea654f3b4fe7e87
3
  size 625664
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4a2636f9f699ebc67a9889925446a795508f82ceebfb1d87cac120fa8ca19765
3
  size 625664
neuronxcc-2.21.18209.0+043b1bf7/MODULE_fef9240b8fbfaca895ba+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:045c207d93d958e506a0c376bfdc149a261dde13610ad749b570dda7b3e4782f
3
  size 216064
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:f01f38f8c4463abe16a1dc652195f36513435e9d0f29861858f2b0803866c913
3
  size 216064
neuronxcc-2.21.18209.0+043b1bf7/MODULE_fef9240b8fbfaca895ba+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:46b4643f0ea2f95c0c99dc5fcba2a4832828e0e0340880b19af5a19bfdf54e8f
3
  size 224412
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:524a1cedcc8a3cfac1dc2a49ff589a00003c931d55aaf9219ca5da0007595eb9
3
  size 224412