Synchronizing local compiler cache.
Browse files
neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7.dev0/e7d4e795523dab7995bc9d349c1acdbf98531773cbd0e38e0113d7443f505bb4/351d893710aee97c9726.json
ADDED
|
@@ -0,0 +1,140 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_entry_class": "SingleModelCacheEntry",
|
| 3 |
+
"_model_id": "unsloth/gemma-3-4b-it",
|
| 4 |
+
"_task": "image-text-to-text",
|
| 5 |
+
"architectures": [
|
| 6 |
+
"Gemma3ForConditionalGeneration"
|
| 7 |
+
],
|
| 8 |
+
"boi_token_index": 255999,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eoi_token_index": 256000,
|
| 11 |
+
"image_token_index": 262144,
|
| 12 |
+
"initializer_range": 0.02,
|
| 13 |
+
"mm_tokens_per_image": 256,
|
| 14 |
+
"model_type": "gemma3",
|
| 15 |
+
"neuron": {
|
| 16 |
+
"_serialized_key": "NxDVLMNeuronConfig",
|
| 17 |
+
"batch_size": 2,
|
| 18 |
+
"capacity_factor": null,
|
| 19 |
+
"checkpoint_id": "unsloth/gemma-3-4b-it",
|
| 20 |
+
"checkpoint_revision": "bf46152c47f5dd20b896357cb51abc4c03b8ee8c",
|
| 21 |
+
"continuous_batching": true,
|
| 22 |
+
"ep_degree": 1,
|
| 23 |
+
"fused_qkv": true,
|
| 24 |
+
"glu_mlp": true,
|
| 25 |
+
"image_seq_len": 256,
|
| 26 |
+
"image_size": 896,
|
| 27 |
+
"local_ranks_size": 2,
|
| 28 |
+
"max_batch_size": 2,
|
| 29 |
+
"max_context_length": 2048,
|
| 30 |
+
"max_num_images": 5,
|
| 31 |
+
"max_topk": 256,
|
| 32 |
+
"n_active_tokens": 2048,
|
| 33 |
+
"neuronxcc_version": "2.26.6360.0+6f180f47",
|
| 34 |
+
"on_device_sampling": true,
|
| 35 |
+
"optimum_neuron_version": "0.4.7.dev0",
|
| 36 |
+
"output_logits": false,
|
| 37 |
+
"pp_degree": 1,
|
| 38 |
+
"prefill_chunk_size": 0,
|
| 39 |
+
"sequence_length": 2048,
|
| 40 |
+
"speculation_length": 0,
|
| 41 |
+
"start_rank_id": 0,
|
| 42 |
+
"target": "trn1",
|
| 43 |
+
"torch_dtype": "bfloat16",
|
| 44 |
+
"tp_degree": 2
|
| 45 |
+
},
|
| 46 |
+
"text_config": {
|
| 47 |
+
"_sliding_window_pattern": 6,
|
| 48 |
+
"attention_bias": false,
|
| 49 |
+
"attention_dropout": 0.0,
|
| 50 |
+
"attn_logit_softcapping": null,
|
| 51 |
+
"bos_token_id": 2,
|
| 52 |
+
"cache_implementation": "hybrid",
|
| 53 |
+
"dtype": "bfloat16",
|
| 54 |
+
"eos_token_id": 1,
|
| 55 |
+
"final_logit_softcapping": null,
|
| 56 |
+
"head_dim": 256,
|
| 57 |
+
"hidden_activation": "gelu_pytorch_tanh",
|
| 58 |
+
"hidden_size": 2560,
|
| 59 |
+
"initializer_range": 0.02,
|
| 60 |
+
"intermediate_size": 10240,
|
| 61 |
+
"layer_types": [
|
| 62 |
+
"sliding_attention",
|
| 63 |
+
"sliding_attention",
|
| 64 |
+
"sliding_attention",
|
| 65 |
+
"sliding_attention",
|
| 66 |
+
"sliding_attention",
|
| 67 |
+
"full_attention",
|
| 68 |
+
"sliding_attention",
|
| 69 |
+
"sliding_attention",
|
| 70 |
+
"sliding_attention",
|
| 71 |
+
"sliding_attention",
|
| 72 |
+
"sliding_attention",
|
| 73 |
+
"full_attention",
|
| 74 |
+
"sliding_attention",
|
| 75 |
+
"sliding_attention",
|
| 76 |
+
"sliding_attention",
|
| 77 |
+
"sliding_attention",
|
| 78 |
+
"sliding_attention",
|
| 79 |
+
"full_attention",
|
| 80 |
+
"sliding_attention",
|
| 81 |
+
"sliding_attention",
|
| 82 |
+
"sliding_attention",
|
| 83 |
+
"sliding_attention",
|
| 84 |
+
"sliding_attention",
|
| 85 |
+
"full_attention",
|
| 86 |
+
"sliding_attention",
|
| 87 |
+
"sliding_attention",
|
| 88 |
+
"sliding_attention",
|
| 89 |
+
"sliding_attention",
|
| 90 |
+
"sliding_attention",
|
| 91 |
+
"full_attention",
|
| 92 |
+
"sliding_attention",
|
| 93 |
+
"sliding_attention",
|
| 94 |
+
"sliding_attention",
|
| 95 |
+
"sliding_attention"
|
| 96 |
+
],
|
| 97 |
+
"max_position_embeddings": 131072,
|
| 98 |
+
"model_type": "gemma3_text",
|
| 99 |
+
"num_attention_heads": 8,
|
| 100 |
+
"num_hidden_layers": 34,
|
| 101 |
+
"num_key_value_heads": 4,
|
| 102 |
+
"pad_token_id": 0,
|
| 103 |
+
"query_pre_attn_scalar": 256,
|
| 104 |
+
"rms_norm_eps": 1e-06,
|
| 105 |
+
"rope_parameters": {
|
| 106 |
+
"full_attention": {
|
| 107 |
+
"factor": 8.0,
|
| 108 |
+
"rope_theta": 1000000.0,
|
| 109 |
+
"rope_type": "linear"
|
| 110 |
+
},
|
| 111 |
+
"sliding_attention": {
|
| 112 |
+
"rope_theta": 10000.0,
|
| 113 |
+
"rope_type": "default"
|
| 114 |
+
}
|
| 115 |
+
},
|
| 116 |
+
"sliding_window": 1024,
|
| 117 |
+
"sliding_window_pattern": 6,
|
| 118 |
+
"tie_word_embeddings": true,
|
| 119 |
+
"use_bidirectional_attention": false,
|
| 120 |
+
"use_cache": true,
|
| 121 |
+
"vocab_size": 262208
|
| 122 |
+
},
|
| 123 |
+
"tie_word_embeddings": true,
|
| 124 |
+
"unsloth_fixed": true,
|
| 125 |
+
"vision_config": {
|
| 126 |
+
"attention_dropout": 0.0,
|
| 127 |
+
"dtype": "bfloat16",
|
| 128 |
+
"hidden_act": "gelu_pytorch_tanh",
|
| 129 |
+
"hidden_size": 1152,
|
| 130 |
+
"image_size": 896,
|
| 131 |
+
"intermediate_size": 4304,
|
| 132 |
+
"layer_norm_eps": 1e-06,
|
| 133 |
+
"model_type": "siglip_vision_model",
|
| 134 |
+
"num_attention_heads": 16,
|
| 135 |
+
"num_channels": 3,
|
| 136 |
+
"num_hidden_layers": 27,
|
| 137 |
+
"patch_size": 14,
|
| 138 |
+
"vision_use_head": false
|
| 139 |
+
}
|
| 140 |
+
}
|
neuronxcc-2.26.6360.0+6f180f47/0_REGISTRY/0.4.7.dev0/gemma3/unsloth/gemma-3-4b-it/351d893710aee97c9726.json
ADDED
|
@@ -0,0 +1,140 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_entry_class": "SingleModelCacheEntry",
|
| 3 |
+
"_model_id": "unsloth/gemma-3-4b-it",
|
| 4 |
+
"_task": "image-text-to-text",
|
| 5 |
+
"architectures": [
|
| 6 |
+
"Gemma3ForConditionalGeneration"
|
| 7 |
+
],
|
| 8 |
+
"boi_token_index": 255999,
|
| 9 |
+
"dtype": "bfloat16",
|
| 10 |
+
"eoi_token_index": 256000,
|
| 11 |
+
"image_token_index": 262144,
|
| 12 |
+
"initializer_range": 0.02,
|
| 13 |
+
"mm_tokens_per_image": 256,
|
| 14 |
+
"model_type": "gemma3",
|
| 15 |
+
"neuron": {
|
| 16 |
+
"_serialized_key": "NxDVLMNeuronConfig",
|
| 17 |
+
"batch_size": 2,
|
| 18 |
+
"capacity_factor": null,
|
| 19 |
+
"checkpoint_id": "unsloth/gemma-3-4b-it",
|
| 20 |
+
"checkpoint_revision": "bf46152c47f5dd20b896357cb51abc4c03b8ee8c",
|
| 21 |
+
"continuous_batching": true,
|
| 22 |
+
"ep_degree": 1,
|
| 23 |
+
"fused_qkv": true,
|
| 24 |
+
"glu_mlp": true,
|
| 25 |
+
"image_seq_len": 256,
|
| 26 |
+
"image_size": 896,
|
| 27 |
+
"local_ranks_size": 2,
|
| 28 |
+
"max_batch_size": 2,
|
| 29 |
+
"max_context_length": 2048,
|
| 30 |
+
"max_num_images": 5,
|
| 31 |
+
"max_topk": 256,
|
| 32 |
+
"n_active_tokens": 2048,
|
| 33 |
+
"neuronxcc_version": "2.26.6360.0+6f180f47",
|
| 34 |
+
"on_device_sampling": true,
|
| 35 |
+
"optimum_neuron_version": "0.4.7.dev0",
|
| 36 |
+
"output_logits": false,
|
| 37 |
+
"pp_degree": 1,
|
| 38 |
+
"prefill_chunk_size": 0,
|
| 39 |
+
"sequence_length": 2048,
|
| 40 |
+
"speculation_length": 0,
|
| 41 |
+
"start_rank_id": 0,
|
| 42 |
+
"target": "trn1",
|
| 43 |
+
"torch_dtype": "bfloat16",
|
| 44 |
+
"tp_degree": 2
|
| 45 |
+
},
|
| 46 |
+
"text_config": {
|
| 47 |
+
"_sliding_window_pattern": 6,
|
| 48 |
+
"attention_bias": false,
|
| 49 |
+
"attention_dropout": 0.0,
|
| 50 |
+
"attn_logit_softcapping": null,
|
| 51 |
+
"bos_token_id": 2,
|
| 52 |
+
"cache_implementation": "hybrid",
|
| 53 |
+
"dtype": "bfloat16",
|
| 54 |
+
"eos_token_id": 1,
|
| 55 |
+
"final_logit_softcapping": null,
|
| 56 |
+
"head_dim": 256,
|
| 57 |
+
"hidden_activation": "gelu_pytorch_tanh",
|
| 58 |
+
"hidden_size": 2560,
|
| 59 |
+
"initializer_range": 0.02,
|
| 60 |
+
"intermediate_size": 10240,
|
| 61 |
+
"layer_types": [
|
| 62 |
+
"sliding_attention",
|
| 63 |
+
"sliding_attention",
|
| 64 |
+
"sliding_attention",
|
| 65 |
+
"sliding_attention",
|
| 66 |
+
"sliding_attention",
|
| 67 |
+
"full_attention",
|
| 68 |
+
"sliding_attention",
|
| 69 |
+
"sliding_attention",
|
| 70 |
+
"sliding_attention",
|
| 71 |
+
"sliding_attention",
|
| 72 |
+
"sliding_attention",
|
| 73 |
+
"full_attention",
|
| 74 |
+
"sliding_attention",
|
| 75 |
+
"sliding_attention",
|
| 76 |
+
"sliding_attention",
|
| 77 |
+
"sliding_attention",
|
| 78 |
+
"sliding_attention",
|
| 79 |
+
"full_attention",
|
| 80 |
+
"sliding_attention",
|
| 81 |
+
"sliding_attention",
|
| 82 |
+
"sliding_attention",
|
| 83 |
+
"sliding_attention",
|
| 84 |
+
"sliding_attention",
|
| 85 |
+
"full_attention",
|
| 86 |
+
"sliding_attention",
|
| 87 |
+
"sliding_attention",
|
| 88 |
+
"sliding_attention",
|
| 89 |
+
"sliding_attention",
|
| 90 |
+
"sliding_attention",
|
| 91 |
+
"full_attention",
|
| 92 |
+
"sliding_attention",
|
| 93 |
+
"sliding_attention",
|
| 94 |
+
"sliding_attention",
|
| 95 |
+
"sliding_attention"
|
| 96 |
+
],
|
| 97 |
+
"max_position_embeddings": 131072,
|
| 98 |
+
"model_type": "gemma3_text",
|
| 99 |
+
"num_attention_heads": 8,
|
| 100 |
+
"num_hidden_layers": 34,
|
| 101 |
+
"num_key_value_heads": 4,
|
| 102 |
+
"pad_token_id": 0,
|
| 103 |
+
"query_pre_attn_scalar": 256,
|
| 104 |
+
"rms_norm_eps": 1e-06,
|
| 105 |
+
"rope_parameters": {
|
| 106 |
+
"full_attention": {
|
| 107 |
+
"factor": 8.0,
|
| 108 |
+
"rope_theta": 1000000.0,
|
| 109 |
+
"rope_type": "linear"
|
| 110 |
+
},
|
| 111 |
+
"sliding_attention": {
|
| 112 |
+
"rope_theta": 10000.0,
|
| 113 |
+
"rope_type": "default"
|
| 114 |
+
}
|
| 115 |
+
},
|
| 116 |
+
"sliding_window": 1024,
|
| 117 |
+
"sliding_window_pattern": 6,
|
| 118 |
+
"tie_word_embeddings": true,
|
| 119 |
+
"use_bidirectional_attention": false,
|
| 120 |
+
"use_cache": true,
|
| 121 |
+
"vocab_size": 262208
|
| 122 |
+
},
|
| 123 |
+
"tie_word_embeddings": true,
|
| 124 |
+
"unsloth_fixed": true,
|
| 125 |
+
"vision_config": {
|
| 126 |
+
"attention_dropout": 0.0,
|
| 127 |
+
"dtype": "bfloat16",
|
| 128 |
+
"hidden_act": "gelu_pytorch_tanh",
|
| 129 |
+
"hidden_size": 1152,
|
| 130 |
+
"image_size": 896,
|
| 131 |
+
"intermediate_size": 4304,
|
| 132 |
+
"layer_norm_eps": 1e-06,
|
| 133 |
+
"model_type": "siglip_vision_model",
|
| 134 |
+
"num_attention_heads": 16,
|
| 135 |
+
"num_channels": 3,
|
| 136 |
+
"num_hidden_layers": 27,
|
| 137 |
+
"patch_size": 14,
|
| 138 |
+
"vision_use_head": false
|
| 139 |
+
}
|
| 140 |
+
}
|