diff --git a/.gitattributes b/.gitattributes index d7ff71acd7c0f97cc35d990afbf2278052d1652b..e39899ed7fabe25dc112236af03b16d67ec02bfe 100644 --- a/.gitattributes +++ b/.gitattributes @@ -15952,3 +15952,15 @@ neuronxcc-2.21.33363.0+82129205/MODULE_8f6a41a7be8028ed3e7f+ad2a59a7/model.neff neuronxcc-2.21.33363.0+82129205/MODULE_8f6a41a7be8028ed3e7f+ad2a59a7/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text neuronxcc-2.21.33363.0+82129205/MODULE_99a48037f942ccd7b85f+1912433e/model.neff filter=lfs diff=lfs merge=lfs -text neuronxcc-2.21.33363.0+82129205/MODULE_99a48037f942ccd7b85f+1912433e/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text +neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/model.neff filter=lfs diff=lfs merge=lfs -text +neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text +neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/model.neff filter=lfs diff=lfs merge=lfs -text +neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text +neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/model.neff filter=lfs diff=lfs merge=lfs -text +neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text +neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/model.neff filter=lfs diff=lfs merge=lfs -text +neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text +neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/model.neff filter=lfs diff=lfs merge=lfs -text +neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text +neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/model.neff filter=lfs diff=lfs merge=lfs -text +neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/2da7a00f0478d50ae1e7f75f085c5b2773b5f355f427c61cf34cb6febd629d96/107ee95948117697f6ea.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/2da7a00f0478d50ae1e7f75f085c5b2773b5f355f427c61cf34cb6febd629d96/107ee95948117697f6ea.json new file mode 100644 index 0000000000000000000000000000000000000000..e1086659e2ac9e11796f5cfcaef5cbf1420c71ac --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/2da7a00f0478d50ae1e7f75f085c5b2773b5f355f427c61cf34cb6febd629d96/107ee95948117697f6ea.json @@ -0,0 +1,59 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "yujiepan/phi-4-tiny-random", + "_task": "text-generation", + "architectures": [ + "Phi3ForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "auto_map": {}, + "dtype": "bfloat16", + "embd_pdrop": 0.0, + "hidden_act": "silu", + "hidden_size": 16, + "initializer_range": 0.02, + "intermediate_size": 32, + "max_position_embeddings": 16384, + "model_type": "phi3", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "yujiepan/phi-4-tiny-random", + "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": true, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "bfloat16", + "tp_degree": 2 + }, + "num_attention_heads": 2, + "num_hidden_layers": 2, + "num_key_value_heads": 1, + "original_max_position_embeddings": 16384, + "partial_rotary_factor": 1.0, + "resid_pdrop": 0.0, + "rms_norm_eps": 1e-05, + "rope_scaling": null, + "rope_theta": 250000, + "sliding_window": null, + "tie_word_embeddings": false, + "use_cache": true, + "vocab_size": 100352 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/4ab8140bc7eb4a553d95855c5c2be2cf8c0fbab21b823d76183b6f51e98b6fc5/9b5786cc5c1d4d1cb1ad.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/4ab8140bc7eb4a553d95855c5c2be2cf8c0fbab21b823d76183b6f51e98b6fc5/9b5786cc5c1d4d1cb1ad.json new file mode 100644 index 0000000000000000000000000000000000000000..059f32dbd358aa382aed534d5f3457f58625e3a7 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/4ab8140bc7eb4a553d95855c5c2be2cf8c0fbab21b823d76183b6f51e98b6fc5/9b5786cc5c1d4d1cb1ad.json @@ -0,0 +1,58 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "dacorvo/Mixtral-tiny", + "_task": "text-generation", + "architectures": [ + "MixtralForCausalLM" + ], + "attention_dropout": 0.0, + "dtype": "float16", + "head_dim": 32, + "hidden_act": "silu", + "hidden_size": 1024, + "initializer_range": 0.02, + "intermediate_size": 3584, + "max_position_embeddings": 1024, + "model_type": "mixtral", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "dacorvo/Mixtral-tiny", + "checkpoint_revision": "c557ba205ddff6ea911f4719e0d543d6c08356b6", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": false, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": false, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "float16", + "tp_degree": 2 + }, + "num_attention_heads": 32, + "num_experts_per_tok": 2, + "num_hidden_layers": 2, + "num_key_value_heads": 8, + "num_local_experts": 8, + "output_router_logits": false, + "rms_norm_eps": 1e-05, + "rope_theta": 10000.0, + "router_aux_loss_coef": 0.001, + "router_jitter_noise": 0.0, + "sliding_window": 4096, + "tie_word_embeddings": false, + "use_cache": true, + "vocab_size": 32000 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/4cb7aff9e2a15c151396f2b684013e39d6739f0dec83e5c9dabbfe9d5fcf77b7/7b7d1ad7b867b7c8a112.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/4cb7aff9e2a15c151396f2b684013e39d6739f0dec83e5c9dabbfe9d5fcf77b7/7b7d1ad7b867b7c8a112.json new file mode 100644 index 0000000000000000000000000000000000000000..e2eaabb1941ef54eae4d0177c5f235fcf5837b33 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/4cb7aff9e2a15c151396f2b684013e39d6739f0dec83e5c9dabbfe9d5fcf77b7/7b7d1ad7b867b7c8a112.json @@ -0,0 +1,82 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "Qwen/Qwen2.5-0.5B", + "_task": "text-generation", + "architectures": [ + "Qwen2ForCausalLM" + ], + "attention_dropout": 0.0, + "dtype": "bfloat16", + "hidden_act": "silu", + "hidden_size": 896, + "initializer_range": 0.02, + "intermediate_size": 4864, + "layer_types": [ + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention" + ], + "max_position_embeddings": 32768, + "max_window_layers": 24, + "model_type": "qwen2", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 2, + "capacity_factor": null, + "checkpoint_id": "Qwen/Qwen2.5-0.5B", + "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987", + "continuous_batching": true, + "ep_degree": 1, + "fused_qkv": false, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 2, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": false, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "bfloat16", + "tp_degree": 2 + }, + "num_attention_heads": 14, + "num_hidden_layers": 24, + "num_key_value_heads": 2, + "rms_norm_eps": 1e-06, + "rope_scaling": null, + "rope_theta": 1000000.0, + "sliding_window": null, + "tie_word_embeddings": true, + "use_cache": true, + "use_mrope": false, + "use_sliding_window": false, + "vocab_size": 151936 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/6454afdf3e9d66c7226c13a575b718845c25e53b0699600ba2bb4f883e9d841b/6aec79a6a472999ccbeb.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/6454afdf3e9d66c7226c13a575b718845c25e53b0699600ba2bb4f883e9d841b/6aec79a6a472999ccbeb.json new file mode 100644 index 0000000000000000000000000000000000000000..7dd4e2ceccc1880ed591d090a2d6aac644efacef --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/6454afdf3e9d66c7226c13a575b718845c25e53b0699600ba2bb4f883e9d841b/6aec79a6a472999ccbeb.json @@ -0,0 +1,62 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "llamafactory/tiny-random-Llama-3", + "_task": "text-generation", + "architectures": [ + "LlamaForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "float16", + "head_dim": 4, + "hidden_act": "silu", + "hidden_size": 16, + "initializer_range": 0.02, + "intermediate_size": 64, + "max_position_embeddings": 131072, + "mlp_bias": false, + "model_type": "llama", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "llamafactory/tiny-random-Llama-3", + "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": true, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "float16", + "tp_degree": 2 + }, + "num_attention_heads": 4, + "num_hidden_layers": 2, + "num_key_value_heads": 4, + "pretraining_tp": 1, + "rms_norm_eps": 1e-05, + "rope_scaling": { + "factor": 8.0, + "high_freq_factor": 4.0, + "low_freq_factor": 1.0, + "original_max_position_embeddings": 8192, + "rope_type": "llama3" + }, + "rope_theta": 500000.0, + "tie_word_embeddings": false, + "use_cache": true, + "vocab_size": 128256 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/6454afdf3e9d66c7226c13a575b718845c25e53b0699600ba2bb4f883e9d841b/8b812851d11ab6e4d2c0.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/6454afdf3e9d66c7226c13a575b718845c25e53b0699600ba2bb4f883e9d841b/8b812851d11ab6e4d2c0.json new file mode 100644 index 0000000000000000000000000000000000000000..72b1e86872154e1996bb66a1c6d6b99c6f486dc3 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/6454afdf3e9d66c7226c13a575b718845c25e53b0699600ba2bb4f883e9d841b/8b812851d11ab6e4d2c0.json @@ -0,0 +1,62 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "llamafactory/tiny-random-Llama-3", + "_task": "text-generation", + "architectures": [ + "LlamaForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "float16", + "head_dim": 4, + "hidden_act": "silu", + "hidden_size": 16, + "initializer_range": 0.02, + "intermediate_size": 64, + "max_position_embeddings": 131072, + "mlp_bias": false, + "model_type": "llama", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "llamafactory/tiny-random-Llama-3", + "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": true, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 512, + "max_topk": 256, + "n_active_tokens": 512, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 512, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "float16", + "tp_degree": 2 + }, + "num_attention_heads": 4, + "num_hidden_layers": 2, + "num_key_value_heads": 4, + "pretraining_tp": 1, + "rms_norm_eps": 1e-05, + "rope_scaling": { + "factor": 8.0, + "high_freq_factor": 4.0, + "low_freq_factor": 1.0, + "original_max_position_embeddings": 8192, + "rope_type": "llama3" + }, + "rope_theta": 500000.0, + "tie_word_embeddings": false, + "use_cache": true, + "vocab_size": 128256 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/6454afdf3e9d66c7226c13a575b718845c25e53b0699600ba2bb4f883e9d841b/e445c13fd907bbaa0de0.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/6454afdf3e9d66c7226c13a575b718845c25e53b0699600ba2bb4f883e9d841b/e445c13fd907bbaa0de0.json new file mode 100644 index 0000000000000000000000000000000000000000..4f559dd45eab352de700050ce4f975f3c1bde49b --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/6454afdf3e9d66c7226c13a575b718845c25e53b0699600ba2bb4f883e9d841b/e445c13fd907bbaa0de0.json @@ -0,0 +1,62 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "llamafactory/tiny-random-Llama-3", + "_task": "text-generation", + "architectures": [ + "LlamaForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "float16", + "head_dim": 4, + "hidden_act": "silu", + "hidden_size": 16, + "initializer_range": 0.02, + "intermediate_size": 64, + "max_position_embeddings": 131072, + "mlp_bias": false, + "model_type": "llama", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 2, + "capacity_factor": null, + "checkpoint_id": "llamafactory/tiny-random-Llama-3", + "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", + "continuous_batching": true, + "ep_degree": 1, + "fused_qkv": true, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 2, + "max_context_length": 128, + "max_topk": 256, + "n_active_tokens": 128, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 128, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn2", + "torch_dtype": "float16", + "tp_degree": 2 + }, + "num_attention_heads": 4, + "num_hidden_layers": 2, + "num_key_value_heads": 4, + "pretraining_tp": 1, + "rms_norm_eps": 1e-05, + "rope_scaling": { + "factor": 8.0, + "high_freq_factor": 4.0, + "low_freq_factor": 1.0, + "original_max_position_embeddings": 8192, + "rope_type": "llama3" + }, + "rope_theta": 500000.0, + "tie_word_embeddings": false, + "use_cache": true, + "vocab_size": 128256 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/73707b485eab9008c7aba7f5dad0ce2384ac685318d5f888c12fa0d81ed90b19/588f7836eb16c9483d90.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/73707b485eab9008c7aba7f5dad0ce2384ac685318d5f888c12fa0d81ed90b19/588f7836eb16c9483d90.json new file mode 100644 index 0000000000000000000000000000000000000000..2e6f473d6b2048c556b258aaa8afd2cd9e2bc2e3 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/73707b485eab9008c7aba7f5dad0ce2384ac685318d5f888c12fa0d81ed90b19/588f7836eb16c9483d90.json @@ -0,0 +1,134 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "HuggingFaceTB/SmolLM3-3B", + "_task": "text-generation", + "architectures": [ + "SmolLM3ForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "bfloat16", + "hidden_act": "silu", + "hidden_size": 2048, + "initializer_range": 0.02, + "intermediate_size": 11008, + "layer_types": [ + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention" + ], + "max_position_embeddings": 65536, + "max_window_layers": 28, + "mlp_bias": false, + "model_type": "smollm3", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "HuggingFaceTB/SmolLM3-3B", + "checkpoint_revision": "a07cc9a04f16550a088caea529712d1d335b0ac1", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": true, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "bfloat16", + "tp_degree": 2 + }, + "no_rope_layer_interval": 4, + "no_rope_layers": [ + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0 + ], + "num_attention_heads": 16, + "num_hidden_layers": 36, + "num_key_value_heads": 4, + "pretraining_tp": 2, + "rms_norm_eps": 1e-06, + "rope_scaling": null, + "rope_theta": 5000000.0, + "sliding_window": null, + "use_cache": false, + "use_sliding_window": false, + "vocab_size": 128256 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/7518518c7e077820070186deda960d8cc49db068cdf0ac70664098fa2b6b698c/176587a2c1bb4b27acdc.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/7518518c7e077820070186deda960d8cc49db068cdf0ac70664098fa2b6b698c/176587a2c1bb4b27acdc.json new file mode 100644 index 0000000000000000000000000000000000000000..bf1f213aa1df9b1c7ebac7c012627a267b026a9f --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/7518518c7e077820070186deda960d8cc49db068cdf0ac70664098fa2b6b698c/176587a2c1bb4b27acdc.json @@ -0,0 +1,64 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "yujiepan/qwen2.5-128k-tiny-random", + "_task": "text-generation", + "architectures": [ + "Qwen2ForCausalLM" + ], + "attention_dropout": 0.0, + "dtype": "bfloat16", + "hidden_act": "silu", + "hidden_size": 8, + "initializer_range": 0.02, + "intermediate_size": 16, + "layer_types": [ + "full_attention", + "full_attention" + ], + "max_position_embeddings": 32768, + "max_window_layers": 1, + "model_type": "qwen2", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "yujiepan/qwen2.5-128k-tiny-random", + "checkpoint_revision": "c8296d4ca3f87782876d2382fbb6481d1beb8ef0", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": false, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "bfloat16", + "tp_degree": 2 + }, + "num_attention_heads": 4, + "num_hidden_layers": 2, + "num_key_value_heads": 2, + "rms_norm_eps": 1e-06, + "rope_scaling": { + "factor": 4.0, + "original_max_position_embeddings": 32768, + "rope_type": "yarn", + "type": "yarn" + }, + "rope_theta": 1000000.0, + "sliding_window": null, + "tie_word_embeddings": false, + "use_cache": true, + "use_sliding_window": false, + "vocab_size": 152064 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/920f44ce6d3e004d1ce547ae06644f7be262180644b04573153aa15d98742edc/7a16cff37a481cbfcd0b.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/920f44ce6d3e004d1ce547ae06644f7be262180644b04573153aa15d98742edc/7a16cff37a481cbfcd0b.json new file mode 100644 index 0000000000000000000000000000000000000000..d95f04fd78ec5b78dd93a5ff64f378653eda16da --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/920f44ce6d3e004d1ce547ae06644f7be262180644b04573153aa15d98742edc/7a16cff37a481cbfcd0b.json @@ -0,0 +1,65 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "optimum-internal-testing/tiny-random-qwen3_moe", + "_task": "text-generation", + "architectures": [ + "Qwen3MoeForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "decoder_sparse_step": 2, + "dtype": "float32", + "head_dim": 32, + "hidden_act": "silu", + "hidden_size": 64, + "initializer_range": 0.02, + "intermediate_size": 128, + "max_position_embeddings": 40960, + "max_window_layers": 1, + "mlp_only_layers": [], + "model_type": "qwen3_moe", + "moe_intermediate_size": 128, + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "optimum-internal-testing/tiny-random-qwen3_moe", + "checkpoint_revision": "e0230be2839556b44b7400a233c73c74b4abb7af", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": false, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "float32", + "tp_degree": 2 + }, + "norm_topk_prob": true, + "num_attention_heads": 2, + "num_experts": 8, + "num_experts_per_tok": 2, + "num_hidden_layers": 2, + "num_key_value_heads": 1, + "output_router_logits": false, + "rms_norm_eps": 1e-06, + "rope_scaling": null, + "rope_theta": 1000000.0, + "router_aux_loss_coef": 0.001, + "sliding_window": null, + "tie_word_embeddings": true, + "use_cache": true, + "use_sliding_window": false, + "vocab_size": 151936 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/9c24a2080ec26e9c74d5bd90866f74131aa4d5c975415b96f5e6600f081d5647/6c7b81b39491fd5f846a.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/9c24a2080ec26e9c74d5bd90866f74131aa4d5c975415b96f5e6600f081d5647/6c7b81b39491fd5f846a.json new file mode 100644 index 0000000000000000000000000000000000000000..286f79be545b140159e21145d2e888f0cbf654fe --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/9c24a2080ec26e9c74d5bd90866f74131aa4d5c975415b96f5e6600f081d5647/6c7b81b39491fd5f846a.json @@ -0,0 +1,81 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "tiny-random/llama-4", + "_task": "text-generation", + "attention_bias": false, + "attention_chunk_size": 128, + "attention_dropout": 0.0, + "attn_scale": 0.1, + "attn_temperature_tuning": 4, + "cache_implementation": "hybrid", + "dtype": "bfloat16", + "floor_scale": 8192, + "for_llm_compressor": false, + "head_dim": 32, + "hidden_act": "silu", + "hidden_size": 32, + "initializer_range": 0.02, + "interleave_moe_layer_step": 2, + "intermediate_size": 64, + "intermediate_size_mlp": 128, + "layer_types": [ + "chunked_attention", + "chunked_attention", + "chunked_attention", + "full_attention" + ], + "max_position_embeddings": 1048576, + "model_type": "llama4_text", + "moe_layers": [ + 1, + 3 + ], + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "tiny-random/llama-4", + "checkpoint_revision": "9e716f5d4d1ffe0a44a15f46f4a12b840439aba4", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": false, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "bfloat16", + "tp_degree": 2 + }, + "no_rope_layers": [ + 1, + 1, + 1, + 0 + ], + "num_attention_heads": 1, + "num_experts_per_tok": 1, + "num_hidden_layers": 4, + "num_key_value_heads": 1, + "num_local_experts": 8, + "output_router_logits": false, + "rms_norm_eps": 1e-05, + "rope_scaling": null, + "rope_theta": 500000.0, + "router_aux_loss_coef": 0.001, + "router_jitter_noise": 0.0, + "tie_word_embeddings": true, + "use_cache": true, + "use_qk_norm": true, + "vocab_size": 202048 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/cf6b9a360dcf294104671106bae2adbd9fd291823bb60a351883163684073231/20d8b3e0a4929b2eea5e.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/cf6b9a360dcf294104671106bae2adbd9fd291823bb60a351883163684073231/20d8b3e0a4929b2eea5e.json new file mode 100644 index 0000000000000000000000000000000000000000..3222bdf97841be546b724786af1a6aa1ee28b5df --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/cf6b9a360dcf294104671106bae2adbd9fd291823bb60a351883163684073231/20d8b3e0a4929b2eea5e.json @@ -0,0 +1,63 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "unsloth/Llama-3.2-1B-Instruct", + "_task": "text-generation", + "architectures": [ + "LlamaForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "bfloat16", + "head_dim": 64, + "hidden_act": "silu", + "hidden_size": 2048, + "initializer_range": 0.02, + "intermediate_size": 8192, + "max_position_embeddings": 131072, + "mlp_bias": false, + "model_type": "llama", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "unsloth/Llama-3.2-1B-Instruct", + "checkpoint_revision": "5a8abab4a5d6f164389b1079fb721cfab8d7126c", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": true, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 4096, + "max_topk": 256, + "n_active_tokens": 4096, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 4096, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "bfloat16", + "tp_degree": 2 + }, + "num_attention_heads": 32, + "num_hidden_layers": 16, + "num_key_value_heads": 8, + "pretraining_tp": 1, + "rms_norm_eps": 1e-05, + "rope_scaling": { + "factor": 32.0, + "high_freq_factor": 4.0, + "low_freq_factor": 1.0, + "original_max_position_embeddings": 8192, + "rope_type": "llama3" + }, + "rope_theta": 500000.0, + "tie_word_embeddings": true, + "unsloth_fixed": true, + "use_cache": true, + "vocab_size": 128256 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/d139acf64685f15794bb983ff6eb881bdd31304bae88b0ce1ed20a54c21f2265/91b521a6f25588a85a5d.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/d139acf64685f15794bb983ff6eb881bdd31304bae88b0ce1ed20a54c21f2265/91b521a6f25588a85a5d.json new file mode 100644 index 0000000000000000000000000000000000000000..53a6552773fc86a3794ad59eba5633a97aa60baf --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/d139acf64685f15794bb983ff6eb881bdd31304bae88b0ce1ed20a54c21f2265/91b521a6f25588a85a5d.json @@ -0,0 +1,58 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "hf-internal-testing/tiny-random-GraniteForCausalLM", + "_task": "text-generation", + "architectures": [ + "GraniteForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "attention_multiplier": 1.0, + "dtype": "float32", + "embedding_multiplier": 1.0, + "hidden_act": "silu", + "hidden_size": 32, + "initializer_range": 0.02, + "intermediate_size": 64, + "logits_scaling": 1.0, + "max_position_embeddings": 2048, + "mlp_bias": false, + "model_type": "granite", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "hf-internal-testing/tiny-random-GraniteForCausalLM", + "checkpoint_revision": "c3074ebc0ac2fe545305f5e5f6cce2cc9b2aa0c5", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": true, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "float32", + "tp_degree": 2 + }, + "num_attention_heads": 4, + "num_hidden_layers": 2, + "num_key_value_heads": 4, + "residual_multiplier": 1.0, + "rms_norm_eps": 1e-06, + "rope_scaling": null, + "rope_theta": 10000.0, + "tie_word_embeddings": false, + "use_cache": true, + "vocab_size": 49152 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/granite/hf-internal-testing/tiny-random-GraniteForCausalLM/91b521a6f25588a85a5d.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/granite/hf-internal-testing/tiny-random-GraniteForCausalLM/91b521a6f25588a85a5d.json new file mode 100644 index 0000000000000000000000000000000000000000..53a6552773fc86a3794ad59eba5633a97aa60baf --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/granite/hf-internal-testing/tiny-random-GraniteForCausalLM/91b521a6f25588a85a5d.json @@ -0,0 +1,58 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "hf-internal-testing/tiny-random-GraniteForCausalLM", + "_task": "text-generation", + "architectures": [ + "GraniteForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "attention_multiplier": 1.0, + "dtype": "float32", + "embedding_multiplier": 1.0, + "hidden_act": "silu", + "hidden_size": 32, + "initializer_range": 0.02, + "intermediate_size": 64, + "logits_scaling": 1.0, + "max_position_embeddings": 2048, + "mlp_bias": false, + "model_type": "granite", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "hf-internal-testing/tiny-random-GraniteForCausalLM", + "checkpoint_revision": "c3074ebc0ac2fe545305f5e5f6cce2cc9b2aa0c5", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": true, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "float32", + "tp_degree": 2 + }, + "num_attention_heads": 4, + "num_hidden_layers": 2, + "num_key_value_heads": 4, + "residual_multiplier": 1.0, + "rms_norm_eps": 1e-06, + "rope_scaling": null, + "rope_theta": 10000.0, + "tie_word_embeddings": false, + "use_cache": true, + "vocab_size": 49152 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/llama/llamafactory/tiny-random-Llama-3/6aec79a6a472999ccbeb.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/llama/llamafactory/tiny-random-Llama-3/6aec79a6a472999ccbeb.json new file mode 100644 index 0000000000000000000000000000000000000000..7dd4e2ceccc1880ed591d090a2d6aac644efacef --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/llama/llamafactory/tiny-random-Llama-3/6aec79a6a472999ccbeb.json @@ -0,0 +1,62 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "llamafactory/tiny-random-Llama-3", + "_task": "text-generation", + "architectures": [ + "LlamaForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "float16", + "head_dim": 4, + "hidden_act": "silu", + "hidden_size": 16, + "initializer_range": 0.02, + "intermediate_size": 64, + "max_position_embeddings": 131072, + "mlp_bias": false, + "model_type": "llama", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "llamafactory/tiny-random-Llama-3", + "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": true, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "float16", + "tp_degree": 2 + }, + "num_attention_heads": 4, + "num_hidden_layers": 2, + "num_key_value_heads": 4, + "pretraining_tp": 1, + "rms_norm_eps": 1e-05, + "rope_scaling": { + "factor": 8.0, + "high_freq_factor": 4.0, + "low_freq_factor": 1.0, + "original_max_position_embeddings": 8192, + "rope_type": "llama3" + }, + "rope_theta": 500000.0, + "tie_word_embeddings": false, + "use_cache": true, + "vocab_size": 128256 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/llama/unsloth/Llama-3.2-1B-Instruct/20d8b3e0a4929b2eea5e.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/llama/unsloth/Llama-3.2-1B-Instruct/20d8b3e0a4929b2eea5e.json new file mode 100644 index 0000000000000000000000000000000000000000..3222bdf97841be546b724786af1a6aa1ee28b5df --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/llama/unsloth/Llama-3.2-1B-Instruct/20d8b3e0a4929b2eea5e.json @@ -0,0 +1,63 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "unsloth/Llama-3.2-1B-Instruct", + "_task": "text-generation", + "architectures": [ + "LlamaForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "bfloat16", + "head_dim": 64, + "hidden_act": "silu", + "hidden_size": 2048, + "initializer_range": 0.02, + "intermediate_size": 8192, + "max_position_embeddings": 131072, + "mlp_bias": false, + "model_type": "llama", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "unsloth/Llama-3.2-1B-Instruct", + "checkpoint_revision": "5a8abab4a5d6f164389b1079fb721cfab8d7126c", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": true, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 4096, + "max_topk": 256, + "n_active_tokens": 4096, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 4096, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "bfloat16", + "tp_degree": 2 + }, + "num_attention_heads": 32, + "num_hidden_layers": 16, + "num_key_value_heads": 8, + "pretraining_tp": 1, + "rms_norm_eps": 1e-05, + "rope_scaling": { + "factor": 32.0, + "high_freq_factor": 4.0, + "low_freq_factor": 1.0, + "original_max_position_embeddings": 8192, + "rope_type": "llama3" + }, + "rope_theta": 500000.0, + "tie_word_embeddings": true, + "unsloth_fixed": true, + "use_cache": true, + "vocab_size": 128256 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/llama4_text/tiny-random/llama-4/6c7b81b39491fd5f846a.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/llama4_text/tiny-random/llama-4/6c7b81b39491fd5f846a.json new file mode 100644 index 0000000000000000000000000000000000000000..286f79be545b140159e21145d2e888f0cbf654fe --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/llama4_text/tiny-random/llama-4/6c7b81b39491fd5f846a.json @@ -0,0 +1,81 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "tiny-random/llama-4", + "_task": "text-generation", + "attention_bias": false, + "attention_chunk_size": 128, + "attention_dropout": 0.0, + "attn_scale": 0.1, + "attn_temperature_tuning": 4, + "cache_implementation": "hybrid", + "dtype": "bfloat16", + "floor_scale": 8192, + "for_llm_compressor": false, + "head_dim": 32, + "hidden_act": "silu", + "hidden_size": 32, + "initializer_range": 0.02, + "interleave_moe_layer_step": 2, + "intermediate_size": 64, + "intermediate_size_mlp": 128, + "layer_types": [ + "chunked_attention", + "chunked_attention", + "chunked_attention", + "full_attention" + ], + "max_position_embeddings": 1048576, + "model_type": "llama4_text", + "moe_layers": [ + 1, + 3 + ], + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "tiny-random/llama-4", + "checkpoint_revision": "9e716f5d4d1ffe0a44a15f46f4a12b840439aba4", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": false, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "bfloat16", + "tp_degree": 2 + }, + "no_rope_layers": [ + 1, + 1, + 1, + 0 + ], + "num_attention_heads": 1, + "num_experts_per_tok": 1, + "num_hidden_layers": 4, + "num_key_value_heads": 1, + "num_local_experts": 8, + "output_router_logits": false, + "rms_norm_eps": 1e-05, + "rope_scaling": null, + "rope_theta": 500000.0, + "router_aux_loss_coef": 0.001, + "router_jitter_noise": 0.0, + "tie_word_embeddings": true, + "use_cache": true, + "use_qk_norm": true, + "vocab_size": 202048 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/mixtral/dacorvo/Mixtral-tiny/9b5786cc5c1d4d1cb1ad.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/mixtral/dacorvo/Mixtral-tiny/9b5786cc5c1d4d1cb1ad.json new file mode 100644 index 0000000000000000000000000000000000000000..059f32dbd358aa382aed534d5f3457f58625e3a7 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/mixtral/dacorvo/Mixtral-tiny/9b5786cc5c1d4d1cb1ad.json @@ -0,0 +1,58 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "dacorvo/Mixtral-tiny", + "_task": "text-generation", + "architectures": [ + "MixtralForCausalLM" + ], + "attention_dropout": 0.0, + "dtype": "float16", + "head_dim": 32, + "hidden_act": "silu", + "hidden_size": 1024, + "initializer_range": 0.02, + "intermediate_size": 3584, + "max_position_embeddings": 1024, + "model_type": "mixtral", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "dacorvo/Mixtral-tiny", + "checkpoint_revision": "c557ba205ddff6ea911f4719e0d543d6c08356b6", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": false, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": false, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "float16", + "tp_degree": 2 + }, + "num_attention_heads": 32, + "num_experts_per_tok": 2, + "num_hidden_layers": 2, + "num_key_value_heads": 8, + "num_local_experts": 8, + "output_router_logits": false, + "rms_norm_eps": 1e-05, + "rope_theta": 10000.0, + "router_aux_loss_coef": 0.001, + "router_jitter_noise": 0.0, + "sliding_window": 4096, + "tie_word_embeddings": false, + "use_cache": true, + "vocab_size": 32000 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/phi3/yujiepan/phi-4-tiny-random/107ee95948117697f6ea.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/phi3/yujiepan/phi-4-tiny-random/107ee95948117697f6ea.json new file mode 100644 index 0000000000000000000000000000000000000000..e1086659e2ac9e11796f5cfcaef5cbf1420c71ac --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/phi3/yujiepan/phi-4-tiny-random/107ee95948117697f6ea.json @@ -0,0 +1,59 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "yujiepan/phi-4-tiny-random", + "_task": "text-generation", + "architectures": [ + "Phi3ForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "auto_map": {}, + "dtype": "bfloat16", + "embd_pdrop": 0.0, + "hidden_act": "silu", + "hidden_size": 16, + "initializer_range": 0.02, + "intermediate_size": 32, + "max_position_embeddings": 16384, + "model_type": "phi3", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "yujiepan/phi-4-tiny-random", + "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": true, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "bfloat16", + "tp_degree": 2 + }, + "num_attention_heads": 2, + "num_hidden_layers": 2, + "num_key_value_heads": 1, + "original_max_position_embeddings": 16384, + "partial_rotary_factor": 1.0, + "resid_pdrop": 0.0, + "rms_norm_eps": 1e-05, + "rope_scaling": null, + "rope_theta": 250000, + "sliding_window": null, + "tie_word_embeddings": false, + "use_cache": true, + "vocab_size": 100352 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/qwen2/Qwen/Qwen2.5-0.5B/7b7d1ad7b867b7c8a112.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/qwen2/Qwen/Qwen2.5-0.5B/7b7d1ad7b867b7c8a112.json new file mode 100644 index 0000000000000000000000000000000000000000..e2eaabb1941ef54eae4d0177c5f235fcf5837b33 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/qwen2/Qwen/Qwen2.5-0.5B/7b7d1ad7b867b7c8a112.json @@ -0,0 +1,82 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "Qwen/Qwen2.5-0.5B", + "_task": "text-generation", + "architectures": [ + "Qwen2ForCausalLM" + ], + "attention_dropout": 0.0, + "dtype": "bfloat16", + "hidden_act": "silu", + "hidden_size": 896, + "initializer_range": 0.02, + "intermediate_size": 4864, + "layer_types": [ + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention" + ], + "max_position_embeddings": 32768, + "max_window_layers": 24, + "model_type": "qwen2", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 2, + "capacity_factor": null, + "checkpoint_id": "Qwen/Qwen2.5-0.5B", + "checkpoint_revision": "060db6499f32faf8b98477b0a26969ef7d8b9987", + "continuous_batching": true, + "ep_degree": 1, + "fused_qkv": false, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 2, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": false, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "bfloat16", + "tp_degree": 2 + }, + "num_attention_heads": 14, + "num_hidden_layers": 24, + "num_key_value_heads": 2, + "rms_norm_eps": 1e-06, + "rope_scaling": null, + "rope_theta": 1000000.0, + "sliding_window": null, + "tie_word_embeddings": true, + "use_cache": true, + "use_mrope": false, + "use_sliding_window": false, + "vocab_size": 151936 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/qwen2/yujiepan/qwen2.5-128k-tiny-random/176587a2c1bb4b27acdc.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/qwen2/yujiepan/qwen2.5-128k-tiny-random/176587a2c1bb4b27acdc.json new file mode 100644 index 0000000000000000000000000000000000000000..bf1f213aa1df9b1c7ebac7c012627a267b026a9f --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/qwen2/yujiepan/qwen2.5-128k-tiny-random/176587a2c1bb4b27acdc.json @@ -0,0 +1,64 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "yujiepan/qwen2.5-128k-tiny-random", + "_task": "text-generation", + "architectures": [ + "Qwen2ForCausalLM" + ], + "attention_dropout": 0.0, + "dtype": "bfloat16", + "hidden_act": "silu", + "hidden_size": 8, + "initializer_range": 0.02, + "intermediate_size": 16, + "layer_types": [ + "full_attention", + "full_attention" + ], + "max_position_embeddings": 32768, + "max_window_layers": 1, + "model_type": "qwen2", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "yujiepan/qwen2.5-128k-tiny-random", + "checkpoint_revision": "c8296d4ca3f87782876d2382fbb6481d1beb8ef0", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": false, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "bfloat16", + "tp_degree": 2 + }, + "num_attention_heads": 4, + "num_hidden_layers": 2, + "num_key_value_heads": 2, + "rms_norm_eps": 1e-06, + "rope_scaling": { + "factor": 4.0, + "original_max_position_embeddings": 32768, + "rope_type": "yarn", + "type": "yarn" + }, + "rope_theta": 1000000.0, + "sliding_window": null, + "tie_word_embeddings": false, + "use_cache": true, + "use_sliding_window": false, + "vocab_size": 152064 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/qwen3_moe/optimum-internal-testing/tiny-random-qwen3_moe/7a16cff37a481cbfcd0b.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/qwen3_moe/optimum-internal-testing/tiny-random-qwen3_moe/7a16cff37a481cbfcd0b.json new file mode 100644 index 0000000000000000000000000000000000000000..d95f04fd78ec5b78dd93a5ff64f378653eda16da --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/qwen3_moe/optimum-internal-testing/tiny-random-qwen3_moe/7a16cff37a481cbfcd0b.json @@ -0,0 +1,65 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "optimum-internal-testing/tiny-random-qwen3_moe", + "_task": "text-generation", + "architectures": [ + "Qwen3MoeForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "decoder_sparse_step": 2, + "dtype": "float32", + "head_dim": 32, + "hidden_act": "silu", + "hidden_size": 64, + "initializer_range": 0.02, + "intermediate_size": 128, + "max_position_embeddings": 40960, + "max_window_layers": 1, + "mlp_only_layers": [], + "model_type": "qwen3_moe", + "moe_intermediate_size": 128, + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "optimum-internal-testing/tiny-random-qwen3_moe", + "checkpoint_revision": "e0230be2839556b44b7400a233c73c74b4abb7af", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": false, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "float32", + "tp_degree": 2 + }, + "norm_topk_prob": true, + "num_attention_heads": 2, + "num_experts": 8, + "num_experts_per_tok": 2, + "num_hidden_layers": 2, + "num_key_value_heads": 1, + "output_router_logits": false, + "rms_norm_eps": 1e-06, + "rope_scaling": null, + "rope_theta": 1000000.0, + "router_aux_loss_coef": 0.001, + "sliding_window": null, + "tie_word_embeddings": true, + "use_cache": true, + "use_sliding_window": false, + "vocab_size": 151936 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/smollm3/HuggingFaceTB/SmolLM3-3B/588f7836eb16c9483d90.json b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/smollm3/HuggingFaceTB/SmolLM3-3B/588f7836eb16c9483d90.json new file mode 100644 index 0000000000000000000000000000000000000000..2e6f473d6b2048c556b258aaa8afd2cd9e2bc2e3 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.5.dev2/smollm3/HuggingFaceTB/SmolLM3-3B/588f7836eb16c9483d90.json @@ -0,0 +1,134 @@ +{ + "_entry_class": "SingleModelCacheEntry", + "_model_id": "HuggingFaceTB/SmolLM3-3B", + "_task": "text-generation", + "architectures": [ + "SmolLM3ForCausalLM" + ], + "attention_bias": false, + "attention_dropout": 0.0, + "dtype": "bfloat16", + "hidden_act": "silu", + "hidden_size": 2048, + "initializer_range": 0.02, + "intermediate_size": 11008, + "layer_types": [ + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention", + "full_attention" + ], + "max_position_embeddings": 65536, + "max_window_layers": 28, + "mlp_bias": false, + "model_type": "smollm3", + "neuron": { + "_serialized_key": "NxDNeuronConfig", + "batch_size": 1, + "capacity_factor": null, + "checkpoint_id": "HuggingFaceTB/SmolLM3-3B", + "checkpoint_revision": "a07cc9a04f16550a088caea529712d1d335b0ac1", + "continuous_batching": false, + "ep_degree": 1, + "fused_qkv": true, + "glu_mlp": true, + "local_ranks_size": 2, + "max_batch_size": 1, + "max_context_length": 1024, + "max_topk": 256, + "n_active_tokens": 1024, + "neuronxcc_version": "2.21.33363.0+82129205", + "on_device_sampling": true, + "optimum_neuron_version": "0.4.5.dev2", + "output_logits": false, + "pp_degree": 1, + "sequence_length": 1024, + "speculation_length": 0, + "start_rank_id": 0, + "target": "trn1", + "torch_dtype": "bfloat16", + "tp_degree": 2 + }, + "no_rope_layer_interval": 4, + "no_rope_layers": [ + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0, + 1, + 1, + 1, + 0 + ], + "num_attention_heads": 16, + "num_hidden_layers": 36, + "num_key_value_heads": 4, + "pretraining_tp": 2, + "rms_norm_eps": 1e-06, + "rope_scaling": null, + "rope_theta": 5000000.0, + "sliding_window": null, + "use_cache": false, + "use_sliding_window": false, + "vocab_size": 128256 +} \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_10217061096959125489+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10217061096959125489+e30acd3a/model.neff index 2502b17c85be4c7021e58ac256d3b39ac7c1190c..c2c3d36ae634a3e92d15eceed67453de56bc5adc 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10217061096959125489+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10217061096959125489+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_10244305442015770634+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10244305442015770634+e30acd3a/model.neff index 17f1026fe84d9206de5a2f0b2dabf90e6fb9d9f0..d29e1de1d8c65af5f410135cf56285b0ebd3db22 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10244305442015770634+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10244305442015770634+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_10606948783918825529+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10606948783918825529+e30acd3a/model.neff index bc30375f38da0a01a8cc4234db0b1432f7d407f6..519db27b6edb0e4a91f17fc6f494a85c456edeb4 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10606948783918825529+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10606948783918825529+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_10645643398657092095+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10645643398657092095+e30acd3a/model.neff index 00947069c07066c23871be22657fcb0edd1947b8..36dc9bbf089f0c2d3cd5ad992bb47e1edf42e32f 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10645643398657092095+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10645643398657092095+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_10661660426924300837+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10661660426924300837+e30acd3a/model.neff index a4a47383162672bfdf0a01c4c65e9c83de71c0a8..4b479dea02e58273fe4ba67e0495d0843d3b8a99 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10661660426924300837+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10661660426924300837+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_10746122569655005679+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_10746122569655005679+e30acd3a/model.neff index 2cb503ea9674503e008b42284334af20264a1761..887b46b5ab5181685b7da50b4415481c1d4c1ab3 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_10746122569655005679+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_10746122569655005679+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_11086318750207148626+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_11086318750207148626+e30acd3a/model.neff index 760a1a7e7bb90063097b816ec799cb72acf68cce..4ee760597d4877e3ebffac8c06ba2e8f77605726 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_11086318750207148626+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_11086318750207148626+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_11909310600244571805+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_11909310600244571805+e30acd3a/model.neff index 44b9b0f75e30c1cc7bf6e8ad6c79d00b5962710a..2dac6aed01cfc22eac79b0939db51dabb8160c95 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_11909310600244571805+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_11909310600244571805+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_11933818254123612383+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_11933818254123612383+e30acd3a/model.neff index c0c13a11a5224a35c3abf4be542cbb02997fff7b..855cd5c70661b4d007cfdb4fdecad4ffd9cba54d 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_11933818254123612383+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_11933818254123612383+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1210392327607194823+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1210392327607194823+e30acd3a/model.neff index 8ca1559942b30bdca331def890a451829a3b6a64..91d7f7e6368f4b3ecdd961119d81780cd89e7c54 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1210392327607194823+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1210392327607194823+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_12487216553200321032+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_12487216553200321032+e30acd3a/model.neff index 60931085580ee99df1e5f35d1fb5c5080fd3c846..72870084f62fcf001a3a38535626a2c0aca6303b 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_12487216553200321032+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_12487216553200321032+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_13085549342645515693+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_13085549342645515693+e30acd3a/model.neff index 1fd0d59eb2d702aae50ca0be3c4ac22e9a8f381f..16a50ea20de0ee392b1ca990ae1cae90558bbf07 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13085549342645515693+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13085549342645515693+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_13102010590082783346+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_13102010590082783346+e30acd3a/model.neff index 2d684b790ce96bf2ac645e1b2c4895c33a50cd0f..1cd80be9a96159d43efedd2d309fb65726234293 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13102010590082783346+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13102010590082783346+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_13475273426270778455+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_13475273426270778455+e30acd3a/model.neff index 3f492645c0e9b1b9b4460e7a4f483fbe1181f1c3..fa226d58d0f49fbc9e4623208c48a56115220fb9 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13475273426270778455+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13475273426270778455+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_13728813963059599796+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_13728813963059599796+e30acd3a/model.neff index 69b02c151807a31e678f89d756b754a3fb6e3f50..8d67d9ed3efdd2a8d9cfbb941205188a77eb2853 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13728813963059599796+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13728813963059599796+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_13783872956654054643+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_13783872956654054643+e30acd3a/model.neff index 692aa6d6e6fd3ba8732f7481198a478656a8cb8d..3e6906514d35ed8236df7178473a35b275f2c993 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_13783872956654054643+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_13783872956654054643+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_14097776130612710282+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_14097776130612710282+e30acd3a/model.neff index 6dad43b163e9e2698c33d96e3ca99ff4e5cf7312..3a5f53d086a28fe8a0f29c662fcb999021f719ac 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14097776130612710282+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14097776130612710282+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_14155105943057125178+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_14155105943057125178+e30acd3a/model.neff index 8d09241ff26a740d7f2aa5ee9e76fa790ba79c73..e104c73c9beda036651e003e3952ead9d6aab580 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14155105943057125178+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14155105943057125178+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_14173682194645188821+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_14173682194645188821+e30acd3a/model.neff index 5b66b03ad43d44a12d2fa09e7e81cf49d814559d..548700cc4f50af2af4e6f59c9a01555c5e0a7b96 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14173682194645188821+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14173682194645188821+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_14274833313744358348+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_14274833313744358348+e30acd3a/model.neff index ea54f54ac8258463bf881a53e96cfe1ac1b5134a..ecff29f9a3eff22138f1669dc0ce310b6cab9353 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14274833313744358348+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14274833313744358348+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_14469716438804935215+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_14469716438804935215+e30acd3a/model.neff index c9165b247947db6911f60eceb7d48ad94c092dfa..88d8a85109f780b283038cf404f43cfd4f7964de 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14469716438804935215+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14469716438804935215+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_14739672283771660808+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_14739672283771660808+e30acd3a/model.neff index 9dc04ffb4e018f3cf11b6f51ad8761beb7072b97..3c9f26d3d7b5721f89ed653c980eb82c52eee9c6 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_14739672283771660808+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_14739672283771660808+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15442663025941492357+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15442663025941492357+e30acd3a/model.neff index 4053b4d34c8318d94122fd4f95d5604e0e8e8fa5..eb31ff40e5f6f5101313a4c350e56ae725d097ea 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15442663025941492357+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15442663025941492357+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15557971880365771457+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15557971880365771457+e30acd3a/model.neff index 89c4ad493645b73ecf37e0ac06771ceda561a37a..05137d9d5bf0312d6698f7137f21de8267e0db55 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15557971880365771457+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15557971880365771457+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15564425602929126510+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15564425602929126510+e30acd3a/model.neff index 67ee0a884b79fd431cba350026d1f382fdc886cc..3fdda3c18dede2ad4f77b08c345a0eefa6bd695f 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15564425602929126510+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15564425602929126510+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15589838460896944293+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15589838460896944293+e30acd3a/model.neff index c809b60997fc3fca508d7b61da6962ead8669d17..5a3c110d3d42a8a091351c080781a9fea245c57c 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15589838460896944293+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15589838460896944293+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15654572849171857535+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15654572849171857535+e30acd3a/model.neff index 9cf84ecf2671accbb209d3563886012bd1c07497..95bee05d8612d0519d9151a06ee179969b92b20d 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15654572849171857535+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15654572849171857535+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15767383571209512795+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15767383571209512795+e30acd3a/model.neff index 065df8ded29e5b20bb72d8e09f5288134f11deb1..43e686161b7e989a7603ca55a02afaaf9dc7dd6c 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15767383571209512795+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15767383571209512795+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15888836045088309511+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15888836045088309511+e30acd3a/model.neff index 9a49189c2b36a0fc95814c572aa40b4501c4b9d2..e4c4dee4c2d9de1ece565abe2634303fc18811ff 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15888836045088309511+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15888836045088309511+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_15929451261464042997+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_15929451261464042997+e30acd3a/model.neff index 08c62b4ebc11e32bd294d311443790040111d5fa..18658a801e547b3ddd9908ee285ec98dabcab410 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_15929451261464042997+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_15929451261464042997+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_16016505958416521648+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_16016505958416521648+e30acd3a/model.neff index 6ce1661191fd241520b7ec215d8e35cfe02c6af7..a6c012aa4b99fc3d48e8bb8ff7822b9be71cc88e 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16016505958416521648+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16016505958416521648+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_16053163479112702088+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_16053163479112702088+e30acd3a/model.neff index f6ff2d2ab556a72ba41aafb98491ef953bb33634..19b71c9a853582a86d5d212ad3cadaf9fa77abaa 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16053163479112702088+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16053163479112702088+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_16204387852795926216+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_16204387852795926216+e30acd3a/model.neff index 3a874acb61aeaf9cb51dc20aaeb7704a88f342f1..7fca27247c9ad3534fd42516ea343e4bc4fb43d9 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16204387852795926216+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16204387852795926216+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1656719109221189948+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1656719109221189948+e30acd3a/model.neff index 0a0a7453d22d095c3b14d50f17da31798120b7a8..fa28ef49b5d6c6d16888167b0f199498a09a95a7 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1656719109221189948+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1656719109221189948+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_16599571375348449904+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_16599571375348449904+e30acd3a/model.neff index dd2d27420147b4ea4259ec7180cdba81dba18f71..525f374e2bfd2d24d0450c4f411b80e0edf0af39 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16599571375348449904+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16599571375348449904+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1663757798483801648+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1663757798483801648+e30acd3a/model.neff index f5477d80bef0fbc270840aad52be57014d1ab8f3..97d4ca28a2d03322eb01f5ae97d6941bcd084f3b 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1663757798483801648+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1663757798483801648+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_16738296820980389103+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_16738296820980389103+e30acd3a/model.neff index d4157ef94cd3361ca571f102ed961e73eca33b28..2c6868808f57ad76c518bbaee194a5ef032478c2 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_16738296820980389103+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_16738296820980389103+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1696825468766062114+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1696825468766062114+e30acd3a/model.neff index bb8ed9df4e06e8999d297d5328bf5b7355cf3cec..420caafb35dfddd14a3380b945b9f178f5035094 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1696825468766062114+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1696825468766062114+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff index bf85f02024cd2b40b290e60839a503ea8d4d5581..cd37c6eac292798f1062064a498c4e7f4469b354 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_17781596253725927902+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_17781596253725927902+e30acd3a/model.neff index 3dd24f774ede9106b32afdc9b4fc2145976d10f0..d37d16bec218f712aaace59072cb0c497a0e5619 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_17781596253725927902+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_17781596253725927902+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1795740353831178306+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1795740353831178306+e30acd3a/model.neff index 1cf100a85408980ad37685ffcee232ebd81e258a..b70ec42643d3d0fa494ae0e496d6d878b958f01b 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1795740353831178306+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1795740353831178306+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1800832390737682969+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1800832390737682969+e30acd3a/model.neff index c0b0d9264220fa6f3838177ddbf5c3ac9b8459f6..9ce693c066acc7b450a873538973aca3ab28eaef 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1800832390737682969+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1800832390737682969+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1805481651134498710+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1805481651134498710+e30acd3a/model.neff index 67fb2bc97e7ee2e3b0e55d68bb45fff58826f1e6..350bd934465b62bcc10da9dbfef4bbd8e3bc3625 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1805481651134498710+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1805481651134498710+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_1805728312057401221+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_1805728312057401221+e30acd3a/model.neff index e6ecce6db62780466ba2ebb9c10a3677ed337252..112dda4de5c296bc71f2096ff7ef8d0942787b29 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_1805728312057401221+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_1805728312057401221+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_18070749384531238695+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_18070749384531238695+e30acd3a/model.neff index bf499228c2cc4be7e99f65f7e8a1e6262ea08989..d034ad4ae22e1d04aa2d5b5ea6090dcc0e6a67f8 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_18070749384531238695+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_18070749384531238695+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_18134517905922687964+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_18134517905922687964+e30acd3a/model.neff index caeb0329fe0c02dba7235e95cbc52c950d44eb38..9a2f0961fdd29f1a8b62eec71e44e7733a828f97 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_18134517905922687964+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_18134517905922687964+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_18207475553554060390+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_18207475553554060390+e30acd3a/model.neff index 3addb2ad8835efe18f718f971a455035e83e5ab2..d65be3e7220e6ab7b6062ca883a1e830aaf65a06 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_18207475553554060390+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_18207475553554060390+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2157613318347839507+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2157613318347839507+e30acd3a/model.neff index d2998a3ab1619e9e79f4ff8e027fed1cee48fa1d..f5d309b3799e3796bdf5173e2ebed7aaa1240ce3 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2157613318347839507+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2157613318347839507+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2203200257461345827+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2203200257461345827+e30acd3a/model.neff index ac19936ca2535edda9c61341f6923969351ade28..7c89e9498dd7d94d703ce784ba341ce6516feca1 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2203200257461345827+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2203200257461345827+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2410426589418994260+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2410426589418994260+e30acd3a/model.neff index a75bd441c73f297c8a23c3281e04712921e75423..820f43dd31f8833f8aa4f6eb3539746566019fbc 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2410426589418994260+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2410426589418994260+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2426414314187505427+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2426414314187505427+e30acd3a/model.neff index 0ddf05511b1d0a8bb9098ee8d4d82af5746a6847..1fbb9a1bc1c4959a9b060cf55418c5eb1eea92ee 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2426414314187505427+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2426414314187505427+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/compile_flags.json new file mode 100644 index 0000000000000000000000000000000000000000..0d2015880aea8c86498952bf11f8da1e3a145e6d --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/compile_flags.json @@ -0,0 +1 @@ +["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_fecdf76b-aa3d-4101-afd4-9f269083439d/compiler_workdir/NeuronRMSNorm/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"] \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/model.done new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/model.hlo_module.pb new file mode 100644 index 0000000000000000000000000000000000000000..6a2752c25c9f7e1092e2b55c010694e168610ce3 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/model.hlo_module.pb @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1c774c6626fc91684e924b843c1fe807dadbbac6d8a8e19443663b43c13eac35 +size 1165 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/model.neff new file mode 100644 index 0000000000000000000000000000000000000000..b6cf79649a20f4969a4bb9c20063e694211e1d7e --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/model.neff @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:037890f0c02a8033bb3ca2c0e530db2870f73f9861e5eb1ff2edbfd346f59559 +size 103424 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/wrapped_neff.hlo new file mode 100644 index 0000000000000000000000000000000000000000..f9007d1da6a12936ace0c999a2b102836fef7e5c --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_24d449ff1eda1ccb1178+e036626c/wrapped_neff.hlo @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:fc37fe7c49b995a45ecd58c0b7b547be56a6924ba5d014bb36b00382c5d91966 +size 104320 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2576759111807165188+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2576759111807165188+e30acd3a/model.neff index 97a6ac2440234308f92ca0167c72c1a88c4644ad..b618bcfcc5d684739254ea064c597c37e9408335 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2576759111807165188+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2576759111807165188+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2650447136112456251+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2650447136112456251+e30acd3a/model.neff index 47dabdcfd476d2414302678220e9ac53bfd005f3..ba7ca8f717d9ec027f07dad526c32bf464a92ba8 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2650447136112456251+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2650447136112456251+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2682953264353234433+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2682953264353234433+e30acd3a/model.neff index c68cd6c95815bd4770664484a49aa3b34e60f63a..7c5acb343457445dddb348cade91fb1e86afb60d 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2682953264353234433+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2682953264353234433+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2699280712748688265+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2699280712748688265+e30acd3a/model.neff index 2ecb9766dae5cc10afdee5585f169a1bc02c33d4..72b9fee8cc6c31961735518e595f6ae2256f6a38 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2699280712748688265+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2699280712748688265+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2811495285170804454+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2811495285170804454+e30acd3a/model.neff index 40b2c73df14a8abcde65d8f3b300842cc2625ee3..198c8b0f14815a5a54e3f08375d2e4b222aaafff 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_2811495285170804454+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_2811495285170804454+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.hlo_module.pb index 4689fc8aa2bdb5983a59d32337c424c2f827a31d..5b31a795fa051fe6382001027af513aab21cff12 100644 --- a/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.hlo_module.pb +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.hlo_module.pb @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:180439fd9660c6cc011351fb67904d48b742214af4187642a641eba66539ffa8 +oid sha256:9a587d5ddb67a42533df39b1b05311d022b9cb500122dfc7b382e9f459a2b148 size 593552 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.neff index 69335340ec59822a30c0c944e665029e868641b1..13d02c16cb5ea0d32ba238579e508d72f0ae51a3 100644 --- a/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.neff +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_2ce3d3065088729c7d7f+24129607/model.neff @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:245045abb2489a33c9495252a741530264c0b8cf56bb8019f073f195705c4d28 +oid sha256:85a6ad470d547c3fc957fb4c49b5f5750462129784a186a9fa82b6c9a67390ce size 3769344 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_302950992539679003+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_302950992539679003+e30acd3a/model.neff index bd0d922655bc2d970b7c5ecaa0c093bc4afdbf69..ab40c226881b30a7a245e1bad7f98aeb5000062c 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_302950992539679003+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_302950992539679003+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4013984518400530928+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4013984518400530928+e30acd3a/model.neff index 06554f14e77d29410bd7258add3e184e1769c45c..42b225177c69fdc55d0b38ae9b30df3f007da08b 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4013984518400530928+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4013984518400530928+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4052264821257342969+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4052264821257342969+e30acd3a/model.neff index bab343c51759ad64dc6fbea6dbc8db460de1bf0a..58b609007446c205a67036b9522d80627254fc1a 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4052264821257342969+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4052264821257342969+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4166695227062860792+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4166695227062860792+e30acd3a/model.neff index e2f48f6c031bed7d8e9c15cffa745e1b9a446543..c23936df5d6ac08858d96cf709261a4497af33fd 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4166695227062860792+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4166695227062860792+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4232557591140652245+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4232557591140652245+e30acd3a/model.neff index 795aaf36c56118edf545ae2e8e7fdef54c4175d0..22e5d4deb4a51065763bb0fd20550ffa2a8f605f 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4232557591140652245+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4232557591140652245+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4606491920565381523+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4606491920565381523+e30acd3a/model.neff index f53ffba2b74c0a07924966983fbd3c0e250c405e..a316d5d04e36c87e814b436586cf88bf1e8da8bd 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4606491920565381523+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4606491920565381523+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4608333438381222049+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4608333438381222049+e30acd3a/model.neff index 0af27c67bc54a5785fcc2bb95ca9e367320cc5cf..d5cb27118c12c0d5da9cf1ef8d77e74c2db37788 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4608333438381222049+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4608333438381222049+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4683016291147007110+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4683016291147007110+e30acd3a/model.neff index 6ee5cd6e3fbd59d8818cf1989af9c8c3b400232c..263048744dfc5f792807bf95590c07b1f2b4c8d7 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4683016291147007110+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4683016291147007110+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4816422433712790690+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4816422433712790690+e30acd3a/model.neff index 51b86a6e3de6e13a3e466b032f67f5482e4bde87..786ff77e243223ee09c5006a57b67cc046cd1722 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4816422433712790690+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4816422433712790690+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4842952601010372434+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4842952601010372434+e30acd3a/model.neff index 7030222b50e062257afea3b887f56d50505e6d14..719567c6490298746bf69ab4c4d9b6c747bca668 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4842952601010372434+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4842952601010372434+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_4939721357859779936+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_4939721357859779936+e30acd3a/model.neff index c6566f06e25abc2dadbe11b5bcbea718b3d1ef3c..709abfed6c23d152b1f8431ccfd9214aaa6546aa 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_4939721357859779936+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_4939721357859779936+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_5397223456911199516+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_5397223456911199516+e30acd3a/model.neff index 4f0d95a80e18bdfd1d277d9b581915d8b053884a..1d075d006ef6bee99d01037a9f35b09bc04f97c2 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_5397223456911199516+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_5397223456911199516+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_5724544066416780383+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_5724544066416780383+e30acd3a/model.neff index da938c497550badaf2683848b0e4e41582023f0b..a6bd5cf2f9f433489c11bf6340e2214bf2d2a30f 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_5724544066416780383+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_5724544066416780383+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_5765916484866650909+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_5765916484866650909+e30acd3a/model.neff index 97f5e8ba348fe3a53c6bdd297517b01de668aec1..ef066097187d3506db6b7c0d1d61195a1d0e4357 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_5765916484866650909+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_5765916484866650909+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6368409506294683105+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_6368409506294683105+e30acd3a/model.neff index 2954b04301c0cb0da373d2de23a1c5800511ccd9..f5a64c81f355ae8cf594c634e152f3916be534ad 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_6368409506294683105+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_6368409506294683105+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6375780938374445148+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_6375780938374445148+e30acd3a/model.neff index dd4105d2e529fcae31e21cfdd6dfa5cbc23af5b7..467f96f1e4b477108d8a0c9255eeb098b2727500 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_6375780938374445148+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_6375780938374445148+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6417297157123190451+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_6417297157123190451+e30acd3a/model.neff index ab77dbaabdd5831ec378a0aa62a70163acc64100..7e11a9055218d170bf3a99bfb74fdfe70699738c 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_6417297157123190451+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_6417297157123190451+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_665145310648895012+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_665145310648895012+e30acd3a/model.neff index 89edb7557f27e084a913be3bded1b2c7fb6c640e..1f0719754c75f15bfa16c36e200ab6d72bcef33f 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_665145310648895012+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_665145310648895012+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_6711765444274762085+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_6711765444274762085+e30acd3a/model.neff index 456afdf6106045d211162394eeb1234235817551..fef17e70f7da8ee39b1306e884d88f9aa3c49183 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_6711765444274762085+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_6711765444274762085+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/compile_flags.json new file mode 100644 index 0000000000000000000000000000000000000000..1388ff87c4279ec609dd385cd63c9f4635a2d3b6 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/compile_flags.json @@ -0,0 +1 @@ +["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_339509f3-d775-4039-89d3-3b620bec195a/compiler_workdir/AttentionModelWrapper/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"] \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/model.done new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/model.hlo_module.pb new file mode 100644 index 0000000000000000000000000000000000000000..4bb45842ea0894b69b62a879499055cb6a5cb9ae --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/model.hlo_module.pb @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:be2fc82d01eae877238b23e1fa62d9bd75d648e5ffdf58b8cf01e4f672c3e27e +size 11280 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/model.neff new file mode 100644 index 0000000000000000000000000000000000000000..84cd618db04268864255e93d596f9f33bcb7ebc5 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/model.neff @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:1bbb7987284337180d70ecebedaa76e0e2f80c02a8f5ff6671330df076b4e433 +size 1444864 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/wrapped_neff.hlo new file mode 100644 index 0000000000000000000000000000000000000000..9c0856643ebeba8eb6f7abf8eee6ca37adf49295 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_69bbb3cca16a406b8ff9+14c7f84d/wrapped_neff.hlo @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:d93b0c899b2dba333f3a4bb598a14fe000bbca77b24f4b1502e6c66b975f57e9 +size 1447845 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_7049343467335223052+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_7049343467335223052+e30acd3a/model.neff index bf1eb9c026ccddcec9b565872962df7c0be09b52..9a7172ffa8f3ae38562303326e6fbc9a00593cd3 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_7049343467335223052+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_7049343467335223052+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_7877866862268499659+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_7877866862268499659+e30acd3a/model.neff index 7d9590f96be2325f32e80651ca75e20e57d90882..3be92774ccac3b1bc4b49a36133ed421e34ad760 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_7877866862268499659+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_7877866862268499659+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_793722132614349680+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_793722132614349680+e30acd3a/model.neff index 92c033e0f750ba9a98b09f0c26c021d2d822d02b..05d841f1b25b09a32f1666e567e4a41b2827a451 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_793722132614349680+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_793722132614349680+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8142158350842512240+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8142158350842512240+e30acd3a/model.neff index a4a104829522fae7cff86eb81075246b0ecfc426..bb0ef978ddaa0528888798a13e818d4cdb60ec23 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_8142158350842512240+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_8142158350842512240+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8146958338478347620+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8146958338478347620+e30acd3a/model.neff index 501a92bfd29fc437574c224ca22901dec15f6d5c..7b665bf88ecced570f8cede80a3df88c46a37848 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_8146958338478347620+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_8146958338478347620+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8245165830758578911+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8245165830758578911+e30acd3a/model.neff index 0a85220d688710fa77a6c778fe9ef621e969f10a..43f832c657b29568b470daf0f5224e454c73c767 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_8245165830758578911+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_8245165830758578911+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8679096817551502409+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8679096817551502409+e30acd3a/model.neff index 77509cc42b608b0d1c08b95fe0d50e4b5ce72205..16599a0918a388bfb3f57207d3db2e898775a8a0 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_8679096817551502409+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_8679096817551502409+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8882592820567938515+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8882592820567938515+e30acd3a/model.neff index 807316d6a36c89da801fd2c727c158baf94ed4a7..834fe4c31d70e891f2403ab1f2e9cf1440ba6674 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_8882592820567938515+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_8882592820567938515+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8d5c7639490b29a58f59+24129607/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_8d5c7639490b29a58f59+24129607/model.hlo_module.pb index de9ee8974f31db4bf8364fa450ebbabf690b8fc7..be952da347de317d364c85e97ed45978ec10f0c5 100644 --- a/neuronxcc-2.21.33363.0+82129205/MODULE_8d5c7639490b29a58f59+24129607/model.hlo_module.pb +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_8d5c7639490b29a58f59+24129607/model.hlo_module.pb @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:c321a653df6a7dcb252c745706002499fca22573f43b59d061ba1456342450fb +oid sha256:395529a99f71d2c53ebdea1ddab9d3c20cb88f4a53eef1dc2f557c1d2f43ea83 size 474402 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_8d5c7639490b29a58f59+24129607/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_8d5c7639490b29a58f59+24129607/model.neff index 51c4269e93fb49ca897477268f92796ea192c728..6e06609701ed9fdcc3d1650d8a7173454d4ac62d 100644 --- a/neuronxcc-2.21.33363.0+82129205/MODULE_8d5c7639490b29a58f59+24129607/model.neff +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_8d5c7639490b29a58f59+24129607/model.neff @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:ebe680ec45bccd0e8d1b0997533df67c7812ccd414f658b2ef6f201bdd0b6ed9 +oid sha256:8173d96dfa1076a582abbbcd3343ebdbd95e1f91ba67cea0a532fea4b16ca179 size 42322944 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_9168843780052024308+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_9168843780052024308+e30acd3a/model.neff index 40a2281a02729a75b9941fe321e55dfd037ab1fd..d620f2151433d4b31166135386281f4ba3cebb62 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_9168843780052024308+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_9168843780052024308+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_9527558113976496538+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_9527558113976496538+e30acd3a/model.neff index 86cf15bfe3f05e18e272af19440668a2d71c4c36..37253e852e14ad1aa6debdbea756f497f1f6695c 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_9527558113976496538+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_9527558113976496538+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_9533302140457976101+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_9533302140457976101+e30acd3a/model.neff index 873ca24f6d55c07e195b8d8730a85adf7c82849d..48a70c37a6c7c4a8fa5bdb0c74f2401938b99a7f 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_9533302140457976101+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_9533302140457976101+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_9770544877851564228+e30acd3a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_9770544877851564228+e30acd3a/model.neff index f548586b321964c793f50dee042247fc639824bf..4ca0650de94cc092b5856e953a163fc51c3f2410 100644 Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_9770544877851564228+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_9770544877851564228+e30acd3a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/compile_flags.json new file mode 100644 index 0000000000000000000000000000000000000000..32166f819e91d4ab13030e40d033167b80ae20f3 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/compile_flags.json @@ -0,0 +1 @@ +["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_3289b018-9440-4b02-a8dd-a74acc85d2d1/compiler_workdir/DecoderLayerWrapper/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"] \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/model.done new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/model.hlo_module.pb new file mode 100644 index 0000000000000000000000000000000000000000..0e51cb39a846e4f58cfe3d5ec50ea34e7babca50 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/model.hlo_module.pb @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c233e7013daa344cb368079ec790fbf10f597bc7f8708f8c91ff6684cdc1cd2e +size 29412 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/model.neff new file mode 100644 index 0000000000000000000000000000000000000000..a80accc9f8cf0847284e7c936b5dfd3f3c5c5686 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/model.neff @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:7702edf9e43a9f8f604d410a78edb9d16e83338c25b1caa3c04af21be42b7de4 +size 328704 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/wrapped_neff.hlo new file mode 100644 index 0000000000000000000000000000000000000000..6df3ebdba3a259d8a827d0ce5fa3182dcf3c7fe7 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_a1abdd54de93009efa29+c1228b08/wrapped_neff.hlo @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:4e9c4398de5cce5a7a71ea29ff972fcea0f762d1d62db0363f3c13f5e20299a9 +size 334452 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/compile_flags.json new file mode 100644 index 0000000000000000000000000000000000000000..4d51960ed1f69ea115ee2389595187600fad95cb --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/compile_flags.json @@ -0,0 +1 @@ +["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_8aba31af-bd5b-46c5-b243-aedafa9ecb43/compiler_workdir/AttentionModelWrapper/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"] \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/model.done new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/model.hlo_module.pb new file mode 100644 index 0000000000000000000000000000000000000000..b7f9f80de01bab3afae30f106439a7600609730b --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/model.hlo_module.pb @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:3a4f4cdcd60e648ee1c6f483fd7d775daeec9d14a980ac23a3446e8df4ef38a3 +size 8979 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/model.neff new file mode 100644 index 0000000000000000000000000000000000000000..228390224c4b59bc8270c3adfaf1fed92cef83ca --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/model.neff @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:e27005b9e8f6c5697230b133aefba97797813d77915f86fe3a77675ed85056fd +size 246784 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/wrapped_neff.hlo new file mode 100644 index 0000000000000000000000000000000000000000..dc996f86a3e7efb1effea5b2450da0ae594afd14 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_c7e1998d928d1169964d+2e7ccd62/wrapped_neff.hlo @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:05358effff0afb911a800bb4461f66e32a1f7cf2a9d38e0d7e95efad4bdfd366 +size 249608 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_dbd0cc3151b636557ebf+313a1d1a/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_dbd0cc3151b636557ebf+313a1d1a/compile_flags.json new file mode 100644 index 0000000000000000000000000000000000000000..1f6fbf74df2461fe47ef50658f665cfb2a949620 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_dbd0cc3151b636557ebf+313a1d1a/compile_flags.json @@ -0,0 +1 @@ +["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_5bf2a387-f4e4-44ce-90ff-a13855a92c9b/compiler_workdir/Llama3RotaryEmbedding/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"] \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_dbd0cc3151b636557ebf+313a1d1a/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_dbd0cc3151b636557ebf+313a1d1a/model.done new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_dbd0cc3151b636557ebf+313a1d1a/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_dbd0cc3151b636557ebf+313a1d1a/model.hlo_module.pb new file mode 100644 index 0000000000000000000000000000000000000000..f1e2473c54cffac949ecf31d3b6ec636c290545b --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_dbd0cc3151b636557ebf+313a1d1a/model.hlo_module.pb @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:f4a29d4cf27aec9af90e53324eff9ee69e033679482f4016a9dd3e5416c7c68a +size 1584 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_dbd0cc3151b636557ebf+313a1d1a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_dbd0cc3151b636557ebf+313a1d1a/model.neff new file mode 100644 index 0000000000000000000000000000000000000000..5e8ceeae44006b7d122efdd1d114e38d2a690a1a Binary files /dev/null and b/neuronxcc-2.21.33363.0+82129205/MODULE_dbd0cc3151b636557ebf+313a1d1a/model.neff differ diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/compile_flags.json new file mode 100644 index 0000000000000000000000000000000000000000..956796adb4d31b270327a3775cdc92f903a51f47 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/compile_flags.json @@ -0,0 +1 @@ +["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_c34c282f-e862-47c5-940c-70976032830c/compiler_workdir/NeuronLlamaMLP/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"] \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/model.done new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/model.hlo_module.pb new file mode 100644 index 0000000000000000000000000000000000000000..6da2cba66863fc471486cc2a5da22b16c00d6480 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/model.hlo_module.pb @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:993f99d2d09d34bd152af4c7f96a1e6e1d8788e1cd4aefb845c601d2f4d5fcfb +size 1931 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/model.neff new file mode 100644 index 0000000000000000000000000000000000000000..b96bd2ef8cfe571ae648fb48fee20407e5e5ee67 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/model.neff @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:c06531949b5a19c8411f652891269424ecbfc1796f8065a0687595b7cbac4761 +size 134144 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/wrapped_neff.hlo new file mode 100644 index 0000000000000000000000000000000000000000..4c3bec5ef5d25fc4dfaf06e3139385e047bc96c7 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_dd763f870ed9238ff61a+c724349a/wrapped_neff.hlo @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:547c669a44a4de1d378857addf74309229a12285841e3b49fcb145303a6e193d +size 136222 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/compile_flags.json b/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/compile_flags.json new file mode 100644 index 0000000000000000000000000000000000000000..d07b695d0120241e6176b65b9832b12e5fc47aaf --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/compile_flags.json @@ -0,0 +1 @@ +["--target=trn1", "--enable-saturate-infinity", "--auto-cast=none", "--model-type=transformer", "-O1", "--logfile=/tmp/nxdi_test_553183f3-3490-4996-bdf8-9b4d6a8c5aa2/compiler_workdir/DecoderLayerWrapper/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"] \ No newline at end of file diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/model.done b/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/model.done new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/model.hlo_module.pb new file mode 100644 index 0000000000000000000000000000000000000000..618358c2b96cde9d0fdd8a5363f4e3756584720d --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/model.hlo_module.pb @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:42a756345a0e47699f8c540a039458d52f4072f93c3d432124a626815c5383b0 +size 14480 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/model.neff new file mode 100644 index 0000000000000000000000000000000000000000..a237652bd7434577a9841191364721cf427b3676 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/model.neff @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:75d55c583d29ff7487756f7276ffa77b85b9cff40b5bf7e6567c64ec50a4d212 +size 267264 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/wrapped_neff.hlo new file mode 100644 index 0000000000000000000000000000000000000000..624a3f4c6d60e8127f809986de371fd92b744660 --- /dev/null +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_e153118345ef4f6faf31+9796177f/wrapped_neff.hlo @@ -0,0 +1,3 @@ +version https://git-lfs.github.com/spec/v1 +oid sha256:16d2073e7f4daa8f990a249f8141edbe0de5c80a05888493917e1f8e46657d26 +size 272962 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.hlo_module.pb index 087adad3be38716000a37f13a671422143669eca..db90fa54202e67f9c2b63a9f960524c316db768c 100644 --- a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.hlo_module.pb +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.hlo_module.pb @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:18a82b3beb86fc5ad8db17a90281a3d8d1e499f2379ca89b897d6d0517a715c4 +oid sha256:8ed2e09754ffce7b5957680d04d07b9b18166a7809582694d1e73f3af4b21434 size 590808 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.neff index 56c65e980ac78c3cfe4d329c0649749f9046f871..51d71de841fa2c3fee2e2834ee67b0d30dd76186 100644 --- a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.neff +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/model.neff @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:3dd763eb1440c95d6ed924548677f6aa098cf9f03d7fbd841ee5ee2f313db171 +oid sha256:0497aa1140c0aea952671ec3a5401e8db65a6ef536b5187562bb9a79c7c83960 size 1547264 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/wrapped_neff.hlo index a3bcd11fef4ae6a107368adf7c0c58bda8c59aa4..ac5322ae42aa54d48baf7cad0cbaea40892c2ba0 100644 --- a/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/wrapped_neff.hlo +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_f19ecf2e2bcc3a4b1bd9+a02c3a36/wrapped_neff.hlo @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:c3cf2970355b636ad91f3ea93d41d6b674e11caa9e82c6e09efb68114e2a3986 +oid sha256:592e9d05b3478e1738837ffc93d4f145b16251d3273004c4e14a3315f61c8f05 size 1703455 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.hlo_module.pb b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.hlo_module.pb index 47dc969631220fdb40e5fe10f7bfae7a5454d613..1f4ba21c9f53777d279e086808276ab93c3b0615 100644 --- a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.hlo_module.pb +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.hlo_module.pb @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:018a62e9a6d7b944baff314113a3d60b4bee472c3df99216573823afea25c4f2 +oid sha256:6661ed9cdcff742b06a6c5ab009981de926925468498a831dec14ba213178f65 size 426072 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.neff b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.neff index c198fd4fbb297faefd5e12f6dec3d9acdc7e3a76..de1b56a6738387a6a0b3f53add79d0b45125da3d 100644 --- a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.neff +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/model.neff @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:cbd90399cdd895e7070d27333128b96a4491c8e03bceefee5d66a7eda041f2c5 +oid sha256:95a0f7e69d9a3a7de7c7cac474f7d2542a72f98680a4ea3fa9cf898287287d13 size 2100224 diff --git a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/wrapped_neff.hlo b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/wrapped_neff.hlo index 96bc20b4ac6958d9d68e157c42e3ea34374ca037..d8fe892b6f8831d91c21d76809db3e1dd891190f 100644 --- a/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/wrapped_neff.hlo +++ b/neuronxcc-2.21.33363.0+82129205/MODULE_fde2b1111ad10bd7504e+a02c3a36/wrapped_neff.hlo @@ -1,3 +1,3 @@ version https://git-lfs.github.com/spec/v1 -oid sha256:4de782c3fa4228d83ee0a52883eb3a2d61e39a862d874dd4f99673d1ac67e79a +oid sha256:9d41879fc891b395c30d54c3b4f4db3b537d4732c6513b0cacd03c20ac3dd55f size 2174311