Synchronizing local compiler cache.

.gitattributes

@@ -7801,3 +7801,6 @@ neuronxcc-2.23.6484.0+3b612583/MODULE_d193749859639bfba97b+84337dd9/wrapped_neff
 neuronxcc-2.23.6484.0+3b612583/MODULE_260997bfa4ae712f60da+519f203d/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.23.6484.0+3b612583/MODULE_39a159365f4f855ed959+84337dd9/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.23.6484.0+3b612583/MODULE_39a159365f4f855ed959+84337dd9/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.23.6484.0+3b612583/MODULE_63f1f0ab865197d09266+84337dd9/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.23.6484.0+3b612583/MODULE_63f1f0ab865197d09266+84337dd9/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.23.6484.0+3b612583/MODULE_dbcb7d4c2f26bbf2e79b+519f203d/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.23.6484.0+3b612583/0_REGISTRY/0.4.6.dev3/527899cc3b15a98022089573f6e42212494e0dbda65770e25bddf8fe48c6bd4a/db3a06d2517c123de766.json

@@ -0,0 +1,59 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "ibm-granite/granite-3.1-2b-base",
+  "_task": "text-generation",
+  "architectures": [
+    "GraniteForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.1,
+  "attention_multiplier": 0.015625,
+  "dtype": "bfloat16",
+  "embedding_multiplier": 12.0,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "logits_scaling": 8.0,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "granite",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 4,
+    "capacity_factor": null,
+    "checkpoint_id": "ibm-granite/granite-3.1-2b-base",
+    "checkpoint_revision": "bcf4e33e7debde263041db41869f9cdbeaae3f01",
+    "continuous_batching": true,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 4,
+    "max_context_length": 1024,
+    "max_topk": 256,
+    "n_active_tokens": 1024,
+    "neuronxcc_version": "2.23.6484.0+3b612583",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.6.dev3",
+    "output_logits": false,
+    "pp_degree": 1,
+    "prefill_chunk_size": 0,
+    "sequence_length": 1024,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2
+  },
+  "num_attention_heads": 32,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 8,
+  "residual_multiplier": 0.22,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 5000000.0,
+  "tie_word_embeddings": true,
+  "use_cache": true,
+  "vocab_size": 49152
+}

neuronxcc-2.23.6484.0+3b612583/0_REGISTRY/0.4.6.dev3/granite/ibm-granite/granite-3.1-2b-base/db3a06d2517c123de766.json

@@ -0,0 +1,59 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "ibm-granite/granite-3.1-2b-base",
+  "_task": "text-generation",
+  "architectures": [
+    "GraniteForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.1,
+  "attention_multiplier": 0.015625,
+  "dtype": "bfloat16",
+  "embedding_multiplier": 12.0,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "logits_scaling": 8.0,
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "granite",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 4,
+    "capacity_factor": null,
+    "checkpoint_id": "ibm-granite/granite-3.1-2b-base",
+    "checkpoint_revision": "bcf4e33e7debde263041db41869f9cdbeaae3f01",
+    "continuous_batching": true,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 4,
+    "max_context_length": 1024,
+    "max_topk": 256,
+    "n_active_tokens": 1024,
+    "neuronxcc_version": "2.23.6484.0+3b612583",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.6.dev3",
+    "output_logits": false,
+    "pp_degree": 1,
+    "prefill_chunk_size": 0,
+    "sequence_length": 1024,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2
+  },
+  "num_attention_heads": 32,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 8,
+  "residual_multiplier": 0.22,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 5000000.0,
+  "tie_word_embeddings": true,
+  "use_cache": true,
+  "vocab_size": 49152
+}

neuronxcc-2.23.6484.0+3b612583/MODULE_63f1f0ab865197d09266+84337dd9/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn1", "--auto-cast=none", "--model-type=transformer", "--tensorizer-options=--enable-ccop-compute-overlap --cc-pipeline-tiling-factor=2 --vectorize-strided-dma ", "-O2", "--lnc=1", "--verbose=35", "--logfile=/tmp/nxd_model/token_generation/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]

neuronxcc-2.23.6484.0+3b612583/MODULE_63f1f0ab865197d09266+84337dd9/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ad1ad421a9a4c28b0a5e51c1240a086177965b69b8c09e9ff426b0fceb1c4ee
+size 3831923

neuronxcc-2.23.6484.0+3b612583/MODULE_63f1f0ab865197d09266+84337dd9/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3defc6c19b9951c98aed2e082225ea4e48a966d1e0997201e40913237b57768
+size 5540864

neuronxcc-2.23.6484.0+3b612583/MODULE_63f1f0ab865197d09266+84337dd9/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:342fa3648a05a4f27d33454d8ec86b40397743a6a9e7a8589d9842aa6122ce25
+size 5725663

neuronxcc-2.23.6484.0+3b612583/MODULE_dbcb7d4c2f26bbf2e79b+519f203d/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn1", "--auto-cast=none", "--model-type=transformer", "--tensorizer-options=--enable-ccop-compute-overlap --cc-pipeline-tiling-factor=2 --vectorize-strided-dma ", "-O2", "--lnc=1", "--verbose=35", "--logfile=/tmp/nxd_model/context_encoding/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.23.6484.0+3b612583/MODULE_dbcb7d4c2f26bbf2e79b+519f203d/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a8982e04f1152ff556e8d7410721d10db124474774c9147c4b74cd8e09b59896
+size 3349107

neuronxcc-2.23.6484.0+3b612583/MODULE_dbcb7d4c2f26bbf2e79b+519f203d/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f60f891a65d2d99cd42e60f2e123b2dba36786e4af0b6c9d6cd56104f89fbd16
+size 3922944