Synchronizing local compiler cache.

neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev0/phi3/microsoft/Phi-3-mini-4k-instruct/72f723e3afb332fbe167.json

@@ -0,0 +1,61 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "microsoft/Phi-3-mini-4k-instruct",
+  "_task": "text-generation",
+  "architectures": [
+    "Phi3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "configuration_phi3.Phi3Config",
+    "AutoModelForCausalLM": "modeling_phi3.Phi3ForCausalLM"
+  },
+  "embd_pdrop": 0.0,
+  "hidden_act": "silu",
+  "hidden_size": 3072,
+  "initializer_range": 0.02,
+  "intermediate_size": 8192,
+  "max_position_embeddings": 4096,
+  "model_type": "phi3",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 4,
+    "capacity_factor": null,
+    "checkpoint_id": "microsoft/Phi-3-mini-4k-instruct",
+    "checkpoint_revision": "0a67737cc96d2554230f90338b163bc6380a2a85",
+    "continuous_batching": true,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 2,
+    "max_batch_size": 4,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "n_active_tokens": 4096,
+    "neuronxcc_version": "2.21.18209.0+043b1bf7",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.2.dev0",
+    "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": 32,
+  "num_key_value_heads": 32,
+  "original_max_position_embeddings": 4096,
+  "partial_rotary_factor": 1.0,
+  "resid_pdrop": 0.0,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "sliding_window": 2047,
+  "tie_word_embeddings": false,
+  "use_cache": true,
+  "vocab_size": 32064
+}

neuronxcc-2.21.18209.0+043b1bf7/MODULE_9d35248a93d4142a3cf1+24129607/model.hlo_module.pb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4c3701ec3661417e95b7a0d48488bffafc155a720c62b2885bc15ad31e9a4dc1
+oid sha256:89d95332373996418dd36c71e92d055f25151179af769bdcbf34f30aa3bcacb6
 size 739558

neuronxcc-2.21.18209.0+043b1bf7/MODULE_9d35248a93d4142a3cf1+24129607/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e118545b87bf0dea461da4fa5ecc6540935443dde798e8227db35b0c315f5967
+oid sha256:dee47034ab783eac59be02bccb1aa5ab2b746cf1bd350578fff367704fec7c2c
 size 26133504

neuronxcc-2.21.18209.0+043b1bf7/MODULE_c2b3afa51c57d431a332+a02c3a36/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:92421ee451d0ae6a58b57769a43978826b242921a10cb1886b5e28c980e35afe
+oid sha256:78410670de86fb85b626bbcd51a5cb4825e7176c757a61fdf64a03dafb9915fd
 size 1926144

neuronxcc-2.21.18209.0+043b1bf7/MODULE_c2b3afa51c57d431a332+a02c3a36/wrapped_neff.hlo

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b3946ae0306fc1753558f9ca642ba3114c1774bbc3f5713830256f08eea19c9c
+oid sha256:3146e765c0a8e68312e1844f7d08a458fe6e0e46ad72afb51d3cade89de7d9e9
 size 2082478