Synchronizing local compiler cache.

.gitattributes

@@ -2784,3 +2784,5 @@ neuronxcc-2.18.121.0+9e31e41a/MODULE_2e229618015e416964c5+613edded/model.neff fi
 neuronxcc-2.18.121.0+9e31e41a/MODULE_4b31f40da9553c6566ed+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.18.121.0+9e31e41a/MODULE_4f54951d7f1d19705651+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.18.121.0+9e31e41a/MODULE_887308291049e2c08d88+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.18.121.0+9e31e41a/MODULE_2db40b5b23a523e16b87+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.18.121.0+9e31e41a/MODULE_967d83f73b8fe30e4d14+613edded/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.18.121.0+9e31e41a/0_REGISTRY/0.2.0.dev7/phi3/microsoft/Phi-3-mini-4k-instruct/12e0ac618ef0e1523616.json

@@ -0,0 +1,55 @@
+{
+  "_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": "microsoft/Phi-3-mini-4k-instruct--configuration_phi3.Phi3Config",
+    "AutoModelForCausalLM": "microsoft/Phi-3-mini-4k-instruct--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": "HloNeuronConfig",
+    "all_reduce_dtype": null,
+    "allow_flash_attention": false,
+    "attention_layout": "HSB",
+    "attn_output_transposed": false,
+    "auto_cast_type": "bf16",
+    "batch_size": 4,
+    "checkpoint_id": "microsoft/Phi-3-mini-4k-instruct",
+    "checkpoint_revision": "0a67737cc96d2554230f90338b163bc6380a2a85",
+    "collectives_layout": "HSB",
+    "continuous_batching": true,
+    "fuse_qkv": true,
+    "group_query_attention": null,
+    "log_softmax_scores": false,
+    "neuronxcc_version": "2.18.121.0+9e31e41a",
+    "optimum_neuron_version": "0.2.0.dev7",
+    "output_all_logits": false,
+    "sequence_length": 4096,
+    "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-2.18.121.0+9e31e41a/MODULE_2db40b5b23a523e16b87+613edded/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]

neuronxcc-2.18.121.0+9e31e41a/MODULE_2db40b5b23a523e16b87+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:43e8d79fc0b4c30b9d62eac049d4746fdb8a322571ce55327784dfadc65c3697
+size 288875

neuronxcc-2.18.121.0+9e31e41a/MODULE_2db40b5b23a523e16b87+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56907c7d20bdea34b2a38b2e7c4de2cf78f452a0bc2925b68417ab15148543d1
+size 6636544

neuronxcc-2.18.121.0+9e31e41a/MODULE_967d83f73b8fe30e4d14+613edded/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn1", "--model-type=transformer", "--auto-cast=none", "--execute-repetition=1"]

neuronxcc-2.18.121.0+9e31e41a/MODULE_967d83f73b8fe30e4d14+613edded/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:317d8f73b199ff24af718b889545a103377e008117878943dfa6f1b0e0bb66e3
+size 338062

neuronxcc-2.18.121.0+9e31e41a/MODULE_967d83f73b8fe30e4d14+613edded/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e4ac2143bee11e6e7f368babadfd3761f59b3cd9e37e5258bdee737a8dfaec5c
+size 14818304