Synchronizing local compiler cache.

.gitattributes

@@ -12421,3 +12421,6 @@ neuronxcc-2.21.18209.0+043b1bf7/MODULE_3c7c7b854a8d23a4106c+24129607/model.neff
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_1c5bb09af30545ba981d+24129607/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_a9765a0de4686386ff33+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_a9765a0de4686386ff33+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_73440ea37044d9b55412+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_73440ea37044d9b55412+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_c0d8c793345bb142d034+24129607/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.1.dev0/phi3/microsoft/phi-4/6e07cc5a5be0652bcef2.json

@@ -0,0 +1,57 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "microsoft/phi-4",
+  "_task": "text-generation",
+  "architectures": [
+    "Phi3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "embd_pdrop": 0.0,
+  "hidden_act": "silu",
+  "hidden_size": 5120,
+  "initializer_range": 0.02,
+  "intermediate_size": 17920,
+  "max_position_embeddings": 16384,
+  "model_type": "phi3",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "microsoft/phi-4",
+    "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61",
+    "continuous_batching": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 10,
+    "max_batch_size": 1,
+    "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": "0.4.1.dev0",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 4096,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn1",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 10
+  },
+  "num_attention_heads": 40,
+  "num_hidden_layers": 40,
+  "num_key_value_heads": 10,
+  "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
+}

neuronxcc-2.21.18209.0+043b1bf7/MODULE_73440ea37044d9b55412+a02c3a36/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", "--logfile=/tmp/nxd_model/token_generation/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]

neuronxcc-2.21.18209.0+043b1bf7/MODULE_73440ea37044d9b55412+a02c3a36/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:997fce42d5fb854e9a2272d02f75583315792f09ae6cc56a044ddaebffd39bb6
+size 1002831

neuronxcc-2.21.18209.0+043b1bf7/MODULE_73440ea37044d9b55412+a02c3a36/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9e6816a02ac217c604a5495910797d15250eb970ebafa93f94d1c581e194a7a3
+size 3810304

neuronxcc-2.21.18209.0+043b1bf7/MODULE_73440ea37044d9b55412+a02c3a36/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:683e2c49173019ef4347c4da0bfeaca0d9b927f1fbdbf237ff3a1b7ecc6d3516
+size 3995906

neuronxcc-2.21.18209.0+043b1bf7/MODULE_c0d8c793345bb142d034+24129607/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", "--logfile=/tmp/nxd_model/context_encoding/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.21.18209.0+043b1bf7/MODULE_c0d8c793345bb142d034+24129607/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e5d445d85c34a609bb559fbe3e83b82a3c415c1153f6c421f57c97284a60f801
+size 1141233

neuronxcc-2.21.18209.0+043b1bf7/MODULE_c0d8c793345bb142d034+24129607/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:171e9f330a7a09d6e98dec9e172211c1eb19f191bead6742bad37340c662a5a3
+size 2632704