Synchronizing local compiler cache.

.gitattributes

@@ -5070,3 +5070,6 @@ neuronxcc-2.21.18209.0+043b1bf7/MODULE_cae14ab78f54c8b4e73f+ed72d204/model.neff
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_5b7172cc10110a470457+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_9361fb1ac241d0d047f1+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_9361fb1ac241d0d047f1+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_47040a4f0bf48fd29c51+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_6a4730946575bae38107+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_6a4730946575bae38107+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.0.dev0/phi3/microsoft/Phi-3-mini-4k-instruct/0fb89ab603a49af953ba.json

@@ -0,0 +1,62 @@
+{
+  "_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,
+    "enable_bucketing": false,
+    "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": "0.4.0.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_47040a4f0bf48fd29c51+ed72d204/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_model/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.21.18209.0+043b1bf7/MODULE_47040a4f0bf48fd29c51+ed72d204/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:42afec99775e6c84e8b9a7acbb37638809fd8f17f1e966bb6aa8992232740a2a
+size 923430

neuronxcc-2.21.18209.0+043b1bf7/MODULE_47040a4f0bf48fd29c51+ed72d204/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:842fbd53cc1579778273f08a3789d50a2ccddb6e0693eb87bb6d15bee807bde8
+size 5213184

neuronxcc-2.21.18209.0+043b1bf7/MODULE_5b7172cc10110a470457+ed72d204/model.hlo_module.pb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:beca9d97bb0817762268910bf4fd6905bc965bbb31b925f7439c5367afef4ca6
+oid sha256:8e44bfe923734a9e692c83a3f6f8bf4e15fc148249c77d1aef3fde47ff55a5d1
 size 1050679

neuronxcc-2.21.18209.0+043b1bf7/MODULE_5b7172cc10110a470457+ed72d204/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb05cd48def23ceeaca42558677439b509041cd6d8f491f4c42a5f28097ad070
+oid sha256:e28886991d12741447b30efa8893e0977fe2de2c7d53584535b6c26f97bc50df
 size 5827584

neuronxcc-2.21.18209.0+043b1bf7/MODULE_6a4730946575bae38107+a9d440f5/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_model/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]

neuronxcc-2.21.18209.0+043b1bf7/MODULE_6a4730946575bae38107+a9d440f5/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ceb5b7f758ce013d84d6b8f019608f3d1c9fb2d520f2bff26682850b5b969a5
+size 691657

neuronxcc-2.21.18209.0+043b1bf7/MODULE_6a4730946575bae38107+a9d440f5/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a9ec2c1618b220063f382e554a8dedeb5bba4237c24cea8c0332531a7914f769
+size 12411904

neuronxcc-2.21.18209.0+043b1bf7/MODULE_6a4730946575bae38107+a9d440f5/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:76b058d2acbfa23a6a8e6c047aeeaf38880f13930bab33f23dca11f747e79a6d
+size 12559063

neuronxcc-2.21.18209.0+043b1bf7/MODULE_9361fb1ac241d0d047f1+a9d440f5/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3742b6d8468bdafeb2688cd11636fd0bd17d5ca2ca200846f9580dc57a43eed6
+oid sha256:d94268e27a81f1ee6de2f8082852f047b480ebd477a491c886d40b594c26b385
 size 6974464

neuronxcc-2.21.18209.0+043b1bf7/MODULE_9361fb1ac241d0d047f1+a9d440f5/wrapped_neff.hlo

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f4fb76c9ca9ec7d660f35c08ab1ffb356bec8c6c9971e884d30305abeac309cb
+oid sha256:8bc1719e07caf3b44da4a897ec365164231cb566c062d8c7a3a02bef630183db
 size 7158800