Synchronizing local compiler cache.

.gitattributes

@@ -10694,3 +10694,7 @@ neuronxcc-2.19.8089.0+8ab9f450/MODULE_5942e3747ac9276edb6a/vae_encoder/model.neu
 neuronxcc-2.20.9961.0+0acef03a/MODULE_9656817a851d5c315204+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.20.9961.0+0acef03a/MODULE_9656817a851d5c315204+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.20.9961.0+0acef03a/MODULE_af0c4f48ed1b0ccccacc+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.20.9961.0+0acef03a/MODULE_099d383796736750b115+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.20.9961.0+0acef03a/MODULE_099d383796736750b115+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.20.9961.0+0acef03a/MODULE_d26e042484ebeb246753+ae6a382b/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.20.9961.0+0acef03a/MODULE_d8d0987cfdbff410b14e+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.20.9961.0+0acef03a/0_REGISTRY/0.3.0/phi3/microsoft/phi-4/1fde4505c9053358f09d.json

@@ -0,0 +1,73 @@
+{
+  "_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",
+    "async_mode": false,
+    "attn_kernel_enabled": false,
+    "batch_size": 10,
+    "capacity_factor": null,
+    "cc_pipeline_tiling_factor": 2,
+    "checkpoint_id": "microsoft/phi-4",
+    "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61",
+    "continuous_batching": true,
+    "enable_bucketing": false,
+    "ep_degree": 1,
+    "flash_decoding_enabled": false,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "is_chunked_prefill": false,
+    "local_ranks_size": 2,
+    "logical_nc_config": 1,
+    "max_batch_size": 10,
+    "max_context_length": 1000,
+    "max_topk": 256,
+    "mlp_kernel_enabled": false,
+    "mlp_kernel_fuse_residual_add": false,
+    "n_active_tokens": 1000,
+    "neuronxcc_version": "2.20.9961.0+0acef03a",
+    "num_cores_per_group": 1,
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.3.0",
+    "output_logits": false,
+    "padding_side": "right",
+    "pp_degree": 1,
+    "qk_layernorm": false,
+    "qkv_kernel_enabled": false,
+    "rpl_reduce_dtype": "bfloat16",
+    "sequence_length": 1000,
+    "sequence_parallel_enabled": false,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": null,
+    "torch_dtype": "bfloat16",
+    "tp_degree": 2,
+    "vocab_parallel": false
+  },
+  "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-2.20.9961.0+0acef03a/MODULE_099d383796736750b115+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-2.20.9961.0+0acef03a/MODULE_099d383796736750b115+a9d440f5/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cba9181699671cc6f09e652243fb8c74df46fca4ea9a8c5395560174989e01a7
+size 991979

neuronxcc-2.20.9961.0+0acef03a/MODULE_099d383796736750b115+a9d440f5/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:774c77b80c890060b749ace33c239a9556daeb31159cc7ceb752d1ca4f7af8a1
+size 13405184

neuronxcc-2.20.9961.0+0acef03a/MODULE_099d383796736750b115+a9d440f5/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ad80d8df7ca9ed45cdcad6b62a96d6abd5a46cf2c17ad979369d598a5c224bb7
+size 13579939

neuronxcc-2.20.9961.0+0acef03a/MODULE_3012861085522900609+e30acd3a/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn1"]

neuronxcc-2.20.9961.0+0acef03a/MODULE_3012861085522900609+e30acd3a/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d4d46b13c515d974b6da2ab08091addfd6ec06f546203669fb6de9ec7ecdd0c6
+size 1814

neuronxcc-2.20.9961.0+0acef03a/MODULE_d26e042484ebeb246753+ae6a382b/compile_flags.json

@@ -0,0 +1 @@
+"--model-type=transformer -O1 --lnc=1 --internal-hlo2tensorizer-options='--experimental-unsafe-fp8e4m3fn-as-fp8e4m3 --verify-hlo=false' --logfile=/tmp/nxd_model/layout_opt/log-neuron-cc.txt"

neuronxcc-2.20.9961.0+0acef03a/MODULE_d26e042484ebeb246753+ae6a382b/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fcaf0594c0bce1ad8a16470bfaf30565995fda03450d600eddcfcac9a3596b67
+size 173651

neuronxcc-2.20.9961.0+0acef03a/MODULE_d26e042484ebeb246753+ae6a382b/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c6ddcb60bd7449d35135ef2a974499cff2255e666306dc9aba34a7324f87cb0c
+size 12780544

neuronxcc-2.20.9961.0+0acef03a/MODULE_d8d0987cfdbff410b14e+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-2.20.9961.0+0acef03a/MODULE_d8d0987cfdbff410b14e+ed72d204/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:757fe9966c7ed8a07950e145a840bc69ea705e126de21e7bb0cc5253c28f6868
+size 956069

neuronxcc-2.20.9961.0+0acef03a/MODULE_d8d0987cfdbff410b14e+ed72d204/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0ea0e114db2e1c982e3a595f9eebeb8fe46644c7d6ba5f064623f45ed70ef18
+size 2571264