Synchronizing local compiler cache.

.gitattributes

@@ -12334,3 +12334,6 @@ neuronxcc-2.21.18209.0+043b1bf7/MODULE_429a245a2402055c1036+24129607/model.neff
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_249255cd06f89a40b35c+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_249255cd06f89a40b35c+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_652817910035c5995871+24129607/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_5a6f8993b105f04bc738+ac10809c/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_5a6f8993b105f04bc738+ac10809c/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_67cd5c19b53565a4d242+a32116a7/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.1.dev0/llama/TinyLlama/TinyLlama-1.1B-Chat-v1.0/4672d42320811fa3d1b5.json

@@ -0,0 +1,55 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 2048,
+  "initializer_range": 0.02,
+  "intermediate_size": 5632,
+  "max_position_embeddings": 2048,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "TinyLlama/TinyLlama-1.1B-Chat-v1.0",
+    "checkpoint_revision": "fe8a4ea1ffedaf415f4da2f062534de366a451e6",
+    "continuous_batching": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 4,
+    "max_batch_size": 1,
+    "max_context_length": 2048,
+    "max_topk": 256,
+    "n_active_tokens": 2048,
+    "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": 2048,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "trn2",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 4
+  },
+  "num_attention_heads": 32,
+  "num_hidden_layers": 22,
+  "num_key_value_heads": 4,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "use_cache": true,
+  "vocab_size": 32000
+}

neuronxcc-2.21.18209.0+043b1bf7/MODULE_5a6f8993b105f04bc738+ac10809c/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn2", "--auto-cast=none", "--model-type=transformer", "--tensorizer-options=--enable-ccop-compute-overlap --cc-pipeline-tiling-factor=2 --vectorize-strided-dma ", "-O2", "--lnc=2", "--logfile=/tmp/nxd_model/token_generation/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]

neuronxcc-2.21.18209.0+043b1bf7/MODULE_5a6f8993b105f04bc738+ac10809c/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9bfb9d3c95a929a108cc48950a8c621e50f6a6f2f29727b99d9dc8d889595939
+size 1554574

neuronxcc-2.21.18209.0+043b1bf7/MODULE_5a6f8993b105f04bc738+ac10809c/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a3fc1a0b08e875b3c19149875140e4e9df0b65b89e4ee2cb7da1b821810a55c
+size 1803264

neuronxcc-2.21.18209.0+043b1bf7/MODULE_5a6f8993b105f04bc738+ac10809c/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c29b65ec28518e41acd7b2651b06d2e224ae27673b9acce92588dd39a72d27aa
+size 1905870

neuronxcc-2.21.18209.0+043b1bf7/MODULE_67cd5c19b53565a4d242+a32116a7/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn2", "--auto-cast=none", "--model-type=transformer", "--tensorizer-options=--enable-ccop-compute-overlap --cc-pipeline-tiling-factor=2 --vectorize-strided-dma ", "-O2", "--lnc=2", "--logfile=/tmp/nxd_model/context_encoding/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.21.18209.0+043b1bf7/MODULE_67cd5c19b53565a4d242+a32116a7/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21a7ce22ddec32105e6d18eccd1ca4101c4405b4d29aaed5bb4c3ff37b6b38a6
+size 1717439

neuronxcc-2.21.18209.0+043b1bf7/MODULE_67cd5c19b53565a4d242+a32116a7/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99a61d3389a00c8008ccc68c86d45d916c5e0e5fd9409a2d5df0e86035040260
+size 9974784