Synchronizing local compiler cache of Maykeye/TinyLLama-v0

.gitattributes

@@ -12510,3 +12510,6 @@ neuronxcc-2.21.18209.0+043b1bf7/MODULE_8623340492324789537+bad9cf09/model.neff f
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_87301aa5994aa3ec3488+ac10809c/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_87301aa5994aa3ec3488+ac10809c/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.18209.0+043b1bf7/MODULE_d520c8be8d8ce950a101+a32116a7/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_1b80091d649fccb735ad+cb7c900e/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_4cba349d50643364f6f6+dfb30fc8/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.18209.0+043b1bf7/MODULE_4cba349d50643364f6f6+dfb30fc8/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.1/llama/Maykeye/TinyLLama-v0/73cd857f742734a1be3d.json

@@ -0,0 +1,55 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "Maykeye/TinyLLama-v0",
+  "_task": "text-generation",
+  "architectures": [
+    "LlamaForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "head_dim": 4,
+  "hidden_act": "silu",
+  "hidden_size": 64,
+  "initializer_range": 0.02,
+  "intermediate_size": 256,
+  "max_position_embeddings": 2048,
+  "mlp_bias": false,
+  "model_type": "llama",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "batch_size": 1,
+    "capacity_factor": null,
+    "checkpoint_id": "Maykeye/TinyLLama-v0",
+    "checkpoint_revision": "298338802ab94432b917bcce11382aa151aee50f",
+    "continuous_batching": false,
+    "ep_degree": 1,
+    "fused_qkv": true,
+    "glu_mlp": true,
+    "local_ranks_size": 4,
+    "max_batch_size": 1,
+    "max_context_length": 128,
+    "max_topk": 256,
+    "n_active_tokens": 128,
+    "neuronxcc_version": "2.21.18209.0+043b1bf7",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.1",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 128,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": "inf2",
+    "torch_dtype": "bfloat16",
+    "tp_degree": 4
+  },
+  "num_attention_heads": 16,
+  "num_hidden_layers": 8,
+  "num_key_value_heads": 16,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 10000.0,
+  "tie_word_embeddings": false,
+  "use_cache": true,
+  "vocab_size": 32000
+}

neuronxcc-2.21.18209.0+043b1bf7/MODULE_1b80091d649fccb735ad+cb7c900e/compile_flags.json

@@ -0,0 +1 @@
+["--target=inf2", "--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_1b80091d649fccb735ad+cb7c900e/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ada42ca0fd71a68477241e3e4e5d855277af1a5f3a01b6d0945f7d2a2eecff56
+size 194123

neuronxcc-2.21.18209.0+043b1bf7/MODULE_1b80091d649fccb735ad+cb7c900e/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe6006e690a235019221dcf218ff217b2e38ec7f865c0906da5130969b61bea6
+size 236544

neuronxcc-2.21.18209.0+043b1bf7/MODULE_4cba349d50643364f6f6+dfb30fc8/compile_flags.json

@@ -0,0 +1 @@
+["--target=inf2", "--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_4cba349d50643364f6f6+dfb30fc8/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bca408abd6e3e7c0268691f26a449ce7eccb2a3ddb05cddbb423ddcc171b6add
+size 204883

neuronxcc-2.21.18209.0+043b1bf7/MODULE_4cba349d50643364f6f6+dfb30fc8/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ab8940444def49d4c8fc9f30462e92288a74d02d6f6d0093dae516e551b04a63
+size 277504

neuronxcc-2.21.18209.0+043b1bf7/MODULE_4cba349d50643364f6f6+dfb30fc8/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6ba2e67a47157cf88b9dfd646b5f267c0e9570cee98958ceca94da62b4d5c111
+size 306640