Synchronizing local compiler cache.

.gitattributes

@@ -5815,3 +5815,5 @@ neuronxcc-2.21.33363.0+82129205/MODULE_fd476e174a39b55eeb18+a02c3a36/model.neff
 neuronxcc-2.21.33363.0+82129205/MODULE_fd476e174a39b55eeb18+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_fff86602d73f8f8ecb89+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.21.33363.0+82129205/MODULE_fff86602d73f8f8ecb89+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_cf7ffe281543e5e4c3d3+a02c3a36/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.21.33363.0+82129205/MODULE_cf7ffe281543e5e4c3d3+a02c3a36/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.4.dev1/phi3/microsoft/Phi-3-mini-4k-instruct/45bac7b83a13be6ba9d0.json

@@ -0,0 +1,63 @@
+{
+  "_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"
+  },
+  "dtype": "bfloat16",
+  "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": "f39ac1d28e925b323eae81227eaba4464caced4e",
+    "continuous_batching": true,
+    "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": "2.21.33363.0+82129205",
+    "on_device_sampling": true,
+    "optimum_neuron_version": "0.4.4.dev1",
+    "output_logits": false,
+    "pp_degree": 1,
+    "sequence_length": 4096,
+    "sequence_parallel_enabled": false,
+    "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-2.21.33363.0+82129205/MODULE_cf7ffe281543e5e4c3d3+a02c3a36/model.hlo_module.pb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:05a4038a70da97389bf754168583601f7fd3c730fe4edc619ed6004e6c5b4c7c
+oid sha256:66355f45b5a564dad737213b1575c3673e50af834b71a571c765842a1e0bf372
 size 700208

neuronxcc-2.21.33363.0+82129205/MODULE_cf7ffe281543e5e4c3d3+a02c3a36/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b79ab853570488a3007f64292decab6f885c357c3e8e2e94f7c6b5fdaf6d1b41
+size 12411904

neuronxcc-2.21.33363.0+82129205/MODULE_cf7ffe281543e5e4c3d3+a02c3a36/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:36206b877cf4645fb5310d17a09b9cc0188b428861c3cac9f14cfeaeb335de96
+size 12558920