Synchronizing local compiler cache.

.gitattributes

@@ -3920,3 +3920,7 @@ neuronxcc-2.19.8089.0+8ab9f450/MODULE_9ddafe9c53dfe7a4a0ca+ca355898/wrapped_neff
 neuronxcc-2.19.8089.0+8ab9f450/MODULE_7aa6e291958ce4b691a4+d859950c/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.19.8089.0+8ab9f450/MODULE_7aa6e291958ce4b691a4+d859950c/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.19.8089.0+8ab9f450/MODULE_b7b5848e8b5f86e50e8c+503f0791/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_734d88d62b55d9271aad+d859950c/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_734d88d62b55d9271aad+d859950c/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_aff1b3d88ed34435f1d7+431f5505/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_de16071d3a627c984e64+503f0791/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.19.8089.0+8ab9f450/0_REGISTRY/0.3.1.dev0/gpt_oss/plop-internal/gpt-oss-20b-trfs-0804/5663ce3c2c64f64092cc.json

@@ -0,0 +1,120 @@
+{
+  "_entry_class": "SingleModelCacheEntry",
+  "_model_id": "plop-internal/gpt-oss-20b-trfs-0804",
+  "_task": "text-generation",
+  "architectures": [
+    "GptOssForCausalLM"
+  ],
+  "attention_bias": true,
+  "attention_dropout": 0.0,
+  "experts_per_token": 4,
+  "head_dim": 64,
+  "hidden_act": "silu",
+  "hidden_size": 2880,
+  "initial_context_length": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 2880,
+  "layer_types": [
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention",
+    "sliding_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 131072,
+  "mlp_bias": false,
+  "model_type": "gpt_oss",
+  "neuron": {
+    "_serialized_key": "NxDNeuronConfig",
+    "async_mode": false,
+    "attn_kernel_enabled": false,
+    "batch_size": 1,
+    "capacity_factor": null,
+    "cc_pipeline_tiling_factor": 2,
+    "checkpoint_id": "plop-internal/gpt-oss-20b-trfs-0804",
+    "checkpoint_revision": "8964acebadbe440e1f70a40383a95d7f9eec20bc",
+    "continuous_batching": false,
+    "enable_bucketing": false,
+    "ep_degree": 1,
+    "flash_decoding_enabled": false,
+    "fused_qkv": false,
+    "glu_mlp": true,
+    "is_chunked_prefill": false,
+    "local_ranks_size": 8,
+    "logical_nc_config": 1,
+    "max_batch_size": 1,
+    "max_context_length": 4096,
+    "max_topk": 256,
+    "mlp_kernel_enabled": false,
+    "mlp_kernel_fuse_residual_add": false,
+    "n_active_tokens": 4096,
+    "neuronxcc_version": "2.19.8089.0+8ab9f450",
+    "num_cores_per_group": 1,
+    "on_device_sampling": false,
+    "optimum_neuron_version": "0.3.1.dev0",
+    "output_logits": false,
+    "padding_side": "right",
+    "pp_degree": 1,
+    "qk_layernorm": false,
+    "qkv_kernel_enabled": false,
+    "rpl_reduce_dtype": "bfloat16",
+    "sequence_length": 4096,
+    "sequence_parallel_enabled": false,
+    "speculation_length": 0,
+    "start_rank_id": 0,
+    "target": null,
+    "torch_dtype": "bfloat16",
+    "tp_degree": 8,
+    "vocab_parallel": false
+  },
+  "num_attention_heads": 64,
+  "num_experts_per_tok": 4,
+  "num_hidden_layers": 1,
+  "num_key_value_heads": 8,
+  "num_local_experts": 32,
+  "output_router_logits": false,
+  "quantization_config": {
+    "modules_to_not_convert": [
+      "model.layers.*.self_attn",
+      "model.layers.*.mlp.router",
+      "model.embed_tokens",
+      "lm_head"
+    ],
+    "quant_method": "mxfp4"
+  },
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": {
+    "beta_fast": 32.0,
+    "beta_slow": 1.0,
+    "factor": 32.0,
+    "original_max_position_embeddings": 4096,
+    "rope_type": "yarn",
+    "truncate": false
+  },
+  "rope_theta": 150000,
+  "router_aux_loss_coef": 0.9,
+  "sliding_window": 128,
+  "swiglu_limit": 7.0,
+  "tie_word_embeddings": false,
+  "use_cache": true,
+  "vocab_size": 201088
+}

neuronxcc-2.19.8089.0+8ab9f450/MODULE_734d88d62b55d9271aad+d859950c/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--enable-mixed-precision-accumulation", "--model-type", "transformer", "-O0", "--tensorizer-options=--enable-ccop-compute-overlap --cc-pipeline-tiling-factor=2", "--auto-cast=none", "--internal-enable-dge-levels", "vector_dynamic_offsets", "--logfile=/tmp/nxd_model/token_generation_model/_tp0_bk0/log-neuron-cc.txt", "--enable-internal-neff-wrapper"]

neuronxcc-2.19.8089.0+8ab9f450/MODULE_734d88d62b55d9271aad+d859950c/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:993d608905679fd8cc0584ab0909aa296a0d59255cfd448cabdefe38746eba4a
+size 49245

neuronxcc-2.19.8089.0+8ab9f450/MODULE_734d88d62b55d9271aad+d859950c/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:15847193b5b8cb700d88f4412361358c3af736ad27b525e3f03d062b5d49ad3f
+size 646144

neuronxcc-2.19.8089.0+8ab9f450/MODULE_734d88d62b55d9271aad+d859950c/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9bf9b5b6e004a0bc52c78cc0e2f8e920e2ab57924a2e8c1124ca65220a2b39ee
+size 653119

neuronxcc-2.19.8089.0+8ab9f450/MODULE_aff1b3d88ed34435f1d7+431f5505/compile_flags.json

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

neuronxcc-2.19.8089.0+8ab9f450/MODULE_aff1b3d88ed34435f1d7+431f5505/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bcdb782afaef94c23bef93d96917de7e7714161d4694f7da03d5322a8f8444c0
+size 7011

neuronxcc-2.19.8089.0+8ab9f450/MODULE_aff1b3d88ed34435f1d7+431f5505/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:99c45fa0baafac8b4b6dc172c5ea1bb7274e8a95317378a1152463c6e094c1a4
+size 144384

neuronxcc-2.19.8089.0+8ab9f450/MODULE_de16071d3a627c984e64+503f0791/compile_flags.json

@@ -0,0 +1 @@
+["--target=trn1", "--enable-saturate-infinity", "--enable-mixed-precision-accumulation", "--model-type", "transformer", "-O0", "--tensorizer-options=--enable-ccop-compute-overlap --cc-pipeline-tiling-factor=2", "--auto-cast=none", "--internal-enable-dge-levels", "vector_dynamic_offsets", "--logfile=/tmp/nxd_model/context_encoding_model/_tp0_bk0/log-neuron-cc.txt"]

neuronxcc-2.19.8089.0+8ab9f450/MODULE_de16071d3a627c984e64+503f0791/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b3d3c536e9816f5383e6376a7e5cbc50acc1acb57e407493b79a471d511a5bdd
+size 17667741

neuronxcc-2.19.8089.0+8ab9f450/MODULE_de16071d3a627c984e64+503f0791/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00a36b0465678a5d8e73f549072c86f8877b9bfedbd42e7577e5802f0adcc9d1
+size 8510464