Synchronizing local compiler cache.

.gitattributes

@@ -3800,3 +3800,7 @@ neuronxcc-2.19.8089.0+8ab9f450/MODULE_41d38d9f3ba18a42f491+a9d440f5/model.neff f
 neuronxcc-2.19.8089.0+8ab9f450/MODULE_41d38d9f3ba18a42f491+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.19.8089.0+8ab9f450/MODULE_6373bd891205842d16f2+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.19.8089.0+8ab9f450/MODULE_f8e9f8e2ab6f66f576c6+431f5505/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_91501952fa085a11a635+ed72d204/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_efca9e0e049746cedc78+a9d440f5/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_efca9e0e049746cedc78+a9d440f5/wrapped_neff.hlo filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.19.8089.0+8ab9f450/MODULE_f5968e810b1a450b5c85+431f5505/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-0804/985cb556b404b45f3491.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": 16,
+    "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": 16,
+    "vocab_parallel": false
+  },
+  "num_attention_heads": 64,
+  "num_experts_per_tok": 4,
+  "num_hidden_layers": 24,
+  "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_2f1fc60d59c6a26326fa+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-2.19.8089.0+8ab9f450/MODULE_2f1fc60d59c6a26326fa+a9d440f5/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb38c2445a124d9b4a32815a23403b62b93182ab58e6b5b311c6464edd861405
+size 1011298

neuronxcc-2.19.8089.0+8ab9f450/MODULE_91501952fa085a11a635+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-2.19.8089.0+8ab9f450/MODULE_91501952fa085a11a635+ed72d204/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be4c07ce17bd7fb7678003beb379838d70b93a3382284c094bb0bfe611917232
+size 223520398

neuronxcc-2.19.8089.0+8ab9f450/MODULE_91501952fa085a11a635+ed72d204/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68b5ab62f260f87637a47e1cd71363a54a545c6aead2ba430f7b3a2f4dfed907
+size 73309184

neuronxcc-2.19.8089.0+8ab9f450/MODULE_efca9e0e049746cedc78+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-2.19.8089.0+8ab9f450/MODULE_efca9e0e049746cedc78+a9d440f5/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cf37a403330c701b06ed4f2d9ad9c4a6a5038dde96ee1b81985b107e627b0f5
+size 1011250

neuronxcc-2.19.8089.0+8ab9f450/MODULE_efca9e0e049746cedc78+a9d440f5/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbc830e09e0b2e2415e5682c8578825e581f48d9a3e5c8a59089dcfb1ea83c33
+size 1700864

neuronxcc-2.19.8089.0+8ab9f450/MODULE_efca9e0e049746cedc78+a9d440f5/wrapped_neff.hlo

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:232f19a83f3b7c8fc96a39dfcf03ea17349d53b4b3aa32a7f0a0f75e40846e18
+size 1858895

neuronxcc-2.19.8089.0+8ab9f450/MODULE_f5968e810b1a450b5c85+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_f5968e810b1a450b5c85+431f5505/model.hlo_module.pb

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9279690ec8df93b7f9fd54403d74d27a7f594d2ed8144e3cee80e4fdccbf404b
+size 167956

neuronxcc-2.19.8089.0+8ab9f450/MODULE_f5968e810b1a450b5c85+431f5505/model.neff

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fd3fad5bbb78249b31638afede862e58a7598eb5623c5b2f37fe2db429d5de1
+size 769024