Synchronizing local compiler cache.

.gitattributes

@@ -1867,3 +1867,61 @@ neuronxcc-2.15.143.0+e39249ad/MODULE_63cbd319a3d911724981+39f12043/model.neff fi
 neuronxcc-2.15.143.0+e39249ad/MODULE_8938e8bef94bcc33f561+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.15.143.0+e39249ad/MODULE_733a6bca18aa9a6325aa+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
 neuronxcc-2.15.143.0+e39249ad/MODULE_967d83f73b8fe30e4d14+39f12043/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_008512cdad534106824b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_0d5e2c0e852741a3814c+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_10bd954bb3f712b9cf93+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_17de9f35d8b452d30d64+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_194792ca42c74e6c357a+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_215d5f9f03ff7f796870+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_2e2fefac0330c92cf935+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_38847bfec695e36ba44b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_399d134629eb34a00d43+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_3aa61c04bd8d9ca8722a+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_40c932379328aee8b619+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_41bd479e657e66f398e1+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_45ee942ce275eb459e1d+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_4723142e6e2041791232+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_478acb1b0aae0b56f401+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_48bfe9ceb9631fdca2d4+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_48c6444adcac1b53fda6+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_4ed4eafdd8b57b2a58ba+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_51c384f52e045bea30bd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_52108672c0a9ba2de711+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_5b45c246024901dcd434+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_626306e0d30994029849+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_66e776ad8f8401a7fa1f+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_6c9037784aa88d983971+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_7323a70130bb87727b84+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_7418d879b0b0dbe9d053+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_767b99765e404104c7bb+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_7f54d4d3885e2a479ab4+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_827f458fd58fe6442166+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_838aaca91716f9f756f4+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_88098bf708a0bb5dbef8/text_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_88098bf708a0bb5dbef8/unet/model.neuron filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_88098bf708a0bb5dbef8/vae_decoder/model.neuron filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_88098bf708a0bb5dbef8/vae_encoder/model.neuron filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_8d1befe3d92435e31fdf+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_922ce3591b342750be6b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_949152ff315aeb616749+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_94932be571ca32c0d6c9+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_94df5568a6ee6b8f721b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_95f239593a81a9baed26+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_9dc8af288609e8374339+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_9e3cdf817e494ba86fd4+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_a307a86e0d389d7d8d9e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_a9bcc20735f75ac1f46e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_b2821092d30cdeb76d6a+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_b2aaa317ae645625c9d1+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_b3ac85ee34a5d9e564ba+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_b4a7a4856dcbd900d6ed+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_c0740fe961da69526f92+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_cb16b651ea9d180d5cfd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_d07d2aeb772cbb9c0ccd+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_de8368a717cfd6dfec57+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_e0765cf6df2204e3664e+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_e40769e5a35d3e5dacff+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_eca56defe775a7981c4b+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_ef26a928b1a6003d9b67+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_f65495564d03f1c90e3a+613edded/model.neff filter=lfs diff=lfs merge=lfs -text
+neuronxcc-2.16.372.0+4a9b2326/MODULE_f914d922823d8ceace3c+613edded/model.neff filter=lfs diff=lfs merge=lfs -text

neuronxcc-2.15.143.0+e39249ad/MODULE_1f6cff7b8c7ee3876236+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:762b9be3f390373aa44c767c26c294ccc69e31cece925752f79fc0a0b311847a
+oid sha256:51bf171f94f9804c13f8b2c5e2dfa8418ba7295354b59ec1d137e594fa7c4de2
 size 3124224

neuronxcc-2.15.143.0+e39249ad/MODULE_2eeecff6f903dd1aaf30+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:22e8825862f5e7f5cdb7a35e2b1fe979a72772d77735bc1469210aa4a48ea6cc
+oid sha256:3142a5e8019c33a5ef1b9cc53abb6dff4f8f1d50d0e04d749b04c81e8e78f6ef
 size 5264384

neuronxcc-2.15.143.0+e39249ad/MODULE_2fafc1ff132d5f941aad+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:af2f9e2ad514b97b416bb6aef9e34afe3d9274d4699802b2151053d474f7499a
+oid sha256:8ecf3dd4fd7bfcd84755db1c6e02e0c8636166a781de06fe106d75e2bf45267d
 size 10845184

neuronxcc-2.15.143.0+e39249ad/MODULE_3af7e7ef3791e98d3c6d+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8b9321a15b876218410414c4c07afb5558a3a841fc22a4d9a188f0b52f0c7c29
+oid sha256:335b430e4841d72c9841f63f75d7d0de45ca610c707c09b557a29554761e19c9
 size 4199424

neuronxcc-2.15.143.0+e39249ad/MODULE_46e098601535b4170185+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5a8a33e23fba0fd52355d4dd08b0e117a0714b7eb2ffeb0ff24c28802a15f8e5
+oid sha256:d05301e480562da305f10a43be768d89386d1b48a8c84340c28f232b45e86fa6
 size 3523584

neuronxcc-2.15.143.0+e39249ad/MODULE_4f8465ac364975d34bcb+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ff10460d516079a69929f2c403e056bf9d8d3ae03e57791fd11a12c034d21a99
+oid sha256:90c4f57defdfb2564d5a6e55524dd2bc6096ebb019a82660921be2fbb57bcc5c
 size 3113984

neuronxcc-2.15.143.0+e39249ad/MODULE_640ad25f4ab9e0ce4ac7+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ceb6500e07b7cbc86470939b6463fe217cfb521293360825d64c29540e3bb1a0
+oid sha256:aed8feb1ed1fe46eb30eadb7c354481d90449089639f2116acd11c72c30ad0b2
 size 3339264

neuronxcc-2.15.143.0+e39249ad/MODULE_722d3c4ebc79e2bbd531+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:216e220c2e421cef516863cefb967c0b2e10e7ba9a1aa0cdfaa6aea8d841cd76
+oid sha256:3fbaf06d4ad87a3f53dc034df20eb74f7a3468d54689007ca059c38b29f145a1
 size 3564544

neuronxcc-2.15.143.0+e39249ad/MODULE_b420f7ff165f4f0517c3+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1df8456803038538c572ee1da28196c61c073660662850a007c297025880e40e
+oid sha256:3a873b6981fc26e2572db4d3e64bb2b31d4ef36aeccf87b8c6c8620e6ef8351f
 size 3615744

neuronxcc-2.15.143.0+e39249ad/MODULE_bb86f1f35879e84a7798+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:be01f3371375ea939c51f2a3bcd524d25d14f673cec78d25a36f30630b3b6e91
+oid sha256:f90139cd0d7490783c7308c8f812756f7a897844edfa675eda4a16bdb6968985
 size 24187904

neuronxcc-2.15.143.0+e39249ad/MODULE_f46c84f4be3f40eafec4+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d04a1b35c7cd84b95185c7d1398e5d2fc81201373e091282dd4676246f959eb5
+oid sha256:a44f7412f20757c4f65770c17f04be082d44549d984ce2d9da57001e84042d6c
 size 3800064

neuronxcc-2.15.143.0+e39249ad/MODULE_f63ff654073837167064+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8f16c4f2138ccda56cd07db50cdd9e4dec9f14451b317ef4dccedc2db487972e
+oid sha256:222d03619526335d9f9c03db8bc7234d21a6c33bbfa89ff0ea9d2d66866b0eb7
 size 3206144

neuronxcc-2.15.143.0+e39249ad/MODULE_fcdf5b6858bafa453910+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7a78d2fb1987a08c0801be5e6fc758c74b2506449f75d1d7c59d07ef90f4ff8d
+oid sha256:cdb285f617fca099f1409e397c39eb94b3d01afc3677fdf3a460f37ad834bc4b
 size 4680704

neuronxcc-2.15.143.0+e39249ad/MODULE_ff275b5606db64ceaea0+39f12043/model.neff

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e2889c2e8ac728c5fdee6d1b3b3994c47be4cc2e4ae2c2a186ec3b6c78ef5638
-size 3083264
+oid sha256:46d1f1d8dc3c1709dcb3ae0c6aac64f95a3524aa66a474765c06c6fad367de2c
+size 3073024

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/granite/ibm-granite/granite-3.1-2b-instruct/0211503669c9e68fbbde.json

@@ -0,0 +1 @@
+{"architectures": ["GraniteForCausalLM"], "attention_bias": false, "attention_dropout": 0.1, "attention_multiplier": 0.015625, "bos_token_id": 0, "embedding_multiplier": 12.0, "eos_token_id": 0, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 8192, "logits_scaling": 8.0, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "granite", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "ibm-granite/granite-3.1-2b-instruct", "checkpoint_revision": "374ef54e020a3ce208c65e96d6213922a87d8952", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 40, "num_key_value_heads": 8, "pad_token_id": 0, "residual_multiplier": 0.22, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 5000000.0, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 49155}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Llama-3.1-8B-Instruct/25395edbfe4cb40ed4e9.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 128, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "meta-llama/Llama-3.1-8B-Instruct", "checkpoint_revision": "0e9e39f249a16976918f6564b8830bc894c89659", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Llama-3.2-1B-Instruct/7a87208395f3db5ba7be.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 64, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "meta-llama/Llama-3.2-1B-Instruct", "checkpoint_revision": "9213176726f574b556790deb65791e0c5aa438b6", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 16, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 32.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Llama-3.2-1B-Instruct/8cd73c74d69aa6440798.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 64, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "meta-llama/Llama-3.2-1B-Instruct", "checkpoint_revision": "9213176726f574b556790deb65791e0c5aa438b6", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 16, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 32.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Llama-3.2-1B-Instruct/ebbf84c3779811da8d2c.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 64, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "meta-llama/Llama-3.2-1B-Instruct", "checkpoint_revision": "9213176726f574b556790deb65791e0c5aa438b6", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 16, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 32.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/meta-llama/Meta-Llama-3.1-8B-Instruct/3b6b0065e24ca9d2eeb0.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 128, "hidden_act": "silu", "hidden_size": 4096, "initializer_range": 0.02, "intermediate_size": 14336, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "meta-llama/Meta-Llama-3.1-8B-Instruct", "checkpoint_revision": "0e9e39f249a16976918f6564b8830bc894c89659", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 8, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/llama/princeton-nlp/Sheared-LLaMA-1.3B/8f3a6f200be91cc32a57.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 1, "eos_token_id": 2, "head_dim": 128, "hidden_act": "silu", "hidden_size": 2048, "initializer_range": 0.02, "intermediate_size": 5504, "max_position_embeddings": 4096, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "princeton-nlp/Sheared-LLaMA-1.3B", "checkpoint_revision": "a4b76938edbf571ea7d7d9904861cbdca08809b4", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 16, "num_hidden_layers": 24, "num_key_value_heads": 16, "pad_token_id": 0, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "tie_word_embeddings": false, "torch_dtype": "float32", "use_cache": true, "vocab_size": 32000}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-medium-4k-instruct/ae83aa0fea9a9d0315d5.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {"AutoConfig": "microsoft/Phi-3-medium-4k-instruct--configuration_phi3.Phi3Config", "AutoModelForCausalLM": "microsoft/Phi-3-medium-4k-instruct--modeling_phi3.Phi3ForCausalLM"}, "bos_token_id": 1, "embd_pdrop": 0.0, "eos_token_id": 32000, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 4096, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "microsoft/Phi-3-medium-4k-instruct", "checkpoint_revision": "ae004ae82eb6eddc32906dfacb1d6dfea8f91996", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 4096, "pad_token_id": 32000, "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, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32064}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-mini-4k-instruct/13bb5f11971eace7eeed.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {"AutoConfig": "microsoft/Phi-3-mini-4k-instruct--configuration_phi3.Phi3Config", "AutoModelForCausalLM": "microsoft/Phi-3-mini-4k-instruct--modeling_phi3.Phi3ForCausalLM"}, "bos_token_id": 1, "embd_pdrop": 0.0, "eos_token_id": 32000, "hidden_act": "silu", "hidden_size": 3072, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 4096, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "microsoft/Phi-3-mini-4k-instruct", "checkpoint_revision": "0a67737cc96d2554230f90338b163bc6380a2a85", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "original_max_position_embeddings": 4096, "pad_token_id": 32000, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "sliding_window": 2047, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32064}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-mini-4k-instruct/29904a0c0edb4679ee76.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {"AutoConfig": "microsoft/Phi-3-mini-4k-instruct--configuration_phi3.Phi3Config", "AutoModelForCausalLM": "microsoft/Phi-3-mini-4k-instruct--modeling_phi3.Phi3ForCausalLM"}, "bos_token_id": 1, "embd_pdrop": 0.0, "eos_token_id": 32000, "hidden_act": "silu", "hidden_size": 3072, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 4096, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "microsoft/Phi-3-mini-4k-instruct", "checkpoint_revision": "0a67737cc96d2554230f90338b163bc6380a2a85", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "original_max_position_embeddings": 4096, "pad_token_id": 32000, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "sliding_window": 2047, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32064}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-mini-4k-instruct/9a957125ff99f484a66e.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {"AutoConfig": "microsoft/Phi-3-mini-4k-instruct--configuration_phi3.Phi3Config", "AutoModelForCausalLM": "microsoft/Phi-3-mini-4k-instruct--modeling_phi3.Phi3ForCausalLM"}, "bos_token_id": 1, "embd_pdrop": 0.0, "eos_token_id": 32000, "hidden_act": "silu", "hidden_size": 3072, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 4096, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "microsoft/Phi-3-mini-4k-instruct", "checkpoint_revision": "0a67737cc96d2554230f90338b163bc6380a2a85", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "original_max_position_embeddings": 4096, "pad_token_id": 32000, "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, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32064}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-3-mini-4k-instruct/d7892042a1da5eaa0020.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {"AutoConfig": "microsoft/Phi-3-mini-4k-instruct--configuration_phi3.Phi3Config", "AutoModelForCausalLM": "microsoft/Phi-3-mini-4k-instruct--modeling_phi3.Phi3ForCausalLM"}, "bos_token_id": 1, "embd_pdrop": 0.0, "eos_token_id": 32000, "hidden_act": "silu", "hidden_size": 3072, "initializer_range": 0.02, "intermediate_size": 8192, "max_position_embeddings": 4096, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "microsoft/Phi-3-mini-4k-instruct", "checkpoint_revision": "0a67737cc96d2554230f90338b163bc6380a2a85", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 32, "num_hidden_layers": 32, "num_key_value_heads": 32, "original_max_position_embeddings": 4096, "pad_token_id": 32000, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 10000.0, "sliding_window": 2047, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 32064}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/Phi-4-mini-instruct/6ff6f9ae27117fd24bd7.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {"AutoConfig": "microsoft/Phi-4-mini-instruct--configuration_phi3.Phi3Config", "AutoModelForCausalLM": "microsoft/Phi-4-mini-instruct--modeling_phi3.Phi3ForCausalLM", "AutoTokenizer": "microsoft/Phi-4-mini-instruct--Xenova/gpt-4o"}, "bos_token_id": 199999, "embd_pdrop": 0.0, "eos_token_id": 199999, "full_attn_mod": 1, "hidden_act": "silu", "hidden_size": 3072, "initializer_range": 0.02, "intermediate_size": 8192, "interpolate_factor": 1, "lm_head_bias": false, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 4, "checkpoint_id": "microsoft/Phi-4-mini-instruct", "checkpoint_revision": "4b00ec8714b0cb224e4fb33380cbf0919f177f3e", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 24, "num_hidden_layers": 32, "num_key_value_heads": 8, "original_max_position_embeddings": 4096, "pad_token_id": 199999, "partial_rotary_factor": 0.75, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": {"long_factor": [1, 1.118320672, 1.250641126, 1.398617824, 1.564103225, 1.74916897, 1.956131817, 2.187582649, 2.446418898, 2.735880826, 3.059592084, 3.421605075, 3.826451687, 4.279200023, 4.785517845, 5.351743533, 5.984965424, 6.693110555, 7.485043894, 8.370679318, 9.36110372, 10.4687158, 11.70738129, 13.09260651, 14.64173252, 16.37415215, 18.31155283, 20.47818807, 22.90118105, 25.61086418, 28.64115884, 32.03, 32.1, 32.13, 32.23, 32.6, 32.61, 32.64, 32.66, 32.7, 32.71, 32.93, 32.97, 33.28, 33.49, 33.5, 44.16, 47.77], "short_factor": [1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0], "type": "longrope"}, "rope_theta": 10000.0, "sliding_window": 262144, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 200064}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/42644c76f78cfd9108b4.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "partial_rotary_factor": 1.0, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/452310f61d209c411ec5.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 1, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/6faff6d0347b9b3197ed.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/aefe5e89e89e5ba40a59.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 24, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 10, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/b05887d807c6f53298d2.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 16, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/c488c1faf1a22992a2ea.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 16, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "partial_rotary_factor": 1.0, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/e21879dbd09a21fb7a2e.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 8, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 10, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/microsoft/phi-4/f6232f240b5b98678a51.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100265, "hidden_act": "silu", "hidden_size": 5120, "initializer_range": 0.02, "intermediate_size": 17920, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "bf16", "batch_size": 16, "checkpoint_id": "microsoft/phi-4", "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 12, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 40, "num_hidden_layers": 40, "num_key_value_heads": 10, "original_max_position_embeddings": 16384, "pad_token_id": 100349, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/yujiepan/phi-4-tiny-random/33891c5b973ec2bfb226.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {}, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100257, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 32, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "yujiepan/phi-4-tiny-random", "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 2, "num_hidden_layers": 2, "num_key_value_heads": 1, "original_max_position_embeddings": 16384, "pad_token_id": 100257, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/yujiepan/phi-4-tiny-random/48d7f1c1e412ea84b60e.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {}, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100257, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 32, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "yujiepan/phi-4-tiny-random", "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 2, "num_hidden_layers": 2, "num_key_value_heads": 1, "original_max_position_embeddings": 16384, "pad_token_id": 100257, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/phi3/yujiepan/phi-4-tiny-random/a37d77e276a4a0f5c4e8.json

@@ -0,0 +1 @@
+{"architectures": ["Phi3ForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "auto_map": {}, "bos_token_id": 100257, "embd_pdrop": 0.0, "eos_token_id": 100257, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 32, "max_position_embeddings": 16384, "model_type": "phi3", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "yujiepan/phi-4-tiny-random", "checkpoint_revision": "18a9a1168dc97ac6d128f811925670c275610f5a", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 2, "num_hidden_layers": 2, "num_key_value_heads": 1, "original_max_position_embeddings": 16384, "pad_token_id": 100257, "resid_pdrop": 0.0, "rms_norm_eps": 1e-05, "rope_scaling": null, "rope_theta": 250000, "sliding_window": null, "tie_word_embeddings": false, "torch_dtype": "bfloat16", "use_cache": true, "vocab_size": 100352}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev1/inference/qwen2/Qwen/Qwen2.5-Math-1.5B/a5b11f2dd182bd813171.json

@@ -0,0 +1 @@
+{"architectures": ["Qwen2ForCausalLM"], "attention_dropout": 0.0, "bos_token_id": 151643, "eos_token_id": 151643, "hidden_act": "silu", "hidden_size": 1536, "initializer_range": 0.02, "intermediate_size": 8960, "max_position_embeddings": 4096, "max_window_layers": 21, "model_type": "qwen2", "neuron": {"auto_cast_type": "bf16", "batch_size": 2, "checkpoint_id": "Qwen/Qwen2.5-Math-1.5B", "checkpoint_revision": "4a83ca6e4526a4f2da3aa259ec36c259f66b2ab2", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 8, "sequence_length": 4096, "task": "text-generation"}, "num_attention_heads": 12, "num_hidden_layers": 28, "num_key_value_heads": 2, "rms_norm_eps": 1e-06, "rope_scaling": null, "rope_theta": 10000, "sliding_window": null, "tie_word_embeddings": true, "torch_dtype": "bfloat16", "use_cache": true, "use_mrope": false, "use_sliding_window": false, "vocab_size": 151936}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/058e3ead348125e7808b.json

@@ -0,0 +1 @@
+{"apply_residual_connection_post_layernorm": false, "architectures": ["BloomForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 1, "dtype": "float32", "eos_token_id": 2, "gradient_checkpointing": false, "hidden_dropout": 0.1, "hidden_size": 32, "id2label": {"0": "LABEL_0", "1": "LABEL_1", "2": "LABEL_2"}, "initializer_range": 0.02, "is_decoder": true, "label2id": {"LABEL_0": 0, "LABEL_1": 1, "LABEL_2": 2}, "layer_norm_epsilon": 1e-05, "model_type": "bloom", "n_head": 4, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM", "checkpoint_revision": "92b07e9b7b4f986fa7c54e2ac3b9201b4ba5212e", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 3, "pretraining_tp": 1, "seq_length": 7, "slow_but_exact": true, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 1024}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/6bf9ef24cedd2d181630.json

@@ -0,0 +1 @@
+{"apply_residual_connection_post_layernorm": false, "architectures": ["BloomForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 1, "dtype": "float32", "eos_token_id": 2, "gradient_checkpointing": false, "hidden_dropout": 0.1, "hidden_size": 32, "id2label": {"0": "LABEL_0", "1": "LABEL_1", "2": "LABEL_2"}, "initializer_range": 0.02, "is_decoder": true, "label2id": {"LABEL_0": 0, "LABEL_1": 1, "LABEL_2": 2}, "layer_norm_epsilon": 1e-05, "model_type": "bloom", "n_head": 4, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM", "checkpoint_revision": "92b07e9b7b4f986fa7c54e2ac3b9201b4ba5212e", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 3, "pretraining_tp": 1, "seq_length": 7, "slow_but_exact": true, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 1024}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/bloom/hf-internal-testing/tiny-random-BloomForCausalLM/7a4c62323b876a6ad082.json

@@ -0,0 +1 @@
+{"apply_residual_connection_post_layernorm": false, "architectures": ["BloomForCausalLM"], "attention_dropout": 0.1, "bos_token_id": 1, "dtype": "float32", "eos_token_id": 2, "gradient_checkpointing": false, "hidden_dropout": 0.1, "hidden_size": 32, "id2label": {"0": "LABEL_0", "1": "LABEL_1", "2": "LABEL_2"}, "initializer_range": 0.02, "is_decoder": true, "label2id": {"LABEL_0": 0, "LABEL_1": 1, "LABEL_2": 2}, "layer_norm_epsilon": 1e-05, "model_type": "bloom", "n_head": 4, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-BloomForCausalLM", "checkpoint_revision": "92b07e9b7b4f986fa7c54e2ac3b9201b4ba5212e", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 3, "pretraining_tp": 1, "seq_length": 7, "slow_but_exact": true, "torch_dtype": "float32", "type_vocab_size": 16, "use_cache": true, "vocab_size": 1024}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/gpt2/gpt2/780a42b2f8414c10d661.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "architectures": ["GPT2LMHeadModel"], "attn_pdrop": 0.1, "bos_token_id": 50256, "embd_pdrop": 0.1, "eos_token_id": 50256, "initializer_range": 0.02, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 1024, "n_embd": 768, "n_head": 12, "n_inner": null, "n_layer": 12, "n_positions": 1024, "neuron": {"auto_cast_type": "fp16", "batch_size": 4, "checkpoint_id": "gpt2", "checkpoint_revision": "607a30d783dfa663caf39e06633721c8d4cfcd7e", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 1024, "task": "text-generation"}, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "task_specific_params": {"text-generation": {"do_sample": true, "max_length": 50}}, "use_cache": true, "vocab_size": 50257}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/3033d959070fc853146f.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "attention_probs_dropout_prob": 0.1, "attn_pdrop": 0.1, "bos_token_id": 98, "embd_pdrop": 0.1, "eos_token_id": 98, "gradient_checkpointing": false, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "initializer_range": 0.02, "intermediate_size": 37, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 512, "n_embd": 32, "n_head": 4, "n_inner": null, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "hf-internal-testing/tiny-random-gpt2", "checkpoint_revision": "71034c5d8bde858ff824298bdedc65515b97d2b9", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 98, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "type_vocab_size": 16, "use_cache": true, "vocab_size": 1000}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/46dc72bec418e48dc0b2.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "attention_probs_dropout_prob": 0.1, "attn_pdrop": 0.1, "bos_token_id": 98, "embd_pdrop": 0.1, "eos_token_id": 98, "gradient_checkpointing": false, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "initializer_range": 0.02, "intermediate_size": 37, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 512, "n_embd": 32, "n_head": 4, "n_inner": null, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-gpt2", "checkpoint_revision": "71034c5d8bde858ff824298bdedc65515b97d2b9", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 98, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "type_vocab_size": 16, "use_cache": true, "vocab_size": 1000}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/gpt2/hf-internal-testing/tiny-random-gpt2/fdfe8a25cf9f41629cb9.json

@@ -0,0 +1 @@
+{"activation_function": "gelu_new", "attention_probs_dropout_prob": 0.1, "attn_pdrop": 0.1, "bos_token_id": 98, "embd_pdrop": 0.1, "eos_token_id": 98, "gradient_checkpointing": false, "hidden_act": "gelu", "hidden_dropout_prob": 0.1, "initializer_range": 0.02, "intermediate_size": 37, "layer_norm_epsilon": 1e-05, "model_type": "gpt2", "n_ctx": 512, "n_embd": 32, "n_head": 4, "n_inner": null, "n_layer": 5, "n_positions": 512, "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "hf-internal-testing/tiny-random-gpt2", "checkpoint_revision": "71034c5d8bde858ff824298bdedc65515b97d2b9", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "pad_token_id": 98, "reorder_and_upcast_attn": false, "resid_pdrop": 0.1, "scale_attn_by_inverse_layer_idx": false, "scale_attn_weights": true, "summary_activation": null, "summary_first_dropout": 0.1, "summary_proj_to_labels": true, "summary_type": "cls_index", "summary_use_proj": true, "type_vocab_size": 16, "use_cache": true, "vocab_size": 1000}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/2e8cb42c92e334dfbaa9.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 4, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 64, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 2, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 4, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/5433fffb45752137c8de.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 4, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 64, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp32", "batch_size": 1, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 4, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 128256}

neuronxcc-2.16.372.0+4a9b2326/0_REGISTRY/0.1.0.dev2/inference/llama/llamafactory/tiny-random-Llama-3/8c48d1da1325809763e8.json

@@ -0,0 +1 @@
+{"architectures": ["LlamaForCausalLM"], "attention_bias": false, "attention_dropout": 0.0, "bos_token_id": 128000, "eos_token_id": [128001, 128008, 128009], "head_dim": 4, "hidden_act": "silu", "hidden_size": 16, "initializer_range": 0.02, "intermediate_size": 64, "max_position_embeddings": 131072, "mlp_bias": false, "model_type": "llama", "neuron": {"auto_cast_type": "fp16", "batch_size": 1, "checkpoint_id": "llamafactory/tiny-random-Llama-3", "checkpoint_revision": "bf2a2e3bf199ad2ee96f02a3c00246c608db22a8", "compiler_type": "neuronx-cc", "compiler_version": "2.16.372.0+4a9b2326", "num_cores": 2, "sequence_length": 100, "task": "text-generation"}, "num_attention_heads": 4, "num_hidden_layers": 2, "num_key_value_heads": 4, "pretraining_tp": 1, "rms_norm_eps": 1e-05, "rope_scaling": {"factor": 8.0, "high_freq_factor": 4.0, "low_freq_factor": 1.0, "original_max_position_embeddings": 8192, "rope_type": "llama3"}, "rope_theta": 500000.0, "tie_word_embeddings": false, "torch_dtype": "float16", "use_cache": true, "vocab_size": 128256}