Synchronizing local compiler cache.
Browse files- neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev1/gemma3_text/unsloth/gemma-3-270m-it/4ef38fe13fd2c7209a1f.json +81 -0
- neuronxcc-2.21.33363.0+82129205/MODULE_b820473962a48c99faec+a02c3a36/model.hlo_module.pb +1 -1
- neuronxcc-2.21.33363.0+82129205/MODULE_b820473962a48c99faec+a02c3a36/model.neff +1 -1
- neuronxcc-2.21.33363.0+82129205/MODULE_b820473962a48c99faec+a02c3a36/wrapped_neff.hlo +1 -1
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev1/gemma3_text/unsloth/gemma-3-270m-it/4ef38fe13fd2c7209a1f.json
ADDED
|
@@ -0,0 +1,81 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_entry_class": "SingleModelCacheEntry",
|
| 3 |
+
"_model_id": "unsloth/gemma-3-270m-it",
|
| 4 |
+
"_sliding_window_pattern": 6,
|
| 5 |
+
"_task": "text-generation",
|
| 6 |
+
"architectures": [
|
| 7 |
+
"Gemma3ForCausalLM"
|
| 8 |
+
],
|
| 9 |
+
"attention_bias": false,
|
| 10 |
+
"attention_dropout": 0.0,
|
| 11 |
+
"attn_logit_softcapping": null,
|
| 12 |
+
"dtype": "bfloat16",
|
| 13 |
+
"final_logit_softcapping": null,
|
| 14 |
+
"head_dim": 256,
|
| 15 |
+
"hidden_activation": "gelu_pytorch_tanh",
|
| 16 |
+
"hidden_size": 640,
|
| 17 |
+
"initializer_range": 0.02,
|
| 18 |
+
"intermediate_size": 2048,
|
| 19 |
+
"layer_types": [
|
| 20 |
+
"sliding_attention",
|
| 21 |
+
"sliding_attention",
|
| 22 |
+
"sliding_attention",
|
| 23 |
+
"sliding_attention",
|
| 24 |
+
"sliding_attention",
|
| 25 |
+
"full_attention",
|
| 26 |
+
"sliding_attention",
|
| 27 |
+
"sliding_attention",
|
| 28 |
+
"sliding_attention",
|
| 29 |
+
"sliding_attention",
|
| 30 |
+
"sliding_attention",
|
| 31 |
+
"full_attention",
|
| 32 |
+
"sliding_attention",
|
| 33 |
+
"sliding_attention",
|
| 34 |
+
"sliding_attention",
|
| 35 |
+
"sliding_attention",
|
| 36 |
+
"sliding_attention",
|
| 37 |
+
"full_attention"
|
| 38 |
+
],
|
| 39 |
+
"max_position_embeddings": 32768,
|
| 40 |
+
"model_type": "gemma3_text",
|
| 41 |
+
"neuron": {
|
| 42 |
+
"_serialized_key": "NxDNeuronConfig",
|
| 43 |
+
"batch_size": 1,
|
| 44 |
+
"capacity_factor": null,
|
| 45 |
+
"checkpoint_id": "unsloth/gemma-3-270m-it",
|
| 46 |
+
"checkpoint_revision": "23cf460f6bb16954176b3ddcc8d4f250501458a9",
|
| 47 |
+
"continuous_batching": false,
|
| 48 |
+
"ep_degree": 1,
|
| 49 |
+
"fused_qkv": true,
|
| 50 |
+
"glu_mlp": true,
|
| 51 |
+
"local_ranks_size": 2,
|
| 52 |
+
"max_batch_size": 1,
|
| 53 |
+
"max_context_length": 8192,
|
| 54 |
+
"max_topk": 256,
|
| 55 |
+
"n_active_tokens": 8192,
|
| 56 |
+
"neuronxcc_version": "2.21.33363.0+82129205",
|
| 57 |
+
"on_device_sampling": true,
|
| 58 |
+
"optimum_neuron_version": "0.4.6.dev1",
|
| 59 |
+
"output_logits": false,
|
| 60 |
+
"pp_degree": 1,
|
| 61 |
+
"sequence_length": 8192,
|
| 62 |
+
"speculation_length": 0,
|
| 63 |
+
"start_rank_id": 0,
|
| 64 |
+
"target": "trn1",
|
| 65 |
+
"torch_dtype": "bfloat16",
|
| 66 |
+
"tp_degree": 2
|
| 67 |
+
},
|
| 68 |
+
"num_attention_heads": 4,
|
| 69 |
+
"num_hidden_layers": 18,
|
| 70 |
+
"num_key_value_heads": 1,
|
| 71 |
+
"query_pre_attn_scalar": 256,
|
| 72 |
+
"rms_norm_eps": 1e-06,
|
| 73 |
+
"rope_local_base_freq": 10000.0,
|
| 74 |
+
"rope_scaling": null,
|
| 75 |
+
"rope_theta": 1000000.0,
|
| 76 |
+
"sliding_window": 512,
|
| 77 |
+
"unsloth_fixed": true,
|
| 78 |
+
"use_bidirectional_attention": false,
|
| 79 |
+
"use_cache": true,
|
| 80 |
+
"vocab_size": 262144
|
| 81 |
+
}
|
neuronxcc-2.21.33363.0+82129205/MODULE_b820473962a48c99faec+a02c3a36/model.hlo_module.pb
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 586816
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2bbe16f6fcfd6c5f6008ed5129756d3690a94340ee41934144703a6ead39c5c3
|
| 3 |
size 586816
|
neuronxcc-2.21.33363.0+82129205/MODULE_b820473962a48c99faec+a02c3a36/model.neff
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 1680384
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:591ec270b8d26567c824444e8adaaaf5eb68608577353b89da6e96d5a79d5b69
|
| 3 |
size 1680384
|
neuronxcc-2.21.33363.0+82129205/MODULE_b820473962a48c99faec+a02c3a36/wrapped_neff.hlo
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
size 1802773
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:c66a0ac72b6d3dfd479e0ae5f5a1af8dfbeea74215d267042c9227c916b19a85
|
| 3 |
size 1802773
|