dacorvo HF Staff commited on
Commit
f8a796c
·
verified ·
1 Parent(s): 7a51042

Synchronizing local compiler cache.

Browse files
neuronxcc-2.21.18209.0+043b1bf7/0_REGISTRY/0.4.2.dev0/phi3/microsoft/phi-4/79e2ff5f1cf2d57f120b.json ADDED
@@ -0,0 +1,57 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "microsoft/phi-4",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Phi3ForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "embd_pdrop": 0.0,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 5120,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 17920,
15
+ "max_position_embeddings": 16384,
16
+ "model_type": "phi3",
17
+ "neuron": {
18
+ "_serialized_key": "NxDNeuronConfig",
19
+ "batch_size": 16,
20
+ "capacity_factor": null,
21
+ "checkpoint_id": "microsoft/phi-4",
22
+ "checkpoint_revision": "187ef0342fff0eb3333be9f00389385e95ef0b61",
23
+ "continuous_batching": true,
24
+ "ep_degree": 1,
25
+ "fused_qkv": true,
26
+ "glu_mlp": true,
27
+ "local_ranks_size": 10,
28
+ "max_batch_size": 16,
29
+ "max_context_length": 4096,
30
+ "max_topk": 256,
31
+ "n_active_tokens": 4096,
32
+ "neuronxcc_version": "2.21.18209.0+043b1bf7",
33
+ "on_device_sampling": true,
34
+ "optimum_neuron_version": "0.4.2.dev0",
35
+ "output_logits": false,
36
+ "pp_degree": 1,
37
+ "sequence_length": 4096,
38
+ "speculation_length": 0,
39
+ "start_rank_id": 0,
40
+ "target": "trn1",
41
+ "torch_dtype": "bfloat16",
42
+ "tp_degree": 10
43
+ },
44
+ "num_attention_heads": 40,
45
+ "num_hidden_layers": 40,
46
+ "num_key_value_heads": 10,
47
+ "original_max_position_embeddings": 16384,
48
+ "partial_rotary_factor": 1.0,
49
+ "resid_pdrop": 0.0,
50
+ "rms_norm_eps": 1e-05,
51
+ "rope_scaling": null,
52
+ "rope_theta": 250000,
53
+ "sliding_window": null,
54
+ "tie_word_embeddings": false,
55
+ "use_cache": true,
56
+ "vocab_size": 100352
57
+ }