optimum-internal-testing-user commited on
Commit
d3949f6
·
verified ·
1 Parent(s): b1f6663

Synchronizing local compiler cache.

Browse files
neuronxcc-2.18.121.0+9e31e41a/0_REGISTRY/0.3.0.dev3/phi3/microsoft/Phi-3-mini-4k-instruct/db5517b1bf390f16f8c2.json ADDED
@@ -0,0 +1,77 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "microsoft/Phi-3-mini-4k-instruct",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Phi3ForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "auto_map": {
11
+ "AutoConfig": "microsoft/Phi-3-mini-4k-instruct--configuration_phi3.Phi3Config",
12
+ "AutoModelForCausalLM": "microsoft/Phi-3-mini-4k-instruct--modeling_phi3.Phi3ForCausalLM"
13
+ },
14
+ "embd_pdrop": 0.0,
15
+ "hidden_act": "silu",
16
+ "hidden_size": 3072,
17
+ "initializer_range": 0.02,
18
+ "intermediate_size": 8192,
19
+ "max_position_embeddings": 4096,
20
+ "model_type": "phi3",
21
+ "neuron": {
22
+ "_serialized_key": "NxDNeuronConfig",
23
+ "async_mode": false,
24
+ "attn_kernel_enabled": false,
25
+ "batch_size": 4,
26
+ "capacity_factor": null,
27
+ "cc_pipeline_tiling_factor": 2,
28
+ "checkpoint_id": "microsoft/Phi-3-mini-4k-instruct",
29
+ "checkpoint_revision": "0a67737cc96d2554230f90338b163bc6380a2a85",
30
+ "continuous_batching": true,
31
+ "enable_bucketing": false,
32
+ "ep_degree": 1,
33
+ "flash_decoding_enabled": false,
34
+ "fused_qkv": true,
35
+ "glu_mlp": true,
36
+ "is_chunked_prefill": false,
37
+ "local_ranks_size": 2,
38
+ "logical_nc_config": 1,
39
+ "max_batch_size": 4,
40
+ "max_context_length": 4096,
41
+ "max_topk": 256,
42
+ "mlp_kernel_enabled": false,
43
+ "mlp_kernel_fuse_residual_add": false,
44
+ "n_active_tokens": 4096,
45
+ "neuronxcc_version": "2.18.121.0+9e31e41a",
46
+ "num_cores_per_group": 1,
47
+ "on_device_sampling": false,
48
+ "optimum_neuron_version": "0.3.0.dev3",
49
+ "output_logits": false,
50
+ "padding_side": "right",
51
+ "pp_degree": 1,
52
+ "qk_layernorm": false,
53
+ "qkv_kernel_enabled": false,
54
+ "rpl_reduce_dtype": "bfloat16",
55
+ "sequence_length": 4096,
56
+ "sequence_parallel_enabled": false,
57
+ "speculation_length": 0,
58
+ "start_rank_id": 0,
59
+ "target": null,
60
+ "torch_dtype": "bfloat16",
61
+ "tp_degree": 2,
62
+ "vocab_parallel": false
63
+ },
64
+ "num_attention_heads": 32,
65
+ "num_hidden_layers": 32,
66
+ "num_key_value_heads": 32,
67
+ "original_max_position_embeddings": 4096,
68
+ "partial_rotary_factor": 1.0,
69
+ "resid_pdrop": 0.0,
70
+ "rms_norm_eps": 1e-05,
71
+ "rope_scaling": null,
72
+ "rope_theta": 10000.0,
73
+ "sliding_window": 2047,
74
+ "tie_word_embeddings": false,
75
+ "use_cache": true,
76
+ "vocab_size": 32064
77
+ }
neuronxcc-2.18.121.0+9e31e41a/MODULE_6ec54d1cb60ff01fc2a1+5be477de/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8aa41113ad8e5d96fc2a524e076a1ba5a09c1311f304175eca6a2984a5a656f6
3
  size 8664064
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:129da04e33caa720203ab681be52afdc21a398c2b2ea8f2524e8b0c1760098e3
3
  size 8664064
neuronxcc-2.18.121.0+9e31e41a/MODULE_6ec54d1cb60ff01fc2a1+5be477de/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2fc962a9d61b4c2ea0f3758227b294bc100b07c62243d30d14818650842dc57b
3
  size 8801315
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b9530594ce0b0e363369210696157d9e75004910eb0095157afa886034de6cf6
3
  size 8801315
neuronxcc-2.18.121.0+9e31e41a/MODULE_ef8771b8447d48519ae4+431f5505/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:21d6fa7a929b019f25647cd5afec98db985b80e41d6e736da384bc6dfade8fef
3
  size 3974144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:52e9e20ee14a50e8fdf217e44003df42d17845af30a61af5250cb5825f35924c
3
  size 3974144
neuronxcc-2.18.121.0+9e31e41a/MODULE_ffcef5c0e4349b0175c7+84f3e719/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:3bca92b24a667c80530089a763f4070157d201b0261a155c36cea611cdd3e3b1
3
  size 841662
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6b92190461c54e01d928e0aabab23b43842b3323d1d4b1d099700e9060877f02
3
  size 841662
neuronxcc-2.18.121.0+9e31e41a/MODULE_ffcef5c0e4349b0175c7+84f3e719/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:17e943bc8d32ee71d5a92739d4d015a09017ca0b976afe93890322650b0422c4
3
  size 6237184
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ac33211a72bfc3472e552004cbd6360a4f17ad9e9f602b66284407424721b7c5
3
  size 6237184