Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files Community
dacorvo HF Staff commited on
Commit
bffcc77
·
verified ·
1 Parent(s): 5ac904c

Synchronizing local compiler cache.

Browse files
Files changed (7) hide show
  1. neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.3.0.dev1/phi3/microsoft/Phi-3-mini-4k-instruct/5b8ae963e4a07386d1f3.json +77 -0
  2. neuronxcc-2.17.194.0+d312836f/MODULE_71d6f1724eb9ca6bbf34+431f5505/model.neff +1 -1
  3. neuronxcc-2.17.194.0+d312836f/MODULE_8d4a4e991c9ba3c051b5+165e9558/model.hlo_module.pb +1 -1
  4. neuronxcc-2.17.194.0+d312836f/MODULE_8d4a4e991c9ba3c051b5+165e9558/model.neff +1 -1
  5. neuronxcc-2.17.194.0+d312836f/MODULE_8d4a4e991c9ba3c051b5+165e9558/wrapped_neff.hlo +1 -1
  6. neuronxcc-2.17.194.0+d312836f/MODULE_ddd8df8ccbf7e2239d19+bfe5714b/model.hlo_module.pb +1 -1
  7. neuronxcc-2.17.194.0+d312836f/MODULE_ddd8df8ccbf7e2239d19+bfe5714b/model.neff +1 -1
neuronxcc-2.17.194.0+d312836f/0_REGISTRY/0.3.0.dev1/phi3/microsoft/Phi-3-mini-4k-instruct/5b8ae963e4a07386d1f3.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": "2.17.194.0+d312836f",
46
+ "num_cores_per_group": 1,
47
+ "on_device_sampling": false,
48
+ "optimum_neuron_version": "0.3.0.dev1",
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-2.17.194.0+d312836f/MODULE_71d6f1724eb9ca6bbf34+431f5505/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e1769f4ed8755d6b56ab33b3ba059b8f67b3f0dff5a098fc5ead75908c8dc238
3
  size 3974144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3e8715e6b2e0ab0cf05ce310d2c592200cdd5b59473ab7526144afa2672036e2
3
  size 3974144
neuronxcc-2.17.194.0+d312836f/MODULE_8d4a4e991c9ba3c051b5+165e9558/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2947ac290d2cd5ee883a3ac6a70100840d60be5d3023a198925ee77c9659a5c9
3
  size 647962
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30d85ea6022cae868c1fc1a3f1d25cf33e3a9691633a11efa31fdb65af4f9b55
3
  size 647962
neuronxcc-2.17.194.0+d312836f/MODULE_8d4a4e991c9ba3c051b5+165e9558/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7ede75e853a41a2fcd7eb69b4808a6532aca457b22ac06d93cdc82d6a6ef02d7
3
  size 10650624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:38d4e4a341b6218a1115413eff52e7b02e100c872230871f19f35b505e7ba152
3
  size 10650624
neuronxcc-2.17.194.0+d312836f/MODULE_8d4a4e991c9ba3c051b5+165e9558/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d7e72eee1aada03be54345608adf36a2a14dd4a09e4ca74b0b4e1f43c48fd340
3
  size 10788002
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3a077f67eb148155c9145a4a1e20d6b158b18b32689048ec19283d37dd03fda3
3
  size 10788002
neuronxcc-2.17.194.0+d312836f/MODULE_ddd8df8ccbf7e2239d19+bfe5714b/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1c9cca486eab4d6045bf1d2ae56c02c9b7cc9027ce7807aec3a99b6ec062e26f
3
  size 835006
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:87b34eae2762cf5b74c8ac51dedbc86cb4b00d76e6f70b35d69fea7be0c2dc0e
3
  size 835006
neuronxcc-2.17.194.0+d312836f/MODULE_ddd8df8ccbf7e2239d19+bfe5714b/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1d80bc8aefb2c260e890a98ae4bd688260b3913b8225b5f70b5787c485c20ff4
3
  size 6216704
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0c5c6587976a01f655486d7be076d69bbf68fe08c4460fdfe0ea9ee5a5f78203
3
  size 6216704