dacorvo HF Staff commited on
Commit
29eb987
·
verified ·
1 Parent(s): bc12444

Synchronizing local compiler cache.

Browse files
neuronxcc-2.19.8089.0+8ab9f450/0_REGISTRY/0.3.0.dev5/qwen3/Qwen/Qwen3-0.6B/bf62d8db5ea46ca61ae6.json ADDED
@@ -0,0 +1,72 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "Qwen/Qwen3-0.6B",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "Qwen3ForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "head_dim": 128,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 1024,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 3072,
15
+ "max_position_embeddings": 40960,
16
+ "max_window_layers": 28,
17
+ "model_type": "qwen3",
18
+ "neuron": {
19
+ "_serialized_key": "NxDNeuronConfig",
20
+ "async_mode": false,
21
+ "attn_kernel_enabled": false,
22
+ "batch_size": 4,
23
+ "capacity_factor": null,
24
+ "cc_pipeline_tiling_factor": 2,
25
+ "checkpoint_id": "Qwen/Qwen3-0.6B",
26
+ "checkpoint_revision": "e6de91484c29aa9480d55605af694f39b081c455",
27
+ "continuous_batching": true,
28
+ "enable_bucketing": false,
29
+ "ep_degree": 1,
30
+ "flash_decoding_enabled": false,
31
+ "fused_qkv": true,
32
+ "glu_mlp": true,
33
+ "is_chunked_prefill": false,
34
+ "local_ranks_size": 2,
35
+ "logical_nc_config": 1,
36
+ "max_batch_size": 4,
37
+ "max_context_length": 4096,
38
+ "max_topk": 256,
39
+ "mlp_kernel_enabled": false,
40
+ "mlp_kernel_fuse_residual_add": false,
41
+ "n_active_tokens": 4096,
42
+ "neuronxcc_version": "2.19.8089.0+8ab9f450",
43
+ "num_cores_per_group": 1,
44
+ "on_device_sampling": false,
45
+ "optimum_neuron_version": "0.3.0.dev5",
46
+ "output_logits": false,
47
+ "padding_side": "right",
48
+ "pp_degree": 1,
49
+ "qk_layernorm": false,
50
+ "qkv_kernel_enabled": false,
51
+ "rpl_reduce_dtype": "bfloat16",
52
+ "sequence_length": 4096,
53
+ "sequence_parallel_enabled": false,
54
+ "speculation_length": 0,
55
+ "start_rank_id": 0,
56
+ "target": null,
57
+ "torch_dtype": "bfloat16",
58
+ "tp_degree": 2,
59
+ "vocab_parallel": false
60
+ },
61
+ "num_attention_heads": 16,
62
+ "num_hidden_layers": 28,
63
+ "num_key_value_heads": 8,
64
+ "rms_norm_eps": 1e-06,
65
+ "rope_scaling": null,
66
+ "rope_theta": 1000000,
67
+ "sliding_window": null,
68
+ "tie_word_embeddings": true,
69
+ "use_cache": true,
70
+ "use_sliding_window": false,
71
+ "vocab_size": 151936
72
+ }
neuronxcc-2.19.8089.0+8ab9f450/MODULE_85a3b00cb47dd0b8aebf+ed72d204/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f8832dcb667870a971ece7d6ecd7d5cae3d0a2b3a1535fe5c75bc29f1f38f2ab
3
  size 847776
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6ffeb8e9ece16ac41c4cf88706d49bf0b121145138f40c6107f5f96f3cb6a275
3
  size 847776
neuronxcc-2.19.8089.0+8ab9f450/MODULE_85a3b00cb47dd0b8aebf+ed72d204/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d5633c860cd257846e5f6b6c4ad8732052efd70fa897001179f2dd7fb1fec700
3
  size 27546624
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:134d14dba6aae5338b4dbbc3b448710798a58d8b15010a09d424dfe2f7a911b5
3
  size 27546624
neuronxcc-2.19.8089.0+8ab9f450/MODULE_9430ff6a35fc239154d1+431f5505/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:de5b2309e51ce7f5c750e4c6404d259868b80de1bff2cb6f59a4354d2b681359
3
  size 492544
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:85374d486908f16bb109c5d132c72b094c8eac4bec64d99c92b6f9f2aebaff89
3
  size 492544
neuronxcc-2.19.8089.0+8ab9f450/MODULE_fdd3c2470676e1b82f01+a9d440f5/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e4a56798cfac9a8b1bfebac54ab7bbf0f85083d40e095b49178ff38502beb872
3
  size 2612224
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:965e436898344a1794a7485a6d22c9270a361cc0d1708fc79a009622a4f1024f
3
  size 2612224
neuronxcc-2.19.8089.0+8ab9f450/MODULE_fdd3c2470676e1b82f01+a9d440f5/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:46bf83f89611b59971f2849923789c5f8a7233397b42f6ebbc2373e1cdbbe9ac
3
  size 2742657
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:89b05680539419dadecda1725b6115918e24721e06b15bd7a0cef2d8306d1d09
3
  size 2742657