Hugging Face's logo

optimum-internal-testing
/
neuron-testing-cache

Model card Files
xet
 Community
optimum-internal-testing-user commited on
Commit
7076807
·
verified ·
1 Parent(s): efacf4d

Synchronizing local compiler cache.

Browse files
Files changed (11) hide show
  1. neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/0cc25526e2cfc37a8875a3752f33c4d7505d8a07b869d0f3f41915cf6e763b74/7e7aa9f507e35c13dd8f.json +58 -0
  2. neuronxcc-2.21.33363.0+82129205/MODULE_6a21c8e3eeb9125f1a9c+a02c3a36/model.hlo_module.pb +1 -1
  3. neuronxcc-2.21.33363.0+82129205/MODULE_6a21c8e3eeb9125f1a9c+a02c3a36/model.neff +1 -1
  4. neuronxcc-2.21.33363.0+82129205/MODULE_6a21c8e3eeb9125f1a9c+a02c3a36/wrapped_neff.hlo +1 -1
  5. neuronxcc-2.21.33363.0+82129205/MODULE_958d41f716f615f7fb01+24129607/model.hlo_module.pb +1 -1
  6. neuronxcc-2.21.33363.0+82129205/MODULE_958d41f716f615f7fb01+24129607/model.neff +1 -1
  7. neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/model.hlo_module.pb +1 -1
  8. neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/model.neff +1 -1
  9. neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/wrapped_neff.hlo +1 -1
  10. neuronxcc-2.21.33363.0+82129205/MODULE_e45e2c512692918b8ac7+24129607/model.hlo_module.pb +1 -1
  11. neuronxcc-2.21.33363.0+82129205/MODULE_e45e2c512692918b8ac7+24129607/model.neff +1 -1
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev0/0cc25526e2cfc37a8875a3752f33c4d7505d8a07b869d0f3f41915cf6e763b74/7e7aa9f507e35c13dd8f.json ADDED
@@ -0,0 +1,58 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "ibm-granite/granite-3.1-2b-instruct",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "GraniteForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.1,
10
+ "attention_multiplier": 0.015625,
11
+ "dtype": "bfloat16",
12
+ "embedding_multiplier": 12.0,
13
+ "hidden_act": "silu",
14
+ "hidden_size": 2048,
15
+ "initializer_range": 0.02,
16
+ "intermediate_size": 8192,
17
+ "logits_scaling": 8.0,
18
+ "max_position_embeddings": 131072,
19
+ "mlp_bias": false,
20
+ "model_type": "granite",
21
+ "neuron": {
22
+ "_serialized_key": "NxDNeuronConfig",
23
+ "batch_size": 1,
24
+ "capacity_factor": null,
25
+ "checkpoint_id": "ibm-granite/granite-3.1-2b-instruct",
26
+ "checkpoint_revision": "bbc2aed595bd38bd770263dc3ab831db9794441d",
27
+ "continuous_batching": false,
28
+ "ep_degree": 1,
29
+ "fused_qkv": true,
30
+ "glu_mlp": true,
31
+ "local_ranks_size": 2,
32
+ "max_batch_size": 1,
33
+ "max_context_length": 8192,
34
+ "max_topk": 256,
35
+ "n_active_tokens": 8192,
36
+ "neuronxcc_version": "2.21.33363.0+82129205",
37
+ "on_device_sampling": true,
38
+ "optimum_neuron_version": "0.4.6.dev0",
39
+ "output_logits": false,
40
+ "pp_degree": 1,
41
+ "sequence_length": 8192,
42
+ "speculation_length": 0,
43
+ "start_rank_id": 0,
44
+ "target": "trn1",
45
+ "torch_dtype": "bfloat16",
46
+ "tp_degree": 2
47
+ },
48
+ "num_attention_heads": 32,
49
+ "num_hidden_layers": 40,
50
+ "num_key_value_heads": 8,
51
+ "residual_multiplier": 0.22,
52
+ "rms_norm_eps": 1e-05,
53
+ "rope_scaling": null,
54
+ "rope_theta": 5000000.0,
55
+ "tie_word_embeddings": true,
56
+ "use_cache": true,
57
+ "vocab_size": 49155
58
+ }
neuronxcc-2.21.33363.0+82129205/MODULE_6a21c8e3eeb9125f1a9c+a02c3a36/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e4fe1fe7fbe05c357ec6ce7787ba7569de3551fd1b854171acd8c6526152830e
3
  size 569536
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:526e8478dff68672c6091200f03d94fa784d760b6af9495a5b5bc52bf238d9bb
3
  size 569536
neuronxcc-2.21.33363.0+82129205/MODULE_6a21c8e3eeb9125f1a9c+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:fd3b6c11b9fedd9b55e5f3ea8c0819381e224337b1da6ba12066ed97de9925ef
3
  size 1373184
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1cdc4e858377a179e32b2566e433818bdf91ddda5ddfe5151dc0e78101c9066b
3
  size 1373184
neuronxcc-2.21.33363.0+82129205/MODULE_6a21c8e3eeb9125f1a9c+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:26ad5befcb8454d72ca0270d00a14a2e838f1830613f5a504b9b4be66ea75b98
3
  size 1495573
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2ee85311c9e3891aa28d60adf06ff5fbc80d7036133daffe4922823d320286cb
3
  size 1495573
neuronxcc-2.21.33363.0+82129205/MODULE_958d41f716f615f7fb01+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:87f55c3f481969e9042baa2996a436c447062c57a41a9bf5f6043febbde67363
3
  size 532500
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:d87aedd12dd5989e8d84f53a3aecb58df8c31e5adc70a3682b0c33ffc3799e7e
3
  size 532500
neuronxcc-2.21.33363.0+82129205/MODULE_958d41f716f615f7fb01+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0b294c0f43a710998b20991ce8dbe6dc73073d74f0a0232e83fcad31f51b959a
3
  size 29942784
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:fa81b8f7bcb32c5911e3bd7fe5883d54ba8e6da6cf13e02c30b845d91f9d37de
3
  size 29942784
neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:66f6450490b10388dafb6e4fb250f49f37b11224dd2e5c49f481a0976f729106
3
  size 562167
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cedde36e9447943a3efdf4ed22ba8e8377c18dc164c05dc9ec3b2404fd285d6c
3
  size 562167
neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ac673781bbf3e2dbc475872bf420164d20d0771203247969452879b64ad49b1f
3
  size 1793024
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:26b19d16e87edce0eb519472854464ea428c8c6621ce680463c7a0576186d64b
3
  size 1793024
neuronxcc-2.21.33363.0+82129205/MODULE_a155b32467ad91efdb56+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:794b24bed8dc827a4275a8e1acbadd1ec8b9d25d8643b83854e9dfaa4b94f296
3
  size 1915413
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:4364f3e30295be20e1b36a9f8b3877323fbdd44fe0154ef2aa2e7e559a991578
3
  size 1915413
neuronxcc-2.21.33363.0+82129205/MODULE_e45e2c512692918b8ac7+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:709aef187eaab30b99967d754304fe6bf08161b2e4820c8d294053c66f68c8ae
3
  size 571822
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:09002100cbee99f3cf9d355464c3b373418336fa7f26fce3379b075a8ff4dd8b
3
  size 571822
neuronxcc-2.21.33363.0+82129205/MODULE_e45e2c512692918b8ac7+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a0946eccda0ccaa11fd83c9e6b1926be6d653f863251c91dca40ee872de00538
3
  size 9503744
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b94048b1e583bea84cb9e311cfe1f6298e201e617dedcd340fbc12ea64ab735d
3
  size 9503744