dacorvo HF Staff commited on
Commit
960497e
·
verified ·
1 Parent(s): 49de9af

Synchronizing local compiler cache.

Browse files
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev3/cf6b9a360dcf294104671106bae2adbd9fd291823bb60a351883163684073231/68b97f4dfb16b6d56549.json ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "unsloth/Llama-3.2-1B",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "LlamaForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "dtype": "bfloat16",
11
+ "head_dim": 64,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 2048,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 8192,
16
+ "max_position_embeddings": 131072,
17
+ "mlp_bias": false,
18
+ "model_type": "llama",
19
+ "neuron": {
20
+ "_serialized_key": "NxDNeuronConfig",
21
+ "batch_size": 4,
22
+ "capacity_factor": null,
23
+ "checkpoint_id": "unsloth/Llama-3.2-1B",
24
+ "checkpoint_revision": "9535bd9b1d1dea6acafbdc4813b728796aeb28da",
25
+ "continuous_batching": true,
26
+ "ep_degree": 1,
27
+ "fused_qkv": true,
28
+ "glu_mlp": true,
29
+ "local_ranks_size": 2,
30
+ "max_batch_size": 4,
31
+ "max_context_length": 1024,
32
+ "max_topk": 256,
33
+ "n_active_tokens": 1024,
34
+ "neuronxcc_version": "2.21.33363.0+82129205",
35
+ "on_device_sampling": true,
36
+ "optimum_neuron_version": "0.4.6.dev3",
37
+ "output_logits": false,
38
+ "pp_degree": 1,
39
+ "prefill_chunk_size": 0,
40
+ "sequence_length": 1024,
41
+ "speculation_length": 0,
42
+ "start_rank_id": 0,
43
+ "target": "trn1",
44
+ "torch_dtype": "bfloat16",
45
+ "tp_degree": 2
46
+ },
47
+ "num_attention_heads": 32,
48
+ "num_hidden_layers": 16,
49
+ "num_key_value_heads": 8,
50
+ "pretraining_tp": 1,
51
+ "rms_norm_eps": 1e-05,
52
+ "rope_scaling": {
53
+ "factor": 32.0,
54
+ "high_freq_factor": 4.0,
55
+ "low_freq_factor": 1.0,
56
+ "original_max_position_embeddings": 8192,
57
+ "rope_type": "llama3"
58
+ },
59
+ "rope_theta": 500000.0,
60
+ "tie_word_embeddings": true,
61
+ "unsloth_fixed": true,
62
+ "use_cache": true,
63
+ "vocab_size": 128256
64
+ }
neuronxcc-2.21.33363.0+82129205/0_REGISTRY/0.4.6.dev3/llama/unsloth/Llama-3.2-1B/68b97f4dfb16b6d56549.json ADDED
@@ -0,0 +1,64 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_entry_class": "SingleModelCacheEntry",
3
+ "_model_id": "unsloth/Llama-3.2-1B",
4
+ "_task": "text-generation",
5
+ "architectures": [
6
+ "LlamaForCausalLM"
7
+ ],
8
+ "attention_bias": false,
9
+ "attention_dropout": 0.0,
10
+ "dtype": "bfloat16",
11
+ "head_dim": 64,
12
+ "hidden_act": "silu",
13
+ "hidden_size": 2048,
14
+ "initializer_range": 0.02,
15
+ "intermediate_size": 8192,
16
+ "max_position_embeddings": 131072,
17
+ "mlp_bias": false,
18
+ "model_type": "llama",
19
+ "neuron": {
20
+ "_serialized_key": "NxDNeuronConfig",
21
+ "batch_size": 4,
22
+ "capacity_factor": null,
23
+ "checkpoint_id": "unsloth/Llama-3.2-1B",
24
+ "checkpoint_revision": "9535bd9b1d1dea6acafbdc4813b728796aeb28da",
25
+ "continuous_batching": true,
26
+ "ep_degree": 1,
27
+ "fused_qkv": true,
28
+ "glu_mlp": true,
29
+ "local_ranks_size": 2,
30
+ "max_batch_size": 4,
31
+ "max_context_length": 1024,
32
+ "max_topk": 256,
33
+ "n_active_tokens": 1024,
34
+ "neuronxcc_version": "2.21.33363.0+82129205",
35
+ "on_device_sampling": true,
36
+ "optimum_neuron_version": "0.4.6.dev3",
37
+ "output_logits": false,
38
+ "pp_degree": 1,
39
+ "prefill_chunk_size": 0,
40
+ "sequence_length": 1024,
41
+ "speculation_length": 0,
42
+ "start_rank_id": 0,
43
+ "target": "trn1",
44
+ "torch_dtype": "bfloat16",
45
+ "tp_degree": 2
46
+ },
47
+ "num_attention_heads": 32,
48
+ "num_hidden_layers": 16,
49
+ "num_key_value_heads": 8,
50
+ "pretraining_tp": 1,
51
+ "rms_norm_eps": 1e-05,
52
+ "rope_scaling": {
53
+ "factor": 32.0,
54
+ "high_freq_factor": 4.0,
55
+ "low_freq_factor": 1.0,
56
+ "original_max_position_embeddings": 8192,
57
+ "rope_type": "llama3"
58
+ },
59
+ "rope_theta": 500000.0,
60
+ "tie_word_embeddings": true,
61
+ "unsloth_fixed": true,
62
+ "use_cache": true,
63
+ "vocab_size": 128256
64
+ }
neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff CHANGED
Binary files a/neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff and b/neuronxcc-2.21.33363.0+82129205/MODULE_17476209562158013765+e30acd3a/model.neff differ
 
neuronxcc-2.21.33363.0+82129205/MODULE_1759fced9dd9a678973e+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ad0ad7a1feec6937b89d224b1cb1bae1381b494f263c93fef0eccc833d6c219e
3
  size 684041
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:9b12e4bc9a9646f83987e5eb925b237f68f4d7a5569ef4a101e91e22748bfcb6
3
  size 684041
neuronxcc-2.21.33363.0+82129205/MODULE_1759fced9dd9a678973e+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:bd34bff87af44bcd41fca25534c69de8e7bd333cc0e1b96f0d9edfbeb3c53218
3
  size 7732224
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a849a1d1b254aef58958b0a532465d7fd2588408c54ce33e87d87c11fb2871cc
3
  size 7732224
neuronxcc-2.21.33363.0+82129205/MODULE_4ff804c2533da46e613e+24129607/model.hlo_module.pb CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ab43318764f4e2ebf9cf45def42efaf46ba58f3aa003a523c065935381c5bc82
3
  size 1050679
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:868cca689a34e99963ba4838928c3d05b3b7ff0c504770b6e3b057f912c499df
3
  size 1050679
neuronxcc-2.21.33363.0+82129205/MODULE_4ff804c2533da46e613e+24129607/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:419f12fc7910d1ff32926746e0b32c9eab0d8e53e7d603bb9911c269f721a520
3
  size 14050304
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7092c426f768f52ed7be2a5ae675118ff348d21dc203e5a969ad6da48a833fb0
3
  size 14050304
neuronxcc-2.21.33363.0+82129205/MODULE_5742ef8f5910f4e17377+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a8e3c51003d9f9d19d502e94cf45a18ca1c9754c88879af7408814ec86250de8
3
  size 5387264
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:29f3a621f39fe257c700d6bfc0b4b4ab251957d63970b10bb8ecf19501d9f36a
3
  size 5387264
neuronxcc-2.21.33363.0+82129205/MODULE_5742ef8f5910f4e17377+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:d6fe363495284476ec399be469a73a5d809286e8e89b4467c26c464eaf17f50d
3
  size 5571457
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:61c0c9873fdc5c05d11f5911176dcef5b6049e8a5c619cd4e51ed757c641e98e
3
  size 5571457
neuronxcc-2.21.33363.0+82129205/MODULE_88833531cb6c582d4830+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:b37f4f8dce34036bb1d283d34e754f4cda54de003b009d27f9fa7deec6470639
3
  size 3472384
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:6e8768b115e2249ec808a67cd0136f0169269ce5e8186bbae6a46c21db2c0594
3
  size 3472384
neuronxcc-2.21.33363.0+82129205/MODULE_88833531cb6c582d4830+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:1d3ad7c791604367db0b662bf985022088dc090536275ff22418ea43588d9d30
3
  size 3610166
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7bf0374f79f6879ee32742c88e40cd6578de4b103c9b299baea56cb034b5e7ab
3
  size 3610166
neuronxcc-2.21.33363.0+82129205/MODULE_8ff1216675c7144590c1+c4f887dc/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:936f7c19a4ff86454a5b83a26a3fb7b07cdcc62518970b6c78f0bf293cc21420
3
  size 30598144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:c7ff4622dc34bd380f361bf88e7c6e21a704134669270047adc1998da3107993
3
  size 30598144
neuronxcc-2.21.33363.0+82129205/MODULE_8ff1216675c7144590c1+c4f887dc/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:c5009b951afa0dace7b72350a1ffd6ed43bfb3919fd73d44c0de35f45a53cc56
3
  size 30708309
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:22de134cd282af4925855d2ae8c1af6e8c9998d51db17f71bb976d60a34e050f
3
  size 30708309
neuronxcc-2.21.33363.0+82129205/MODULE_ad16674dd3fc5a51c226+a02c3a36/model.neff CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:2b8be74d754d3877ff4c9586c81611124a2989d55ff021b4915372155a27fe1a
3
  size 3073024
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2f7ccb3c563a4c0fdae234892df68dcd6c60cd41674bf983c7a5693709d2af9c
3
  size 3073024
neuronxcc-2.21.33363.0+82129205/MODULE_ad16674dd3fc5a51c226+a02c3a36/wrapped_neff.hlo CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:a1a2dfe758abc52214d97cbae6fd8bfe738893d16164557e687784ac1c1a80a9
3
  size 3147125
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1a6ac7c0556c0e218f9e95cabc999fa5d73cb79aabcc96ccbc57e3c5325e36e9
3
  size 3147125