Jingya HF Staff commited on
Commit
dd31b30
·
verified ·
1 Parent(s): 872dff7

Upload folder using huggingface_hub

Browse files
Files changed (3) hide show
  1. config.json +37 -0
  2. model.pt +3 -0
  3. neuron_config.json +27 -0
config.json ADDED
@@ -0,0 +1,37 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "architectures": [
3
+ "LlamaForCausalLM"
4
+ ],
5
+ "attention_bias": false,
6
+ "attention_dropout": 0.0,
7
+ "bos_token_id": 128000,
8
+ "dtype": "bfloat16",
9
+ "eos_token_id": 128009,
10
+ "head_dim": 64,
11
+ "hidden_act": "silu",
12
+ "hidden_size": 2048,
13
+ "initializer_range": 0.02,
14
+ "intermediate_size": 8192,
15
+ "max_position_embeddings": 131072,
16
+ "mlp_bias": false,
17
+ "model_type": "llama",
18
+ "num_attention_heads": 32,
19
+ "num_hidden_layers": 16,
20
+ "num_key_value_heads": 8,
21
+ "pad_token_id": 128004,
22
+ "pretraining_tp": 1,
23
+ "rms_norm_eps": 1e-05,
24
+ "rope_scaling": {
25
+ "factor": 32.0,
26
+ "high_freq_factor": 4.0,
27
+ "low_freq_factor": 1.0,
28
+ "original_max_position_embeddings": 8192,
29
+ "rope_type": "llama3"
30
+ },
31
+ "rope_theta": 500000.0,
32
+ "tie_word_embeddings": true,
33
+ "transformers_version": "4.57.1",
34
+ "unsloth_fixed": true,
35
+ "use_cache": true,
36
+ "vocab_size": 128256
37
+ }
model.pt ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:42d7758e106550ec7dcb1fb217d6c06a1c4511d38a95adae3df4421a45b16198
3
+ size 45359109
neuron_config.json ADDED
@@ -0,0 +1,27 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "_serialized_key": "NxDNeuronConfig",
3
+ "batch_size": 1,
4
+ "capacity_factor": null,
5
+ "checkpoint_id": "unsloth/Llama-3.2-1B-Instruct",
6
+ "checkpoint_revision": null,
7
+ "continuous_batching": false,
8
+ "ep_degree": 1,
9
+ "fused_qkv": false,
10
+ "glu_mlp": true,
11
+ "local_ranks_size": 2,
12
+ "max_batch_size": 1,
13
+ "max_context_length": 4096,
14
+ "max_topk": 256,
15
+ "n_active_tokens": 4096,
16
+ "neuronxcc_version": "2.21.18209.0+043b1bf7",
17
+ "on_device_sampling": false,
18
+ "optimum_neuron_version": "0.4.2.dev0",
19
+ "output_logits": false,
20
+ "pp_degree": 1,
21
+ "sequence_length": 4096,
22
+ "speculation_length": 0,
23
+ "start_rank_id": 0,
24
+ "target": "trn1",
25
+ "torch_dtype": "bfloat16",
26
+ "tp_degree": 2
27
+ }