0x7o commited on
Commit
3c99efa
1 Parent(s): 2f88a56

Correcting from Llama to Mistral

Browse files

Correcting from Llama to Mistral saves memory when training/inference

[Check](https://huggingface.co/Vikhrmodels/Vikhr-7b-0.1/discussions/2)

Files changed (1) hide show
  1. config.json +2 -2
config.json CHANGED
@@ -1,7 +1,7 @@
1
  {
2
  "_name_or_path": "Vikhrmodels/Vikhr-7b-0.1",
3
  "architectures": [
4
- "LlamaForCausalLM"
5
  ],
6
  "attention_bias": false,
7
  "attention_dropout": 0.0,
@@ -12,7 +12,7 @@
12
  "initializer_range": 0.02,
13
  "intermediate_size": 14336,
14
  "max_position_embeddings": 32768,
15
- "model_type": "llama",
16
  "num_attention_heads": 32,
17
  "num_hidden_layers": 32,
18
  "num_key_value_heads": 8,
 
1
  {
2
  "_name_or_path": "Vikhrmodels/Vikhr-7b-0.1",
3
  "architectures": [
4
+ "MistralForCausalLM"
5
  ],
6
  "attention_bias": false,
7
  "attention_dropout": 0.0,
 
12
  "initializer_range": 0.02,
13
  "intermediate_size": 14336,
14
  "max_position_embeddings": 32768,
15
+ "model_type": "mistral",
16
  "num_attention_heads": 32,
17
  "num_hidden_layers": 32,
18
  "num_key_value_heads": 8,