Create config.json
Browse files- config.json +30 -0
config.json
ADDED
|
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"LlamaForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"model_type": "llama",
|
| 6 |
+
"hidden_size": 4096,
|
| 7 |
+
"intermediate_size": 11008,
|
| 8 |
+
"num_hidden_layers": 32,
|
| 9 |
+
"num_attention_heads": 32,
|
| 10 |
+
"max_position_embeddings": 4096,
|
| 11 |
+
"rms_norm_eps": 1e-06,
|
| 12 |
+
"vocab_size": 32000,
|
| 13 |
+
"bos_token_id": 1,
|
| 14 |
+
"eos_token_id": 2,
|
| 15 |
+
"quantization_config": {
|
| 16 |
+
"load_in_4bit": true,
|
| 17 |
+
"bnb_4bit_compute_dtype": "float16",
|
| 18 |
+
"bnb_4bit_quant_type": "nf4",
|
| 19 |
+
"bnb_4bit_use_double_quant": true
|
| 20 |
+
},
|
| 21 |
+
"fine_tuned": true,
|
| 22 |
+
"base_model": "unsloth/llama-2-7b-bnb-4bit",
|
| 23 |
+
"training_args": {
|
| 24 |
+
"num_epochs": 2,
|
| 25 |
+
"learning_rate": 2e-4,
|
| 26 |
+
"max_seq_length": 700,
|
| 27 |
+
"batch_size": 1,
|
| 28 |
+
"gradient_accumulation_steps": 8
|
| 29 |
+
}
|
| 30 |
+
}
|