Shouldn't CodeLlama 34B have 16K context and rope_theta 1M?
Browse files- config.json +2 -1
config.json
CHANGED
@@ -9,7 +9,7 @@
|
|
9 |
"hidden_size": 8192,
|
10 |
"initializer_range": 0.02,
|
11 |
"intermediate_size": 22016,
|
12 |
-
"max_position_embeddings":
|
13 |
"model_type": "llama",
|
14 |
"num_attention_heads": 64,
|
15 |
"num_hidden_layers": 48,
|
@@ -18,6 +18,7 @@
|
|
18 |
"pretraining_tp": 1,
|
19 |
"rms_norm_eps": 1e-05,
|
20 |
"rope_scaling": null,
|
|
|
21 |
"tie_word_embeddings": false,
|
22 |
"torch_dtype": "float16",
|
23 |
"transformers_version": "4.31.0",
|
|
|
9 |
"hidden_size": 8192,
|
10 |
"initializer_range": 0.02,
|
11 |
"intermediate_size": 22016,
|
12 |
+
"max_position_embeddings": 16384,
|
13 |
"model_type": "llama",
|
14 |
"num_attention_heads": 64,
|
15 |
"num_hidden_layers": 48,
|
|
|
18 |
"pretraining_tp": 1,
|
19 |
"rms_norm_eps": 1e-05,
|
20 |
"rope_scaling": null,
|
21 |
+
"rope_theta": 1000000,
|
22 |
"tie_word_embeddings": false,
|
23 |
"torch_dtype": "float16",
|
24 |
"transformers_version": "4.31.0",
|