Modify quantization_config to include modules_to_not_convert
Browse files- config.json +10 -2
config.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"_name_or_path": "/home/
|
| 3 |
"architectures": [
|
| 4 |
"LlamaForCausalLM"
|
| 5 |
],
|
|
@@ -21,7 +21,15 @@
|
|
| 21 |
"num_key_value_heads": 8,
|
| 22 |
"pretraining_tp": 1,
|
| 23 |
"quantization_config": {
|
| 24 |
-
"modules_to_not_convert":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 25 |
"quant_method": "bitnet"
|
| 26 |
},
|
| 27 |
"rms_norm_eps": 1e-05,
|
|
|
|
| 1 |
{
|
| 2 |
+
"_name_or_path": "/home/user/Documents/nanotron/models/Llama3-8B-1.58-100B-tokens",
|
| 3 |
"architectures": [
|
| 4 |
"LlamaForCausalLM"
|
| 5 |
],
|
|
|
|
| 21 |
"num_key_value_heads": 8,
|
| 22 |
"pretraining_tp": 1,
|
| 23 |
"quantization_config": {
|
| 24 |
+
"modules_to_not_convert": [
|
| 25 |
+
"self_attn.q_proj",
|
| 26 |
+
"self_attn.k_proj",
|
| 27 |
+
"self_attn.v_proj",
|
| 28 |
+
"self_attn.o_proj",
|
| 29 |
+
"mlp.gate_proj",
|
| 30 |
+
"mlp.up_proj",
|
| 31 |
+
"mlp.down_proj"
|
| 32 |
+
],
|
| 33 |
"quant_method": "bitnet"
|
| 34 |
},
|
| 35 |
"rms_norm_eps": 1e-05,
|