Upload LlamaForCausalLM
Browse files- config.json +9 -2
- generation_config.json +3 -1
config.json
CHANGED
|
@@ -22,6 +22,7 @@
|
|
| 22 |
"num_attention_heads": 24,
|
| 23 |
"num_hidden_layers": 28,
|
| 24 |
"num_key_value_heads": 8,
|
|
|
|
| 25 |
"pretraining_tp": 1,
|
| 26 |
"quantization_config": {
|
| 27 |
"_load_in_4bit": true,
|
|
@@ -32,7 +33,12 @@
|
|
| 32 |
"bnb_4bit_use_double_quant": true,
|
| 33 |
"llm_int8_enable_fp32_cpu_offload": false,
|
| 34 |
"llm_int8_has_fp16_weight": false,
|
| 35 |
-
"llm_int8_skip_modules":
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 36 |
"llm_int8_threshold": 6.0,
|
| 37 |
"load_in_4bit": true,
|
| 38 |
"load_in_8bit": false,
|
|
@@ -49,7 +55,8 @@
|
|
| 49 |
"rope_theta": 500000.0,
|
| 50 |
"tie_word_embeddings": true,
|
| 51 |
"torch_dtype": "bfloat16",
|
| 52 |
-
"transformers_version": "4.
|
|
|
|
| 53 |
"use_cache": true,
|
| 54 |
"vocab_size": 128256
|
| 55 |
}
|
|
|
|
| 22 |
"num_attention_heads": 24,
|
| 23 |
"num_hidden_layers": 28,
|
| 24 |
"num_key_value_heads": 8,
|
| 25 |
+
"pad_token_id": 128004,
|
| 26 |
"pretraining_tp": 1,
|
| 27 |
"quantization_config": {
|
| 28 |
"_load_in_4bit": true,
|
|
|
|
| 33 |
"bnb_4bit_use_double_quant": true,
|
| 34 |
"llm_int8_enable_fp32_cpu_offload": false,
|
| 35 |
"llm_int8_has_fp16_weight": false,
|
| 36 |
+
"llm_int8_skip_modules": [
|
| 37 |
+
"lm_head",
|
| 38 |
+
"multi_modal_projector",
|
| 39 |
+
"merger",
|
| 40 |
+
"modality_projection"
|
| 41 |
+
],
|
| 42 |
"llm_int8_threshold": 6.0,
|
| 43 |
"load_in_4bit": true,
|
| 44 |
"load_in_8bit": false,
|
|
|
|
| 55 |
"rope_theta": 500000.0,
|
| 56 |
"tie_word_embeddings": true,
|
| 57 |
"torch_dtype": "bfloat16",
|
| 58 |
+
"transformers_version": "4.48.0",
|
| 59 |
+
"unsloth_fixed": true,
|
| 60 |
"use_cache": true,
|
| 61 |
"vocab_size": 128256
|
| 62 |
}
|
generation_config.json
CHANGED
|
@@ -6,7 +6,9 @@
|
|
| 6 |
128008,
|
| 7 |
128009
|
| 8 |
],
|
|
|
|
|
|
|
| 9 |
"temperature": 0.6,
|
| 10 |
"top_p": 0.9,
|
| 11 |
-
"transformers_version": "4.
|
| 12 |
}
|
|
|
|
| 6 |
128008,
|
| 7 |
128009
|
| 8 |
],
|
| 9 |
+
"max_length": 131072,
|
| 10 |
+
"pad_token_id": 128004,
|
| 11 |
"temperature": 0.6,
|
| 12 |
"top_p": 0.9,
|
| 13 |
+
"transformers_version": "4.48.0"
|
| 14 |
}
|