instruction fine-tuned for 50 steps on OPUS dataset.
Browse files- config.json +43 -0
config.json
ADDED
|
@@ -0,0 +1,43 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_name_or_path": "CohereForAI/aya-23-8B",
|
| 3 |
+
"architectures": [
|
| 4 |
+
"CohereForCausalLM"
|
| 5 |
+
],
|
| 6 |
+
"attention_bias": false,
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 5,
|
| 9 |
+
"eos_token_id": 255001,
|
| 10 |
+
"hidden_act": "silu",
|
| 11 |
+
"hidden_size": 4096,
|
| 12 |
+
"initializer_range": 0.02,
|
| 13 |
+
"intermediate_size": 14336,
|
| 14 |
+
"layer_norm_eps": 1e-05,
|
| 15 |
+
"logit_scale": 0.0625,
|
| 16 |
+
"max_position_embeddings": 8192,
|
| 17 |
+
"model_type": "cohere",
|
| 18 |
+
"num_attention_heads": 32,
|
| 19 |
+
"num_hidden_layers": 32,
|
| 20 |
+
"num_key_value_heads": 8,
|
| 21 |
+
"pad_token_id": 0,
|
| 22 |
+
"quantization_config": {
|
| 23 |
+
"_load_in_4bit": true,
|
| 24 |
+
"_load_in_8bit": false,
|
| 25 |
+
"bnb_4bit_compute_dtype": "bfloat16",
|
| 26 |
+
"bnb_4bit_quant_storage": "uint8",
|
| 27 |
+
"bnb_4bit_quant_type": "nf4",
|
| 28 |
+
"bnb_4bit_use_double_quant": true,
|
| 29 |
+
"llm_int8_enable_fp32_cpu_offload": false,
|
| 30 |
+
"llm_int8_has_fp16_weight": false,
|
| 31 |
+
"llm_int8_skip_modules": null,
|
| 32 |
+
"llm_int8_threshold": 6.0,
|
| 33 |
+
"load_in_4bit": true,
|
| 34 |
+
"load_in_8bit": false,
|
| 35 |
+
"quant_method": "bitsandbytes"
|
| 36 |
+
},
|
| 37 |
+
"rope_theta": 10000,
|
| 38 |
+
"torch_dtype": "float16",
|
| 39 |
+
"transformers_version": "4.42.4",
|
| 40 |
+
"use_cache": true,
|
| 41 |
+
"use_qk_norm": false,
|
| 42 |
+
"vocab_size": 256000
|
| 43 |
+
}
|