Hugging Face's logo

0chanly
/
mistral-large-constitution-adapters

Safetensors
Model card Files
xet
 Community
mistral-large-constitution-adapters
File size: 952 Bytes
e6c75f4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
{
  "model_name": "ML-VIR-MIX2000-STACKCTL",
  "model_id": "mistralai/Mistral-Large-Instruct-2411",
  "dataset": "data/virtue_ethics_combined_2000_mistral-large_clean.jsonl",
  "architecture": "1-stage mixed (clean retrain)",
  "train_samples": 1798,
  "eval_samples": 200,
  "lora_r": 64,
  "lora_alpha": 128,
  "lora_targets": [
    "q_proj",
    "v_proj",
    "k_proj",
    "o_proj"
  ],
  "lora_dropout": 0.1,
  "epochs": 3,
  "batch_size": 1,
  "grad_accum": 8,
  "lr": 0.0002,
  "warmup_steps": 50,
  "max_seq_length": 1024,
  "quantization": "4bit_nf4",
  "seed": 42,
  "method": "self-teaching (Mistral Large generates + critiques + revises) \u2014 clean helpful base_cai slot",
  "final_train_loss": 0.1299,
  "final_eval_loss": 0.7978425621986389,
  "best_eval_loss": 0.7340231537818909,
  "lib_versions": {
    "torch": "2.4.1+cu124",
    "cuda": "12.4",
    "transformers": "4.48.3",
    "peft": "0.19.1",
    "bitsandbytes": "0.49.0"
  }
}