Edit model card

Johan-7B

Not suitable for interfernce ,experimienting with fine-tuning only

Model Details

QLoRA 4bit Fine-Tuned Version of Mistral-7B using Unsloth

Trying To Be a Friendly Neighbourhood Doctor.

Training Details

LoRA modules=['v_proj', 'down_proj', 'up_proj', 'o_proj', 'q_proj', 'gate_proj', 'k_proj']

  1. Samantha

    -Advice
    
    -Philosiphy
    
    -Therapy
    

    Special Thanks To Cognitive Computations

batch_size=2
learning_rate=1.5e-5,
warmup_steps=5,
max_steps=25, #per_subject
optim = "adamw_8bit",
weight_decay = 1e-3,
lr_scheduler_type = "cosine",
neftune_noise_alpha=1
  1. LinhDuong/chatdoctor-200k
batch_size=2
learning_rate=1.25e-5,
warmup_steps=5,
max_steps=35,
optim = "adamw_8bit",
weight_decay = 1e-3,
lr_scheduler_type = "cosine",
neftune_noise_alpha=1
  1. Disease_remedies

    Created using GPT 3.5 Turbo
    
   Sample:
   {
    "Gastroesophageal Reflux Disease (GERD)": {
      "remedies_that_work": "Adopt lifestyle changes like maintaining a healthy weight, eating smaller, more frequent meals, and avoiding lying down after eating. Elevate the head of your bed to reduce nighttime symptoms. Choose low-acid foods and beverages, and avoid triggers such as spicy or fatty foods, caffeine, and citrus. Stay upright for at least 2-3 hours after meals. Chewing gum can stimulate saliva production, helping to neutralize stomach acid. Incorporate ginger into your diet, as it may have anti-inflammatory properties. Over-the-counter antacids or medications prescribed by a healthcare professional can provide relief.",
      "remedies_that_dont_work": "Avoid overeating and late-night snacks. Smoking and excessive alcohol consumption can exacerbate symptoms. Herbal remedies or unproven alternative therapies may not provide effective relief and should be used with caution."
    }
  }

Training-Params

  epoch=2
  batch_size=2
  learning_rate=2.5e-5,
  warmup_steps=5,
  optim = "adamw_8bit",
  weight_decay = 1e-3,
  lr_scheduler_type = "cosine",
  neftune_noise_alpha=1
Downloads last month
2
Safetensors
Model size
7.24B params
Tensor type
FP16
·

Datasets used to train fhai50032/Johan-7B-v0.1

Collection including fhai50032/Johan-7B-v0.1