mpasila/BadVibesNemo-LoRA-12B

Uses this dataset: mpasila/BadVibesV1-16k-context

Details about the dataset:

It is a combination of these datasets (which have been filtered/processed for ShareGPT format and made sure they don't exceed 16k context length based on unsloth/Ministral-3-8B-Base-2512's tokenizer):

• 3216 entries from adamo1139/4chan_archive_ShareGPT_fixed_newlines_unfiltered
• 19962 entries from Fizzarolli/fse-raw-dump
• 11547 entries from R-Arfin/Depression
• 5060 entries from ShiniChien/creepypasta

The data was also combined and shuffled. Total entries: 39785

Prompt format: ChatML (may be messed up by Unsloth atm)

Merged: mpasila/BadVibesNemo-12B

Training params

Trained at 16384 context window in 4-bit.

model = FastLanguageModel.get_peft_model(
    model,
    r = 128, # Choose any number > 0 ! Suggested 8, 16, 32, 64, 128
    target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
                      "gate_proj", "up_proj", "down_proj",],
    lora_alpha = 32,
    lora_dropout = 0, # Supports any, but = 0 is optimized
    bias = "none",    # Supports any, but = "none" is optimized
    # [NEW] "unsloth" uses 30% less VRAM, fits 2x larger batch sizes!
    use_gradient_checkpointing = "unsloth", # True or "unsloth" for very long context
    random_state = 3407,
    use_rslora = False,  # We support rank stabilized LoRA
    loftq_config = None, # And LoftQ
)

from trl import SFTTrainer, SFTConfig
trainer = SFTTrainer(
    model = model,
    tokenizer = tokenizer,
    train_dataset = dataset,
    eval_dataset = None, # Can set up evaluation!
    args = SFTConfig(
        dataset_text_field = "text",
        per_device_train_batch_size = 2,
        gradient_accumulation_steps = 4, # Use GA to mimic batch size!
        warmup_steps = 10,
        num_train_epochs = 1, # Set this for 1 full training run.
        #max_steps = 60,
        learning_rate = 2e-4, # Reduce to 2e-5 for long training runs
        logging_steps = 1,
        optim = "adamw_8bit",
        weight_decay = 0.001,
        lr_scheduler_type = "linear",
        seed = 3407,
        report_to = "none", # Use TrackIO/WandB etc
    ),
)

Uploaded BadVibesNemo-LoRA-12B model

• Developed by: mpasila
• License: apache-2.0
• Finetuned from model : unsloth/mistral-nemo-base-2407-bnb-4bit

This mistral model was trained 2x faster with Unsloth