metadata

library_name: transformers
license: apache-2.0
base_model:
  - microsoft/Phi-3-medium-128k-instruct
datasets:
  - flammenai/FlameMix-DPO-v1
  - flammenai/Grill-preprod-v1_chatML
  - flammenai/Grill-preprod-v2_chatML

Exllamav2 quant (exl2 / 3.0 bpw) made with ExLlamaV2 v0.0.21

Other EXL2 quants:

Quant	Model Size	lm_head
2.2	4032 MB	6
2.5	4485 MB	6
3.0	5312 MB	6
3.5	6116 MB	6
3.75	6526 MB	6
4.0	6936 MB	6
4.25	7312 MB	6
5.0	8559 MB	6
6.0	10220 MB	8
6.5	11013 MB	8
8.0	12726 MB	8

Mahou-1.2-phi-14B

Please note: this is an untested, experimental release.

Mahou is our attempt to build a production-ready conversational/roleplay LLM.

Future versions will be released iteratively and finetuned from flammen.ai conversational data.

Chat Format

This model has been trained to use ChatML format.

<|im_start|>system
{{system}}<|im_end|>
<|im_start|>{{char}}
{{message}}<|im_end|>
<|im_start|>{{user}}
{{message}}<|im_end|>

Roleplay Format

Speech without quotes.
Actions in *asterisks*

*leans against wall cooly* so like, i just casted a super strong spell at magician academy today, not gonna lie, felt badass.

ST Settings

Use ChatML for the Context Template.
Turn on Instruct Mode for ChatML.
Use the following stopping strings: ["<", "|", "<|", "\n"]

Method

Finetuned using an A100 on Google Colab.

Fine-tune a Mistral-7b model with Direct Preference Optimization - Maxime Labonne

Configuration

LoRA, model, and training settings:

# LoRA configuration
peft_config = LoraConfig(
    r=16,
    lora_alpha=16,
    lora_dropout=0.05,
    bias="none",
    task_type="CAUSAL_LM",
    target_modules=['k_proj', 'gate_proj', 'v_proj', 'up_proj', 'q_proj', 'o_proj', 'down_proj']
)

# Model to fine-tune
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    load_in_4bit=True
)
model.config.use_cache = False

# Reference model
ref_model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    load_in_4bit=True
)

# Training arguments
training_args = TrainingArguments(
    per_device_train_batch_size=4,
    gradient_accumulation_steps=4,
    gradient_checkpointing=True,
    learning_rate=5e-5,
    lr_scheduler_type="cosine",
    max_steps=2000,
    save_strategy="no",
    logging_steps=1,
    output_dir=new_model,
    optim="paged_adamw_32bit",
    warmup_steps=100,
    bf16=True,
    report_to="wandb",
)

# Create DPO trainer
dpo_trainer = DPOTrainer(
    model,
    ref_model,
    args=training_args,
    train_dataset=dataset,
    tokenizer=tokenizer,
    peft_config=peft_config,
    beta=0.1,
    force_use_ref_model=True
)

# Fine-tune model with DPO
dpo_trainer.train()