Modelpublic

RichardErkhov/flammenai_-_flammen15X-mistral-7B-gguf

sourceHugging Faceupdated 2y agoView on Hugging Face

0likes4.3kdownloads

Model Card

Quantization made by Richard Erkhov.

Github

Discord

Request more models

flammen15X-mistral-7B - GGUF

—Model creator: https://huggingface.co/flammenai/
—Original model: https://huggingface.co/flammenai/flammen15X-mistral-7B/

Name	Quant method	Size
flammen15X-mistral-7B.Q2_K.gguf	Q2_K	2.53GB
flammen15X-mistral-7B.IQ3_XS.gguf	IQ3_XS	2.81GB
flammen15X-mistral-7B.IQ3_S.gguf	IQ3_S	2.96GB
flammen15X-mistral-7B.Q3_K_S.gguf	Q3KS	2.95GB
flammen15X-mistral-7B.IQ3_M.gguf	IQ3_M	3.06GB
flammen15X-mistral-7B.Q3_K.gguf	Q3_K	3.28GB
flammen15X-mistral-7B.Q3_K_M.gguf	Q3KM	3.28GB
flammen15X-mistral-7B.Q3_K_L.gguf	Q3KL	3.56GB
flammen15X-mistral-7B.IQ4_XS.gguf	IQ4_XS	3.67GB
flammen15X-mistral-7B.Q4_0.gguf	Q4_0	3.83GB
flammen15X-mistral-7B.IQ4_NL.gguf	IQ4_NL	3.87GB
flammen15X-mistral-7B.Q4_K_S.gguf	Q4KS	3.86GB
flammen15X-mistral-7B.Q4_K.gguf	Q4_K	4.07GB
flammen15X-mistral-7B.Q4_K_M.gguf	Q4KM	4.07GB
flammen15X-mistral-7B.Q4_1.gguf	Q4_1	4.24GB
flammen15X-mistral-7B.Q5_0.gguf	Q5_0	4.65GB
flammen15X-mistral-7B.Q5_K_S.gguf	Q5KS	4.65GB
flammen15X-mistral-7B.Q5_K.gguf	Q5_K	4.78GB
flammen15X-mistral-7B.Q5_K_M.gguf	Q5KM	4.78GB
flammen15X-mistral-7B.Q5_1.gguf	Q5_1	5.07GB
flammen15X-mistral-7B.Q6_K.gguf	Q6_K	5.53GB
flammen15X-mistral-7B.Q8_0.gguf	Q8_0	7.17GB

Original model description: --- libraryname: transformers license: apache-2.0 basemodel:

—nbeerbower/flammen15-gutenberg-DPO-v1-7B datasets:
—chargoddard/chai-dpo ---

image/png

flammen15X-mistral-7B

A Mistral 7B LLM built from merging pretrained models and finetuning on Jon Durbin's Gutenberg DPO set and Charles Goddard's Chai DPO set. Flammen specializes in exceptional character roleplay, creative writing, and general intelligence

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:

python

# 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=2,
    gradient_accumulation_steps=2,
    gradient_checkpointing=True,
    learning_rate=2e-5,
    lr_scheduler_type="cosine",
    max_steps=200,
    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,
    max_prompt_length=1024,
    max_length=1536,
    force_use_ref_model=True
)

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