You can deploy this model and start earning money today!

RichardErkhov/
nbeerbower_-_llama-3-bophades-v2-8B-gguf

Name: nbeerbower_-_llama-3-bophades-v2-8B-gguf
Availability: InStock

Views

No views yet

Quantization made by Richard Erkhov.

Github

Discord

Request more models

llama-3-bophades-v2-8B - GGUF

Model creator: https://huggingface.co/nbeerbower/
Original model: https://huggingface.co/nbeerbower/llama-3-bophades-v2-8B/

Name	Quant method	Size
llama-3-bophades-v2-8B.Q2_K.gguf	Q2_K	2.96GB
llama-3-bophades-v2-8B.IQ3_XS.gguf	IQ3_XS	3.28GB
llama-3-bophades-v2-8B.IQ3_S.gguf	IQ3_S	3.43GB
llama-3-bophades-v2-8B.Q3_K_S.gguf	Q3_K_S	3.41GB
llama-3-bophades-v2-8B.IQ3_M.gguf	IQ3_M	3.52GB
llama-3-bophades-v2-8B.Q3_K.gguf	Q3_K	3.74GB
llama-3-bophades-v2-8B.Q3_K_M.gguf	Q3_K_M	3.74GB
llama-3-bophades-v2-8B.Q3_K_L.gguf	Q3_K_L	4.03GB
llama-3-bophades-v2-8B.IQ4_XS.gguf	IQ4_XS	4.18GB
llama-3-bophades-v2-8B.Q4_0.gguf	Q4_0	4.34GB
llama-3-bophades-v2-8B.IQ4_NL.gguf	IQ4_NL	4.38GB
llama-3-bophades-v2-8B.Q4_K_S.gguf	Q4_K_S	4.37GB
llama-3-bophades-v2-8B.Q4_K.gguf	Q4_K	4.58GB
llama-3-bophades-v2-8B.Q4_K_M.gguf	Q4_K_M	4.58GB
llama-3-bophades-v2-8B.Q4_1.gguf	Q4_1	4.78GB
llama-3-bophades-v2-8B.Q5_0.gguf	Q5_0	5.21GB
llama-3-bophades-v2-8B.Q5_K_S.gguf	Q5_K_S	5.21GB
llama-3-bophades-v2-8B.Q5_K.gguf	Q5_K	5.34GB
llama-3-bophades-v2-8B.Q5_K_M.gguf	Q5_K_M	5.34GB
llama-3-bophades-v2-8B.Q5_1.gguf	Q5_1	5.65GB
llama-3-bophades-v2-8B.Q6_K.gguf	Q6_K	6.14GB
llama-3-bophades-v2-8B.Q8_0.gguf	Q8_0	7.95GB

Original model description:

library_name: transformers base_model:

nbeerbower/llama-3-sauce-v1-8B datasets:
jondurbin/truthy-dpo-v0.1
kyujinpy/orca_math_dpo license: other license_name: llama3

llama-3-bophades-v2-8B

This model is based on Llama-3-8b, and is governed by META LLAMA 3 COMMUNITY LICENSE AGREEMENT

llama-3-sauce-v1-8B finetuned on jondurbin/truthy-dpo-v0.1 and kyujinpy/orca_math_dpo.

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=4,
    gradient_checkpointing=True,
    learning_rate=3e-5,
    lr_scheduler_type="cosine",
    max_steps=420,
    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=2048,
    max_length=4096,
    force_use_ref_model=True
)

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

RichardErkhov/nbeerbower_-_llama-3-bophades-v2-8B-gguf

Original model description:

llama-3-bophades-v2-8B

Method

Configuration

RichardErkhov/
nbeerbower_-_llama-3-bophades-v2-8B-gguf