DOL DevOps LLaMA-3 8B — Version 3

Model được fine-tune từ Meta-Llama-3-8B-Instruct bằng kỹ thuật LoRA trên bộ dataset DevErrBench.

Thông tin training

Thông số Giá trị
Base model mlx-community/Meta-Llama-3-8B-Instruct-4bit
Method LoRA (rank=8, dropout=0.05)
Optimizer AdamW (weight_decay=0.01)
LR Schedule Cosine Decay + Warmup 100
Learning Rate 2e-6
Best Val Loss 0.1620 (Iter 1000)
Early Stopping Patience=2
Dataset 31k Hard-Negative + 60k Replay = ~92k samples
Framework MLX (Apple Silicon)

Sử dụng trên Kaggle / Colab (GPU)

from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

model_id = "vdtry06/deveerbench-llama3-v3"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    device_map="auto",
)

Sử dụng trên Mac (MLX)

mlx_lm.server \
  --model vdtry06/deveerbench-llama3-v3 \
  --host 0.0.0.0 --port 8080
Downloads last month
29
Safetensors
Model size
8B params
Tensor type
U32
·
F16
·
MLX
Hardware compatibility
Log In to add your hardware

4-bit

Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for vdtry06/deveerbench-llama3-v3

Adapter
(1206)
this model