models/Qwen1.5-MoE-A2.7B-Wikihow

This model is a fine-tuned version of Qwen/Qwen1.5-MoE-A2.7B on the HuggingFaceTB/cosmopedia dataset.

How to use it

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="MaziyarPanahi/Qwen1.5-MoE-A2.7B-Wikihow")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("MaziyarPanahi/Qwen1.5-MoE-A2.7B-Wikihow")
model = AutoModelForCausalLM.from_pretrained("MaziyarPanahi/Qwen1.5-MoE-A2.7B-Wikihow")

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• total_eval_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 1

Training results

See axolotl config

axolotl version: 0.4.0

base_model: Qwen/Qwen1.5-MoE-A2.7B
trust_remote_code: true

load_in_8bit: false
load_in_4bit: true
strict: false

# hub_model_id: MaziyarPanahi/Qwen1.5-MoE-A2.7B-Wikihow
# hf_use_auth_token: true

chat_template: chatml

datasets:
  - path: HuggingFaceTB/cosmopedia
    name: wikihow
    type:
      system_prompt: ""
      field_instruction: prompt
      field_output: text
      format: "<|im_start|>user\n{instruction}<|im_end|>\n<|im_start|>assistant\n"
      no_input_format: "<|im_start|>user\n{instruction}<|im_end|>\n<|im_start|>assistant\n"
    
dataset_prepared_path:
val_set_size: 0.0
output_dir: ./models/Qwen1.5-MoE-A2.7B-Wikihow

sequence_len: 2048
sample_packing: false
pad_to_sequence_len: false

adapter: lora
lora_model_dir:
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 1
optimizer: paged_adamw_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: true

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 10
evals_per_epoch: 4
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:

Framework versions

• PEFT 0.10.0
• Transformers 4.40.0.dev0
• Pytorch 2.2.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2