See axolotl config

axolotl version: 0.6.0

# git clone https://github.com/axolotl-ai-cloud/axolotl
# cd axolotl
# git checkout 844331005c1ef45430ff26b9f42f757dce6ee66a
# pip3 install packaging ninja huggingface_hub[cli]
# pip3 install -e '.[flash-attn,deepspeed]'
# apt update && apt install libopenmpi-dev 
# pip install mpi4py
# huggingface-cli login --token $hf_key && wandb login $wandb_key
# python -m axolotl.cli.preprocess nemo-rp-test-human.yml
# accelerate launch -m axolotl.cli.train qwen-rp-test-human-lite.yml
# python -m axolotl.cli.merge_lora qwen-rp-test-human.yml
# huggingface-cli upload ToastyPigeon/tqi-some-rp-40 train-workspace/merged . --exclude "*.md"
# sleep 10h; runpodctl stop pod $RUNPOD_POD_ID &

# git clone https://github.com/axolotl-ai-cloud/axolotl && cd axolotl && pip3 install packaging ninja huggingface_hub[cli] && pip3 install -e '.[flash-attn,deepspeed]' && cd ..
# apt update && apt install libopenmpi-dev && pip install mpi4py

# Model
base_model: Qwen/Qwen2.5-14B-Instruct
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: true
strict: false
bf16: true
fp16:
tf32: false
flash_attention: true
special_tokens:

# Output
output_dir: ./train-workspace
hub_model_id: ToastyPigeon/qwen-inst-rp-lite-test-h-qlora
hub_strategy: "checkpoint"
auto_resume_from_checkpoint: true
#resume_from_checkpoint: ./train-workspace/checkpoint-304
saves_per_epoch: 5
save_total_limit: 3

# Data
sequence_len: 8192 # fits
min_sample_len: 2048
chat_template: chatml
dataset_prepared_path: last_run_prepared
datasets:
  - path: ToastyPigeon/some-rp
    type: chat_template
    field_messages: conversations
    message_field_role: from
    message_field_content: value
warmup_steps: 20
shuffle_merged_datasets: true
sample_packing: true
pad_to_sequence_len: true

# Batching
num_epochs: 1
gradient_accumulation_steps: 1
micro_batch_size: 4
eval_batch_size: 1

# Evaluation
val_set_size: 80
evals_per_epoch: 10
eval_table_size:
eval_max_new_tokens: 256
eval_sample_packing: false

save_safetensors: true

# WandB
wandb_project: Qwen-Rp-Test
#wandb_entity:

gradient_checkpointing: 'unsloth'
gradient_checkpointing_kwargs:
  use_reentrant: false

unsloth_cross_entropy_loss: true
#unsloth_lora_mlp: true
#unsloth_lora_qkv: true
#unsloth_lora_o: true

# LoRA
adapter: qlora
lora_r: 16
lora_alpha: 32
lora_dropout: 0.125
lora_target_linear: true
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj
lora_modules_to_save:
#peft_use_rslora: true
#loraplus_lr_ratio: 8

# Optimizer
optimizer: paged_ademamix_8bit
lr_scheduler: cosine
learning_rate: 5e-5
cosine_min_lr_ratio: 0.1
weight_decay: 0.01
max_grad_norm: 1.0

# Misc
train_on_inputs: false
group_by_length: false
early_stopping_patience:
local_rank:
logging_steps: 1
xformers_attention:
#debug:
deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16.json # previously blank
fsdp:
fsdp_config:

plugins:
  - axolotl.integrations.liger.LigerPlugin
liger_rope: true
liger_rms_norm: true
liger_layer_norm: true
liger_glu_activation: true
liger_fused_linear_cross_entropy: true

gc_steps: 10
seed: 69

qwen-inst-rp-lite-test-h-qlora

This model is a fine-tuned version of Qwen/Qwen2.5-14B-Instruct on the ToastyPigeon/some-rp dataset. It achieves the following results on the evaluation set:

• Loss: 2.4369

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 4
• eval_batch_size: 1
• seed: 69
• distributed_type: multi-GPU
• num_devices: 2
• total_train_batch_size: 8
• total_eval_batch_size: 2
• optimizer: Use OptimizerNames.PAGED_ADEMAMIX_8BIT and the args are: No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 20
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss
2.6892	0.0033	1	2.8023
2.5375	0.1030	31	2.4737
2.5817	0.2060	62	2.4597
2.3864	0.3090	93	2.4535
2.3276	0.4120	124	2.4488
2.4463	0.5150	155	2.4445
2.2724	0.6179	186	2.4414
2.3688	0.7209	217	2.4394
2.3059	0.8239	248	2.4378
2.4368	0.9269	279	2.4369

Framework versions

• PEFT 0.14.0
• Transformers 4.47.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0