---
library_name: peft
license: apache-2.0
base_model: arcee-ai/Virtuoso-Small
tags:
- axolotl
- generated_from_trainer
datasets:
- ToastyPigeon/some-rp
model-index:
- name: qwen-rp-test-h-qlora
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

[<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/axolotl-ai-cloud/axolotl)
<details><summary>See axolotl config</summary>

axolotl version: `0.6.0`
```yaml
# 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]'
# 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.yml
# python -m axolotl.cli.merge_lora nemo-rp-test-human.yml
# huggingface-cli upload Columbidae/nemo-rp-test-human 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 ..

# Model
base_model: arcee-ai/Virtuoso-Small
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-rp-test-h-qlora
hub_strategy: "checkpoint"
auto_resume_from_checkpoint: true
#resume_from_checkpoint: ./train-workspace/checkpoint-304
saves_per_epoch: 10
save_total_limit: 3

# Data
sequence_len: 8192 # fits
min_sample_len: 128
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: 1
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: 64
lora_alpha: 128
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: 1e-4
cosine_min_lr_ratio: 0.1
weight_decay: 0.1
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

# Debug config
debug: true
seed: 69
```

</details><br>

# qwen-rp-test-h-qlora

This model is a fine-tuned version of [arcee-ai/Virtuoso-Small](https://huggingface.co/arcee-ai/Virtuoso-Small) on the ToastyPigeon/some-rp dataset.
It achieves the following results on the evaluation set:
- Loss: 2.3971

## 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: 0.0001
- train_batch_size: 1
- eval_batch_size: 1
- seed: 69
- distributed_type: multi-GPU
- num_devices: 8
- total_train_batch_size: 8
- total_eval_batch_size: 8
- 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 |
|:-------------:|:------:|:----:|:---------------:|
| 2.4903        | 0.0026 | 1    | 2.5546          |
| 2.2882        | 0.1016 | 39   | 2.4302          |
| 2.3752        | 0.2031 | 78   | 2.4171          |
| 2.3249        | 0.3047 | 117  | 2.4119          |
| 2.2504        | 0.4062 | 156  | 2.4081          |
| 2.3905        | 0.5078 | 195  | 2.4030          |
| 2.3354        | 0.6094 | 234  | 2.4018          |
| 2.5473        | 0.7109 | 273  | 2.3996          |
| 2.4123        | 0.8125 | 312  | 2.3982          |
| 2.2878        | 0.9141 | 351  | 2.3971          |


### Framework versions

- PEFT 0.14.0
- Transformers 4.47.1
- Pytorch 2.5.1+cu124
- Datasets 3.1.0
- Tokenizers 0.21.0