minpeter/QLoRA-Llama-3.2-1B-chatml-tool-v3

See axolotl config

axolotl version: 0.6.0

base_model: minpeter/Llama-3.2-1B-AlternateTokenizer-tool-chatml
# revision_of_model: 10f4b4db33e5221f75678044d524bdc2d8b5b056

load_in_8bit: false
load_in_4bit: true
strict: false

lora_modules_to_save:
  - embed_tokens
  - lm_head

datasets:
  # - path: teknium/OpenHermes-2.5
  #   type: chat_template
  #   chat_template: chatml
  #   field_messages: conversations
  #   message_field_role: from
  #   message_field_content: value
  #   shards: 800

  - path: minpeter/hermes-function-calling-v1-jsonl
    data_files:
      - func-calling-singleturn.jsonl
      - func-calling.jsonl
    type: chat_template
    chat_template: chatml
    field_messages: conversations
    message_field_role: from
    message_field_content: value

  - path: minpeter/hermes-function-calling-v1-jsonl
    data_files:
      - glaive-function-calling-5k.jsonl
    type: chat_template
    chat_template: chatml
    field_messages: conversations
    message_field_role: from
    message_field_content: value

save_safetensors: true
auto_resume_from_checkpoints: false
save_steps: 200

chat_template: chatml
dataset_prepared_path: last_run_prepared
val_set_size: 0.05
output_dir: ./output

adapter: qlora
lora_model_dir:

sequence_len: 4096
sample_packing: true
eval_sample_packing: true
pad_to_sequence_len: true

lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_fan_in_fan_out:
lora_target_modules:
  - gate_proj
  - down_proj
  - up_proj
  - q_proj
  - v_proj
  - k_proj
  - o_proj

wandb_project: "axolotl"
wandb_entity: "kasfiekfs-e"
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 3
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

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

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

loss_watchdog_threshold: 5.0
loss_watchdog_patience: 3

warmup_steps: 10
evals_per_epoch: 2
eval_table_size:
eval_max_new_tokens: 128
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:

special_tokens:
  bos_token: <|begin_of_text|>
  eos_token: <|im_end|>
  pad_token: <|end_of_text|>

# <--- unsloth config --->
unsloth_lora_mlp: true
unsloth_lora_qkv: true
unsloth_lora_o: true

output

This model is a fine-tuned version of minpeter/Llama-3.2-1B-AlternateTokenizer-tool-chatml on the minpeter/hermes-function-calling-v1-jsonl and the minpeter/hermes-function-calling-v1-jsonl datasets. It achieves the following results on the evaluation set:

• Loss: 0.3821

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: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 3.0

Training results

Training Loss	Epoch	Step	Validation Loss
1.3359	0.0028	1	1.2171
0.5638	0.4997	179	0.4399
0.385	0.9993	358	0.3942
0.2324	1.4969	537	0.3905
0.1998	1.9965	716	0.3729
0.0984	2.4941	895	0.3817
0.2157	2.9937	1074	0.3821

Framework versions

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