legionlm/Dora-Smol

See axolotl config

axolotl version: 0.4.0

base_model: /media/renfroe/llms/SmolLM-360M/

model_type: LlamaForCausalLM
tokenizer_type: GPT2Tokenizer
seed: 122887 
load_in_8bit: false
load_in_4bit: false
strict: false

max_steps: 0
resume_from_checkpoint: 
datasets:
  - path: /home/renfroe/Desktop/sqa_tiny-llama_dataset/Dynamic_Optimization_Methods_with_Applications_sqa_answers_only.json
    type:
      field_instruction: question
      field_output: answer
      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"
  - path: /home/renfroe/Dev/tinyllama-models/dataset/open_hermes_top_tech.json
    type: sharegpt
  - path:  /home/renfroe/Desktop/sqa_tiny-llama_dataset/hermes_prior_knowledge_question_expansion_with_answers.json
    type:
      field_instruction: question
      field_output: answer
      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"
  - path:  /home/renfroe/Desktop/sqa_tiny-llama_dataset/hermes_prior_knowledge_question_expansion_with_answers.json
    type:
      field_instruction: question
      field_output: answer
      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"
  - path:  /home/renfroe/Desktop/sqa_tiny-llama_dataset/or-farm_sharegpt.json
    type: sharegpt
  


dataset_prepared_path:
val_set_size: 0.2
output_dir: ./SmolLM-Ora
auto_resume_from_checkpoints: false

sequence_len: 2048
sample_packing: true
chat_template: chatml

wandb_project: SmolLM-Ora
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 1
micro_batch_size: 10
num_epochs: 4
optimizer: adamw_bnb_8bit
lr_scheduler: linear
weight_decay: 0.0000001
learning_rate: 0.0001
lr_scheduler_kwargs:
  #  num_cycles: 3 

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

eval_sample_packing: False

warmup_steps: 50
evals_per_epoch: 4
eval_table_size:
saves_per_epoch: 4
debug:
deepspeed:

fsdp:
fsdp_config:
special_tokens:
  bos_token: "<|endoftext|>"
  eos_token: "<|endoftext|>"
  pad_token: "<|endoftext|>"

SmolLM-Ora

This model was trained from scratch on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8298

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: 10
• eval_batch_size: 10
• seed: 122887
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 50
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
1.0131	0.01	1	1.0419
0.9727	0.25	27	0.9962
0.953	0.5	54	0.9076
0.8494	0.75	81	0.8792
0.9297	1.0	108	0.8632
0.8801	1.22	135	0.8527
0.8133	1.47	162	0.8459
0.8342	1.72	189	0.8410
0.8973	1.97	216	0.8376
0.7731	2.19	243	0.8350
0.8207	2.44	270	0.8332
0.7963	2.69	297	0.8318
0.81	2.94	324	0.8309
0.8351	3.18	351	0.8302
0.8104	3.43	378	0.8299
0.9019	3.68	405	0.8298
0.7828	3.93	432	0.8298

Framework versions

• Transformers 4.40.0.dev0
• Pytorch 2.0.1+cu117
• Datasets 2.18.0
• Tokenizers 0.15.0