See axolotl config

axolotl version: 0.4.1

base_model: meta-llama/Llama-3.2-1B
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

chat_template: chatml
datasets:
  - path: datasets/airoboros_3.2_without_contextual_slimorca_orca_sharegpt.json
    ds_type: json
    type: sharegpt
    conversation: chatml

  - path: datasets/allenai_wild_chat_gpt4_english_toxic_random_half_4k_sharegpt.json
    ds_type: json
    type: sharegpt
    strict: false
    conversation: chatml

  - path: datasets/buzz_unstacked_chosen_math_removed_filtered.json
    ds_type: json
    type: alpaca
    conversation: chatml

  - path: datasets/capybara_sharegpt.json
    ds_type: json
    type: sharegpt
    conversation: chatml

  - path: datasets/cot_alpaca_gpt4_extracted_openhermes_2.5_sharegpt.json
    ds_type: json
    type: sharegpt
    conversation: chatml

  - path: datasets/everythinglm-data-v3_sharegpt.json
    ds_type: json
    type: sharegpt
    strict: false
    conversation: chatml

  - path: datasets/gpt4_data_lmys_1m_sharegpt.json
    ds_type: json
    type: sharegpt
    conversation: chatml

  - path: datasets/gpteacher-instruct-special-alpaca.json
    ds_type: json
    type: gpteacher
    conversation: chatml

  - path: datasets/merged_all.json
    ds_type: json
    type: alpaca
    conversation: chatml

  - path: datasets/no_robots_sharegpt.json
    ds_type: json
    type: sharegpt
    strict: false
    conversation: chatml

  - path: datasets/oasst_top1_from_fusechatmixture_sharegpt.json
    ds_type: json
    type: sharegpt
    strict: false
    conversation: chatml

  - path: datasets/pippa_bagel_repo_3k_sharegpt.json
    ds_type: json
    type: sharegpt
    conversation: chatml

  - path: datasets/rpguild_quarter_alignment_lab_sharegpt.json
    ds_type: json
    type: sharegpt
    conversation: chatml

  - path: datasets/sharegpt_gpt4_english.json
    ds_type: json
    type: sharegpt
    conversation: chatml

  - path: datasets/slimorca_dedup_filtered_95k_sharegpt.json
    ds_type: json
    type: sharegpt
    conversation: chatml

  - path: datasets/soda_diaolog_longest_tenth_buzz_sharegpt.json
    ds_type: json
    type: sharegpt
    conversation: chatml

  - path: datasets/synthia-v1.3_sharegpt_12500.json
    ds_type: json
    type: sharegpt
    conversation: chatml

  - path: datasets/system_conversations_dolphin_sharegpt.json
    ds_type: json
    type: sharegpt
    conversation: chatml
  
  - path: datasets/NuminaMath-CoT-olympiads-40k_alpaca.json
    ds_type: json
    type: alpaca
    conversation: chatml

  - path: datasets/math-gpt-4o-40k_alpaca.json
    ds_type: json
    type: alpaca
    conversation: chatml

  - path: datasets/sonnet3.5_science_conversations_sharegpt.json
    ds_type: json
    type: sharegpt
    conversation: chatml

  - path: datasets/reasoning-0.01_sharegpt.jsonl
    ds_type: json
    type: sharegpt
    conversation: chatml

dataset_prepared_path: last_run_prepared
val_set_size: 0.002

output_dir: ./Einstein-v8-Llama3.2-1B-model

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true
eval_sample_packing: false

wandb_project: Einstein
wandb_entity:
wandb_watch:
wandb_name: Einstein-v8-Llama3.2-1B-2-epoch
wandb_log_model:
hub_model_id: Weyaxi/Einstein-v8-Llama3.2-1B

save_safetensors: true

gradient_accumulation_steps: 4
micro_batch_size: 4 
num_epochs: 2
optimizer: adamw_bnb_8bit # look
lr_scheduler: cosine
learning_rate: 0.000005 # look

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

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

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_table_max_new_tokens: 128
saves_per_epoch: 1
debug:

deepspeed: axolotl/deepspeed_configs/zero3_bf16.json
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "<|im_end|>"
  unk_token: "<unk>"
  pad_token: <|end_of_text|> # changed
tokens:
  - "<|im_start|>"

Einstein-v8-Llama3.2-1B

This model is a fine-tuned version of meta-llama/Llama-3.2-1B on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.9292

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-06
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 4
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• total_eval_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 2

Training results

Training Loss	Epoch	Step	Validation Loss
1.4261	0.0009	1	1.4028
1.0487	0.2501	268	0.9917
1.0484	0.5001	536	0.9652
1.0039	0.7502	804	0.9499
1.0528	1.0002	1072	0.9399
0.9559	1.2481	1340	0.9345
0.9078	1.4981	1608	0.9309
0.9702	1.7481	1876	0.9295
0.929	1.9981	2144	0.9292

Framework versions

• Transformers 4.45.0
• Pytorch 2.3.1+cu121
• Datasets 2.21.0
• Tokenizers 0.20.0

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	4.63
IFEval (0-Shot)	18.62
BBH (3-Shot)	3.01
MATH Lvl 5 (4-Shot)	0.00
GPQA (0-shot)	1.12
MuSR (0-shot)	3.22
MMLU-PRO (5-shot)	1.79