See axolotl config

axolotl version: 0.10.0.dev0

base_model: RedHatAI/Sparse-Llama-3.1-8B-2of4

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: trl-lib/tldr
    type:
      system_prompt: "Give a TL;DR of the following Reddit post."
      field_system: system
      field_instruction: prompt
      field_output: completion
      format: "<|user|>\n{instruction}\n<|assistant|>\n"
      no_input_format: "<|user|>\n{instruction}\n<|assistant|>\n"
    split: train

dataset_prepared_path: last_run_prepared
output_dir: Sparse-Llama-3.1-8B-2of4-tldr

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

torch.compile: true
gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 1
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 1e-5
max_grad_norm: 1.0

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false

train_on_inputs: false
bf16: auto
fp16:
tf32: false

early_stopping_patience:
resume_from_checkpoint:
logging_steps: 1
flash_attention: true

warmup_ratio: 0.05
eval_steps: 0.05
val_set_size: 0.05
save_strategy: "best"
metric_for_best_model: "loss"

debug:
deepspeed:
weight_decay: 0.0
special_tokens:
  pad_token: "<|end_of_text|>"

seed: 0

plugins:
  - axolotl.integrations.liger.LigerPlugin
  - axolotl.integrations.llm_compressor.LLMCompressorPlugin

liger_rope: true
liger_rms_norm: true
liger_glu_activation: true
liger_layer_norm: true
liger_fused_linear_cross_entropy: true

llmcompressor:
  recipe:
    finetuning_stage:
      finetuning_modifiers:
        ConstantPruningModifier:
          targets: [
            're:.*q_proj.weight',
            're:.*k_proj.weight',
            're:.*v_proj.weight',
            're:.*o_proj.weight',
            're:.*gate_proj.weight',
            're:.*up_proj.weight',
            're:.*down_proj.weight',
          ]
          start: 0
  save_compressed: true

Sparse-Llama-3.1-8B-2of4-tldr

This model is a fine-tuned version of RedHatAI/Sparse-Llama-3.1-8B-2of4 on the trl-lib/tldr dataset. It achieves the following results on the evaluation set:

• Loss: 1.8295

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: 1e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 0
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 8
• total_eval_batch_size: 8
• optimizer: Use adamw_bnb_8bit 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: 66
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
2.4149	0.0008	1	2.2321
1.9603	0.0505	67	1.8758
1.8909	0.1010	134	1.8560
1.8109	0.1515	201	1.8491
1.7688	0.2020	268	1.8441
1.8535	0.2524	335	1.8411
1.773	0.3029	402	1.8381
1.8349	0.3534	469	1.8360
1.8382	0.4039	536	1.8342
1.7975	0.4544	603	1.8328
1.8171	0.5049	670	1.8317
1.8309	0.5554	737	1.8309
1.8158	0.6059	804	1.8303
1.8684	0.6564	871	1.8298
1.7743	0.7069	938	1.8296
1.7132	0.7573	1005	1.8295
1.7912	0.8078	1072	1.8294
1.9432	0.8583	1139	1.8295
1.7789	0.9088	1206	1.8294
1.8084	0.9593	1273	1.8295

Framework versions

• Transformers 4.51.3
• Pytorch 2.7.0+cu126
• Datasets 3.5.1
• Tokenizers 0.21.1