cosmicoptima/computer-1

1---2library_name: transformers3tags:4- generated_from_trainer5datasets:6- /home/athuser/modelC_train/sft_modelC.jsonl7model-index:8- name: models/modelC_out/70B_fft_e19  results: []10---11 12<!-- This model card has been generated automatically according to the information the Trainer had access to. You13should probably proofread and complete it, then remove this comment. -->14 15[<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)16<details><summary>See axolotl config</summary>17 18axolotl version: `0.12.2`19```yaml20# Model C anchor run: Llama-3.1-70B FFT on the assembled keeper set.21# Derived from /models/axolot/llama3_70b_fsdp.yaml (the out_FFT_E precedent);22# dataset swapped to Model C keepers in completion format (full-doc LM loss,23# both speakers, 15% header dropout baked into the jsonl by export_sft.py).24base_model: /models/Llama-3.1-70B25model_type: LlamaForCausalLM26tokenizer_type: AutoTokenizer27load_in_8bit: false28load_in_4bit: false29datasets:30  - path: /home/athuser/modelC_train/sft_modelC.jsonl31    type: completion32    field: text33dataset_prepared_path: /home/athuser/modelC_train/last_run_prepared34val_set_size: 0.0235output_dir: /models/modelC_out/70B_fft_e136sequence_len: 409637sample_packing: true38tf32: true39gradient_accumulation_steps: 440micro_batch_size: 141num_epochs: 142optimizer: adamw_torch_fused43lr_scheduler: constant_with_warmup44learning_rate: 2.0e-0545bf16: true46resume_from_checkpoint:47logging_steps: 148flash_attention: true49warmup_ratio: 0.0350evals_per_epoch: 451saves_per_epoch: 152save_only_model: true53weight_decay: 0.054ddp_backend: nccl55fsdp_version: 256fsdp_config:57  offload_params: false58  cpu_ram_efficient_loading: true59  auto_wrap_policy: TRANSFORMER_BASED_WRAP60  transformer_layer_cls_to_wrap: LlamaDecoderLayer61  state_dict_type: FULL_STATE_DICT62  reshard_after_forward: true63  activation_checkpointing: true64special_tokens:65  pad_token: <|end_of_text|>66 67```68 69</details><br>70 71# models/modelC_out/70B_fft_e172 73This model was trained from scratch on the /home/athuser/modelC_train/sft_modelC.jsonl dataset.74It achieves the following results on the evaluation set:75- Loss: 1.497076- Memory/max Mem Active(gib): 89.1577- Memory/max Mem Allocated(gib): 89.1578- Memory/device Mem Reserved(gib): 94.1579 80## Model description81 82More information needed83 84## Intended uses & limitations85 86More information needed87 88## Training and evaluation data89 90More information needed91 92## Training procedure93 94### Training hyperparameters95 96The following hyperparameters were used during training:97- learning_rate: 2e-0598- train_batch_size: 199- eval_batch_size: 1100- seed: 42101- distributed_type: multi-GPU102- num_devices: 8103- gradient_accumulation_steps: 4104- total_train_batch_size: 32105- total_eval_batch_size: 8106- optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments107- lr_scheduler_type: constant_with_warmup108- lr_scheduler_warmup_steps: 3109- training_steps: 113110 111### Training results112 113| Training Loss | Epoch  | Step | Validation Loss | Mem Active(gib) | Mem Allocated(gib) | Mem Reserved(gib) |114|:-------------:|:------:|:----:|:---------------:|:---------------:|:------------------:|:-----------------:|115| No log        | 0      | 0    | 1.7114          | 27.73           | 27.73              | 31.33             |116| 1.53          | 0.2549 | 29   | 1.5095          | 89.15           | 89.15              | 94.15             |117| 1.49          | 0.5099 | 58   | 1.5013          | 89.15           | 89.15              | 94.15             |118| 1.4778        | 0.7648 | 87   | 1.4970          | 89.15           | 89.15              | 94.15             |119 120 121### Framework versions122 123- Transformers 4.55.2124- Pytorch 2.7.0+cu128125- Datasets 4.0.0126- Tokenizers 0.21.2127