gpt-oss-120b-multilingual-reasoning

<!-- This model card has been generated automatically according to the information the Trainer had access to. You should probably proofread and complete it, then remove this comment. -->

<img src="https://raw.githubusercontent.com/axolotl-ai-cloud/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>

<details><summary>See axolotl config</summary>

axolotl version: 0.12.0

# the original mxfp4 quantized model is not supported with FSDP cpu_ram_efficient_loading
# FSDP cpu_ram_efficient_loading is used to reduce the initial CPU memory usage when loading the model
base_model: axolotl-ai-co/gpt-oss-120b-dequantized

use_kernels: false

dp_shard_size: 8  # requires 2x8xH100 nodes

plugins:
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin

experimental_skip_move_to_device: true  # prevent OOM by NOT putting model to GPU before sharding

datasets:
  - path: HuggingFaceH4/Multilingual-Thinking
    type: chat_template
    field_thinking: thinking
    template_thinking_key: thinking

dataset_prepared_path: last_run_prepared
val_set_size: 0
output_dir: ./outputs/gpt-oss-120b/
#save_only_model: true

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

wandb_project: gpt-oss-20b
wandb_name: fft-20b

gradient_accumulation_steps: 1
micro_batch_size: 4
num_epochs: 1

optimizer: adamw_torch_fused  # 8bit optimizers do not work with FSDP2 offload
lr_scheduler: constant_with_warmup
learning_rate: 2e-5
load_best_model_at_end: false

bf16: true
tf32: true

flash_attention: true
attn_implementation: kernels-community/vllm-flash-attn3

gradient_checkpointing: true
activation_offloading: true

logging_steps: 1
saves_per_epoch: 1

warmup_ratio: 0.03

special_tokens:
eot_tokens:
  - "<|end|>"

#deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16.json
fsdp_version: 2
fsdp_config:
  offload_params: true
  state_dict_type: SHARDED_STATE_DICT
  auto_wrap_policy: TRANSFORMER_BASED_WRAP
  transformer_layer_cls_to_wrap: GptOssDecoderLayer
  reshard_after_forward: true
  cpu_ram_efficient_loading: true

</details><br>

outputs/gpt-oss-120b/

This model is a fine-tuned version of axolotl-ai-co/gpt-oss-120b-dequantized on the HuggingFaceH4/Multilingual-Thinking dataset.

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: 2e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 32
• total_eval_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant_with_warmup
• training_steps: 8

Training results

Framework versions

• Transformers 4.55.0
• Pytorch 2.8.0+cu128
• Datasets 4.0.0
• Tokenizers 0.21.4