ba3dba3e4f
* add kernels for gpt oss models * add support for gpt-oss * typo incorrect package * fix: layout for configs and added wandb/epochs * add gptoss example w offload and set moe leaf for z3 * add support for Mxfp4Config from yaml * update yaml to use official model * fix lora and don't allow triton to go above 3.3.1 * fix lr and tweak vram use * fix range for triton since pinned wasn't compatible with toch 2.6.0 * update cce with gpt oss patches --------- Co-authored-by: NanoCode012 <nano@axolotl.ai>
63 lines
1.2 KiB
YAML
63 lines
1.2 KiB
YAML
base_model: openai/gpt-oss-20b
|
|
use_kernels: true
|
|
model_quantization_config: Mxfp4Config
|
|
model_quantization_config_kwargs:
|
|
dequantize: true
|
|
|
|
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: winglian/pirate-ultrachat-10k
|
|
type: chat_template
|
|
split: train
|
|
|
|
dataset_prepared_path: last_run_prepared
|
|
val_set_size: 0
|
|
output_dir: ./outputs/gpt-oss-out/
|
|
|
|
sequence_len: 4096
|
|
sample_packing: true
|
|
|
|
wandb_project:
|
|
wandb_entity:
|
|
wandb_watch:
|
|
wandb_name:
|
|
wandb_log_model:
|
|
|
|
gradient_accumulation_steps: 2
|
|
micro_batch_size: 1
|
|
num_epochs: 1
|
|
|
|
optimizer: adamw_torch_8bit
|
|
lr_scheduler: constant_with_warmup
|
|
learning_rate: 2e-5
|
|
|
|
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: 0.1
|
|
|
|
special_tokens:
|
|
eot_tokens:
|
|
- "<|end|>"
|
|
|
|
fsdp_version: 2
|
|
fsdp_config:
|
|
offload_params: false
|
|
state_dict_type: SHARDED_STATE_DICT
|
|
auto_wrap_policy: TRANSFORMER_BASED_WRAP
|
|
transformer_layer_cls_to_wrap: GptOssDecoderLayer
|
|
reshard_after_forward: true
|