README.md

---
library_name: transformers
license: apache-2.0
base_model: Qwen/Qwen3-4B-Base
tags:
- generated_from_trainer
datasets:
- winglian/OpenThoughts-114k-math-correct-qwen3-14b-math-prepared-topk128-normalized
model-index:
- name: outputs/out-kd-4b-offline-t1-topk128-rkl
  results: []
---

<!-- 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"/>](https://github.com/axolotl-ai-cloud/axolotl)
<details><summary>See axolotl config</summary>

axolotl version: `0.10.0.dev0`
```yaml
base_model: Qwen/Qwen3-4B-Base

plugins:
  - axolotl.integrations.kd.KDPlugin
  - axolotl.integrations.liger.LigerPlugin

liger_rms_norm: true
liger_glu_activation: true

  # torch_compile: true

strict: false

kd_trainer: true
kd_ce_alpha: 0.1
kd_alpha: 1.0
kd_temperature: 1.0
kd_beta: 1.0
kd_normalize_topk: true

dataloader_prefetch_factor: 1
dataloader_num_workers: 2
dataloader_pin_memory: true
dataloader_drop_last: true

gc_steps: -1  # gc at the end of each epoch

chat_template: qwen3
datasets:
  - path: winglian/OpenThoughts-114k-math-correct-qwen3-14b-math-prepared-topk128-normalized
    type: chat_template
    split: train
    split_thinking: true
    eot_tokens:
      - "<|im_end|>"

skip_prepare_dataset: true
dataset_prepared_path: last_run_prepared
val_set_size: 0.0
output_dir: ./outputs/out-kd-4b-offline-t1-topk128-rkl

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

wandb_project: kd-4b-math
wandb_entity: axolotl-ai
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 1
micro_batch_size: 4
num_epochs: 2
optimizer: adamw_torch_fused
adam_beta2: 0.999
lr_scheduler: rex
learning_rate: 3e-5
max_grad_norm: 0.2
save_safetensors: true

bf16: true
tf32: true

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: false
logging_steps: 1
flash_attention: true

warmup_steps: 100
evals_per_epoch: 4
saves_per_epoch: 2
debug:
weight_decay: 0.0
special_tokens:
  eos_token: <|im_end|>
deepspeed: deepspeed_configs/zero2_torch_compile.json


```

</details><br>

# outputs/out-kd-4b-offline-t1-topk128-rkl

This model is a fine-tuned version of [Qwen/Qwen3-4B-Base](https://huggingface.co/Qwen/Qwen3-4B-Base) on the winglian/OpenThoughts-114k-math-correct-qwen3-14b-math-prepared-topk128-normalized 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: 3e-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: cosine
- lr_scheduler_warmup_steps: 100
- training_steps: 1282

### Training results



### Framework versions

- Transformers 4.52.3
- Pytorch 2.7.0+cu128
- Datasets 3.6.0
- Tokenizers 0.21.1