README.md

---
library_name: transformers
license: gemma
base_model: unsloth/gemma-3-270m-it
tags:
- axolotl
- generated_from_trainer
datasets:
- allura-org/EU01-S2
- allenai/tulu-3-sft-personas-instruction-following
- ToastyPigeon/mixed-medical-reasoning-formatted
- ToastyPigeon/steve-and-marvin
- ToastyPigeon/kimi-stories-instruct
- ToastyPigeon/new-story-dataset
- allura-org/fujin-instruct-v2
- ToastyPigeon/gutenberg-sft
- ToastyPigeon/SpringDragon
- ToastyPigeon/some-erotica
model-index:
- name: micro-glitter
  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: `0.11.0.dev0`
```yaml
# === Model Configuration ===
base_model: unsloth/gemma-3-270m-it
load_in_8bit: false
load_in_4bit: false

# === HF Configuration === 
hub_model_id: allura-forge/micro-glitter
hub_strategy: "checkpoint"
output_dir: /workspace/aibox-standalone-pool/axolotl/lilglitter-ckpts

# === Training Setup ===
num_epochs: 2
micro_batch_size: 4
gradient_accumulation_steps: 8
sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true
#max_steps: 10
# === Evaluation ===
val_set_size: 0.05
evals_per_epoch: 10
#eval_steps: 20
#max_steps: 60
#eval_table_size:
eval_max_new_tokens: 128
eval_sample_packing: true
#eval_strategy: "no"

# === LoRA Configuration ===
#adapter: qlora
#lora_model_dir:
#lora_r: 128
#lora_alpha: 16
#lora_dropout: 0.25
#lora_target_linear: true
#lora_target_modules:
#  - embed_tokens
#  - lm_head
lora_fan_in_fan_out:
lora_target_modules:
#peft_use_rslora: true
lora_modules_to_save:
#  - embed_tokens
#  - lm_head
#fix_untrained_tokens: true
#lora_mlp_kernel: true
#lora_qkv_kernel: true
#lora_o_kernel: true

# === Hyperparameter Configuration ===
#optimizer: apollo_adamw_layerwise
warmup_steps: 0
optimizer: adamw_torch_fused
#optimizer: paged_adamw_8bit
#optim_args:
#  enable_stochastic_rounding: true
#  enable_cautious: true
#  enable_8bit: true
# Apollo-mini configuration:
#optim_args: "proj=random,rank=128,scale=128.0,scale_type=tensor,update_proj_gap=100"
# Regular Apollo configuration:
# optim_args: 
#optim_target_modules: all_linear
learning_rate: 1e-5
lr_scheduler: cosine
#cosine_min_lr_ratio: 0.2
#lr_scheduler: cosine_with_min_lr
#lr_scheduler_kwargs:
#  cosine_min_lr: 1e-6
weight_decay: 0.01
max_grad_norm: 2.0
#warmup_steps: 0
#warmup_ratio: 0.025


# === Data Configuration ===
#
#chat_template: jinja
#chat_template_jinja: "{% for message in messages %}{% if not loop.first %}{{' \n\n' }}{% endif %}{% if message['role'] == 'system' %}{{ '### System:\n' + message['content'].strip() }}{% elif message['role'] == 'user' %}{{ '### Instruction:\n' + message['content'].strip() }}{% elif message['role'] == 'assistant' %}{{ '### Response:\n' + message['content'].strip() + eos_token }}{% endif %}{% endfor %}"

#chat_template_jinja: "{%- set default_system_message = \"You are Mistral Small 3, a Large Language Model (LLM) created by Mistral AI, a French startup headquartered in Paris. You obediently fulfill the user's requests.\" %}\n\n{{- bos_token }}\n\n{%- if messages[0]['role'] == 'system' %}\n    {%- if messages[0]['content'] is string %}\n        {%- set system_message = messages[0]['content'] %}\n    {%- else %}\n        {%- set system_message = messages[0]['content'][0]['text'] %}\n    {%- endif %}\n    {%- set loop_messages = messages[1:] %}\n{%- else %}\n    {%- set system_message = default_system_message %}\n    {%- set loop_messages = messages %}\n{%- endif %}\n{{- '[SYSTEM_PROMPT]' + system_message + '[/SYSTEM_PROMPT]' }}\n\n{%- for message in loop_messages %}\n    {%- if message['role'] == 'user' %}\n        {%- if message['content'] is string %}\n            {{- '[INST]' + message['content'] + '[/INST]' }}\n        {%- else %}\n            {{- '[INST]' }}\n            {%- for bl (line truncated to 1000 characters)
#chat_template: chatml
#special_tokens:
#  eos_token: "<|im_end|>"
#  eos_token: "</s>"
#tokenizer_use_mistral_common: true
shuffle_merged_datasets: true
datasets:
  - path: allura-org/EU01-S2
    type: chat_template
    field_messages: conversations
    message_property_mappings:
      role: from
      content: value
  - path: allenai/tulu-3-sft-personas-instruction-following
    type: chat_template
    split: train[:10%]
  - path: ToastyPigeon/mixed-medical-reasoning-formatted
    type: chat_template
    data_files: mixed-medical-thinking.json
    split: train[:10%]
  - path: ToastyPigeon/steve-and-marvin
    type: completion
    data_files: marvin.json
  - path: ToastyPigeon/kimi-stories-instruct
    type: chat_template
  - path: ToastyPigeon/new-story-dataset
 #   type: customcompletion-regex
    type: completion
    data_files: new-story-dataset-v2.json
  - path: allura-org/fujin-instruct-v2
#    type: customchatml-regex
    type: chat_template
    field_messages: conversations
    message_property_mappings:
      role: from
      content: value
#  - path: ToastyPigeon/some-rp-extended
 #   type: customchatml-regex
#    type: chat_template
#    field_messages: conversations
#    message_property_mappings:
#      role: from
#      content: value
#    roles_to_train: ["user","assistant"]
  - path: ToastyPigeon/gutenberg-sft
#    type: customchatml-regex
    type: chat_template
    field_messages: conversations
    message_property_mappings:
      role: from
      content: value
  - path: ToastyPigeon/SpringDragon
#    type: customcompletion-regex
    type: completion
    split: train
  - path: ToastyPigeon/some-erotica
#    type: customcompletion-regex
    type: completion
    split: train[:10%]

dataset_prepared_path: last_run_prepared


# === Plugins ===
plugins:
  - axolotl.integrations.liger.LigerPlugin
  - axolotl.integrations.cut_cross_entropy.CutCrossEntropyPlugin

# === Hardware Optimization ===
#gradient_checkpointing: offload
#gradient_checkpointing_kwargs:
#  use_reentrant: false
liger_rope: true
liger_rms_norm: true
liger_layer_norm: true
liger_glu_activation: true
#liger_fused_linear_cross_entropy: true
cut_cross_entropy: true

#deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16.json

# === FSDP Config === 
#fsdp:
#  - full_shard
#  - auto_wrap
#fsdp_config:
#  fsdp_limit_all_gathers: true
#  fsdp_sync_module_states: true
#  fsdp_offload_params: true
#  fsdp_activation_checkpointing: true
#  fsdp_use_orig_params: false
#  fsdp_cpu_ram_efficient_loading: true
#  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
#  fsdp_transformer_layer_cls_to_wrap: MistralDecoderLayer
#  fsdp_state_dict_type: FULL_STATE_DICT
#  fsdp_sharding_strategy: FULL_SHARD
#  fsdp_version: 2
# === Wandb Tracking ===
wandb_project: TinyGemma
# wandb_entity: [WANDB_ENTITY]
# wandb_name: [WANDB_RUN_NAME]

# === Checkpointing ===
#save_steps: 10
saves_per_epoch: 10
save_total_limit: 1

# === Advanced Settings ===
bf16: auto
flash_attention: true
train_on_inputs: false
group_by_length: false
save_safetensors: true
logging_steps: 1
gc_steps: 10
seed: 69



```

</details><br>

# micro-glitter

This model is a fine-tuned version of [unsloth/gemma-3-270m-it](https://huggingface.co/unsloth/gemma-3-270m-it) on the allura-org/EU01-S2, the allenai/tulu-3-sft-personas-instruction-following, the ToastyPigeon/mixed-medical-reasoning-formatted, the ToastyPigeon/steve-and-marvin, the ToastyPigeon/kimi-stories-instruct, the ToastyPigeon/new-story-dataset, the allura-org/fujin-instruct-v2, the ToastyPigeon/gutenberg-sft, the ToastyPigeon/SpringDragon and the ToastyPigeon/some-erotica datasets.
It achieves the following results on the evaluation set:
- Loss: 3.7387

## 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: 4
- eval_batch_size: 4
- seed: 69
- distributed_type: multi-GPU
- num_devices: 2
- gradient_accumulation_steps: 8
- total_train_batch_size: 64
- total_eval_batch_size: 8
- 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: 8
- training_steps: 296

### Training results

| Training Loss | Epoch  | Step | Validation Loss |
|:-------------:|:------:|:----:|:---------------:|
| No log        | 0      | 0    | 3.8582          |
| 3.4802        | 0.1008 | 15   | 3.5118          |
| 3.4608        | 0.2017 | 30   | 3.4890          |
| 3.5272        | 0.3025 | 45   | 3.5189          |
| 3.559         | 0.4034 | 60   | 3.5753          |
| 3.5817        | 0.5042 | 75   | 3.6121          |
| 3.6349        | 0.6050 | 90   | 3.6471          |
| 3.68          | 0.7059 | 105  | 3.6721          |
| 3.6597        | 0.8067 | 120  | 3.6970          |
| 3.6462        | 0.9076 | 135  | 3.7068          |
| 3.7009        | 1.0067 | 150  | 3.7213          |
| 3.6717        | 1.1076 | 165  | 3.7313          |
| 3.7631        | 1.2084 | 180  | 3.7338          |
| 3.7535        | 1.3092 | 195  | 3.7346          |
| 3.668         | 1.4101 | 210  | 3.7375          |
| 3.679         | 1.5109 | 225  | 3.7383          |
| 3.6539        | 1.6118 | 240  | 3.7386          |
| 3.6547        | 1.7126 | 255  | 3.7386          |
| 3.7533        | 1.8134 | 270  | 3.7400          |
| 3.6983        | 1.9143 | 285  | 3.7387          |


### Framework versions

- Transformers 4.52.4
- Pytorch 2.7.0+cu126
- Datasets 3.6.0
- Tokenizers 0.21.1