metadata

license: other
library_name: transformers
datasets:
  - vicgalle/alpaca-gpt4
extra_gated_heading: Access Gemma on Hugging Face
extra_gated_prompt: >-
  To access Gemma on Hugging Face, you’re required to review and agree to
  Google’s usage license. To do this, please ensure you’re logged-in to Hugging
  Face and click below. Requests are processed immediately.
extra_gated_button_content: Acknowledge license
license_name: gemma-terms-of-use
license_link: https://ai.google.dev/gemma/terms
base_model:
  - google/gemma-2b
model-index:
  - name: Gemmalpaca-2B
    results:
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: AI2 Reasoning Challenge (25-Shot)
          type: ai2_arc
          config: ARC-Challenge
          split: test
          args:
            num_few_shot: 25
        metrics:
          - type: acc_norm
            value: 48.72
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/Gemmalpaca-2B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: HellaSwag (10-Shot)
          type: hellaswag
          split: validation
          args:
            num_few_shot: 10
        metrics:
          - type: acc_norm
            value: 71.36
            name: normalized accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/Gemmalpaca-2B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: MMLU (5-Shot)
          type: cais/mmlu
          config: all
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 36.3
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/Gemmalpaca-2B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: TruthfulQA (0-shot)
          type: truthful_qa
          config: multiple_choice
          split: validation
          args:
            num_few_shot: 0
        metrics:
          - type: mc2
            value: 41.24
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/Gemmalpaca-2B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: Winogrande (5-shot)
          type: winogrande
          config: winogrande_xl
          split: validation
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 65.59
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/Gemmalpaca-2B
          name: Open LLM Leaderboard
      - task:
          type: text-generation
          name: Text Generation
        dataset:
          name: GSM8k (5-shot)
          type: gsm8k
          config: main
          split: test
          args:
            num_few_shot: 5
        metrics:
          - type: acc
            value: 10.69
            name: accuracy
        source:
          url: >-
            https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=mlabonne/Gemmalpaca-2B
          name: Open LLM Leaderboard

Gemmalpaca-2B

This is gemma-2b model supervised fine-tuned on the vicgalle/alpaca-gpt4 dataset. It outperforms gemma-2b-it, Google's chat version, on Nous' benchmark suite.

It's mostly a test to see how fine-tuning works with Gemma models on a well-known dataset. It turned out better than expected. :)

🔍 Applications

This model has a context length of 8k. I recommend using it with the Alpaca chat template and NOT the Gemma Instruct template (works perfectly with LM Studio). You also want to add </s> as a stop token.

⚡ Quantized models

GGUF: https://huggingface.co/mlabonne/Gemmalpaca-2B-GGUF

🏆 Evaluation

Nous

Gemmalpaca-2B outperforms gemma-2b and gemma-2b-it on Nous' benchmark suite (evaluation performed using LLM AutoEval). See the entire leaderboard here.

Model	Average	AGIEval	GPT4All	TruthfulQA	Bigbench
mlabonne/Gemmalpaca-2B 📄	38.39	24.48	51.22	47.02	30.85
google/gemma-2b-it 📄	36.1	23.76	43.6	47.64	29.41
google/gemma-2b 📄	34.26	22.7	43.35	39.96	31.03

🧩 Configuration

It was trained using Axolotl with the following configuration.

base_model: alpindale/gemma-2b
model_type: GemmaForCausalLM
tokenizer_type: GemmaTokenizer

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: vicgalle/alpaca-gpt4
    type: alpaca

dataset_prepared_path:
val_set_size: 0.01
output_dir: ./out

sequence_len: 2048
sample_packing: true
pad_to_sequence_len: true

adapter: qlora
lora_model_dir:
lora_r: 32
lora_alpha: 64
lora_dropout: 0.05
lora_target_linear: true

wandb_project: axolotl
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 3
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention:

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_table_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.1
fsdp:
fsdp_config:
special_tokens:
  bos_token: <s>
  eos_token: </s>
  unk_token: <unk>

Open LLM Leaderboard Evaluation Results

Detailed results can be found here

Metric	Value
Avg.	45.65
AI2 Reasoning Challenge (25-Shot)	48.72
HellaSwag (10-Shot)	71.36
MMLU (5-Shot)	36.30
TruthfulQA (0-shot)	41.24
Winogrande (5-shot)	65.59
GSM8k (5-shot)	10.69