--- license: cc-by-nc-4.0 base_model: google/gemma-2b-it tags: - generated_from_trainer - axolotl - gemma - instruct - finetune - chatml - gpt4 - synthetic data - distillation model-index: - name: gemma-2b-openhermes results: [] datasets: - mlabonne/chatml-OpenHermes2.5-dpo-binarized-alpha language: - en library_name: transformers pipeline_tag: text-generation --- # gemma-2b-openhermes ![image/jpeg](https://cdn-uploads.huggingface.co/production/uploads/64e380b2e12618b261fa6ba0/9bmxL8Lt7hBaKlKHVxtew.jpeg) gemma-2b-openhermes is a variant of the Gemma 2B language model, which has been further fine-tuned on the OpenHermes-2.5 preference dataset using QLoRA. * [google/gemma-2b-it](https://huggingface.co/google/gemma-2b-it) * [mlabonne/chatml-OpenHermes2.5-dpo-binarized-alpha](https://huggingface.co/datasets/mlabonne/chatml-OpenHermes2.5-dpo-binarized-alpha)
## Usage ### Chat Template The instruction-tuned models use a chat template that must be adhered to for conversational use. The easiest way to apply it is using the tokenizer's built-in chat template, as shown in the following snippet. Let's load the model and apply the chat template to a conversation. In this example, we'll start with a single user interaction: ```py from transformers import AutoTokenizer, AutoModelForCausalLM import transformers import torch model_id = "abideen/gemma-2b-openhermes" dtype = torch.bfloat16 tokenizer = AutoTokenizer.from_pretrained(model_id) model = AutoModelForCausalLM.from_pretrained( model_id, device_map="cuda", torch_dtype=dtype, ) chat = [{ "role": "user", "content": "What is a Language Model?" }] prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True) ``` After the prompt is ready, generation can be performed like this: ```py inputs = tokenizer.encode(prompt, add_special_tokens=True, return_tensors="pt") outputs = model.generate(input_ids=inputs.to(model.device), max_new_tokens=250) print(tokenizer.decode(outputs[0])) ``` ### Inputs and outputs * **Input:** Text string, such as a question, a prompt, or a document to be summarized. * **Output:** Generated English-language text in response to the input, such as an answer to a question, or a summary of a document. ## 🏆 Evaluation results # Nous Benchmark Agieval | Task | Version | Metric | Value | | StdErr | |-------------------------------------------|---------|--------|-------|---|---------| | agieval\_aqua\_rat | 0 | acc | 24.02 | _ | 2.69 | | agieval\_aqua\_rat | 0 | acc\_norm | 24.02 | _ | 2.69 | | agieval\_logiqa\_en | 0 | acc | 23.20 | _ | 1.66 | | agieval\_logiqa\_en | 0 | acc\_norm | 24.42 | _ | 1.69 | | agieval\_lsat\_ar | 0 | acc | 18.26 | _ | 2.55 | | agieval\_lsat\_ar | 0 | acc\_norm | 18.70 | _ | 2.58 | | agieval\_lsat\_lr | 0 | acc | 22.35 | _ | 1.85 | | agieval\_lsat\_lr | 0 | acc\_norm | 23.53 | _ | 1.88 | | agieval\_lsat\_rc | 0 | acc | 20.82 | _ | 2.48 | | agieval\_lsat\_rc | 0 | acc\_norm | 20.07 | _ | 2.45 | | agieval\_sat\_en | 0 | acc | 32.52 | _ | 3.27 | | agieval\_sat\_en | 0 | acc\_norm | 32.52 | _ | 3.27 | | agieval\_sat\_en\_without\_passage | 0 | acc | 25.73 | _ | 3.05 | | agieval\_sat\_en\_without\_passage | 0 | acc\_norm | 24.27 | _ | 2.99 | | agieval\_sat\_math | 0 | acc | 25.00 | _ | 2.93 | | agieval\_sat\_math | 0 | acc\_norm | 20.91 | _ | 2.75 | Average: 24.11 GPT4ALL | Task | Version | Metric | Value | | StdErr | |----------------------|---------|--------|-------|---|---------| | arc\_challenge | 0 | acc | 21.77 | _ | 1.21 | | arc\_challenge | 0 | acc\_norm | 24.15 | _ | 1.25 | | arc\_easy | 0 | acc | 37.37 | _ | 0.99 | | arc\_easy | 0 | acc\_norm | 36.95 | _ | 0.99 | | boolq | 1 | acc | 65.60 | _ | 0.83 | | hellaswag | 0 | acc | 34.54 | _ | 0.47 | | hellaswag | 0 | acc\_norm | 40.54 | _ | 0.49 | | openbookqa | 0 | acc | 15.00 | _ | 1.59 | | openbookqa | 0 | acc\_norm | 27.40 | _ | 2.00 | | piqa | 0 | acc | 60.88 | _ | 1.14 | | piqa | 0 | acc\_norm | 60.55 | _ | 1.14 | | winogrande | 0 | acc | 50.91 | _ | 1.41 | Average: 40.01 BigBench | Task | Version | Metric | Value | Std Err | |-----------------------------------|---------|--------|--------|---------| | bigbench\_causal\_judgement | 0 | MCG | 50 | 2.26 | | bigbench\_date\_understanding | 0 | MCG | 49.14 | 2.18 | | bigbench\_disambiguation\_qa | 0 | MCG | 49.31 | 2.74 | | bigbench\_geometric\_shapes | 0 | MCG | 14.18 | 1.37 | | bigbench\_logical\_deduction\_5objs | 0 | MCG | 49.41 | 2.73 | | bigbench\_logical\_deduction\_7objs | 0 | MCG | 41.48 | 2.46 | | bigbench\_logical\_deduction\_3objs | 0 | MCG | 69.33 | 2.75 | | bigbench\_movie\_recommendation | 0 | MCG | 51.71 | 2.25 | | bigbench\_navigate | 0 | MCG | 50 | 1.58 | | bigbench\_reasoning\_colored\_obj | 0 | MCG | 51.92 | 0.99 | | bigbench\_ruin\_names | 0 | MCG | 48.14 | 2.01 | | bigbench\_salient\_trans\_err\_detec | 0 | MCG | 39.92 | 1.2 | | bigbench\_snarks | 0 | MCG | 64.14 | 3.71 | | bigbench\_sports\_understanding | 0 | MCG | 55.31 | 1.59 | | bigbench\_temporal\_sequences | 0 | MCG | 46.92 | 1.4 | | bigbench\_tsk\_shuff\_objs\_5 | 0 | MCG | 25.04 | 1.01 | | bigbench\_tsk\_shuff\_objs\_7 | 0 | MCG | 15.04 | 0.72 | | bigbench\_tsk\_shuff\_objs\_3 | 0 | MCG | 55.33 | 2.75 | Average: 44.75 TruthfulQA | Task | Version | Metric | Value | Std Err | |----------------------------------|---------|--------|--------|----------| | truthfulqa\_mc | 1 | mc1 | 30.11 | 1.61 | | truthfulqa\_mc | 1 | mc2 | 47.69 | 1.61 | Average: 38.90 # Openllm Benchmark | Task |Version| Metric |Value| |Stderr| |-------------|------:|--------|----:|---|-----:| |arc_challenge| 0|acc |40.44|± | 1.43| | | |acc_norm|43.81|± | 1.34| |hellaswag | 0|acc |48.1 |± | 0.45| | | |acc_norm|62.73|± | 0.32| |gsm8k | 0|acc |5.6 |± | 0.6 | |winogrande | 0|acc |60.91|± | 1.3 | |mmlu | 0|acc |37.62 |±| 0.6 | Average: 73.5% ### TruthfulQA | Task |Version|Metric|Value| |Stderr| |-------------|------:|------|----:|---|-----:| |truthfulqa_mc| 1|mc1 |29.00|± | 1.58| | | |mc2 |45.83|± | 1.59| ### Training hyperparameters The following hyperparameters were used during training: - learning_rate: 5e-07 - train_batch_size: 1 - eval_batch_size: 8 - seed: 42 - gradient_accumulation_steps: 8 - total_train_batch_size: 8 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 - lr_scheduler_type: cosine - lr_scheduler_warmup_steps: 100 - training_steps: 1300 ### 📝 Axolotl Configuration ```yaml base_model: google/gemma-2b-it model_type: GemmaForCausalLM tokenizer_type: GemmaTokenizer trust_remote_code: true load_in_8bit: false load_in_4bit: true strict: false rl: dpo chat_template: chatml datasets: - path: mlabonne/chatml-OpenHermes2.5-dpo-binarized-alpha split: train type: chatml.intel dataset_prepared_path: val_set_size: 0.01 output_dir: ./out adapter: qlora lora_model_dir: sequence_len: 1800 sample_packing: false pad_to_sequence_len: false lora_r: 16 lora_alpha: 16 lora_dropout: 0.05 lora_target_linear: true lora_fan_in_fan_out: lora_target_modules: wandb_project: gemma wandb_entity: wandb_watch: wandb_name: wandb_log_model: gradient_accumulation_steps: 8 micro_batch_size: 1 num_epochs: 1 optimizer: paged_adamw_32bit lr_scheduler: cosine learning_rate: 5e-7 train_on_inputs: false group_by_length: false bf16: true fp16: false tf32: true gradient_checkpointing: true early_stopping_patience: resume_from_checkpoint: local_rank: logging_steps: 1 xformers_attention: flash_attention: false warmup_steps: 100 evals_per_epoch: 1 eval_table_size: eval_table_max_new_tokens: 128 save_steps: 1000 max_steps: 1300 debug: deepspeed: weight_decay: 0.0 fsdp: fsdp_config: special_tokens: ``` ### Framework versions - Transformers 4.39.0.dev0 - Pytorch 2.1.2+cu118 - Datasets 2.17.0 - Tokenizers 0.15.0 - axolotl: 0.4.0 [Built with Axolotl](https://github.com/OpenAccess-AI-Collective/axolotl)