openhermes-gemma-2b-it

openhermes-gemma-2b-it is a variant of the Gemma 2B language model, which has been further fine-tuned on the OpenHermes-2.5 preference dataset using QLoRA. This fine-tuning process enhances the model's ability to understand and generate responses that align with user preferences in conversational settings.

• google/gemma-2b-it
• 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:

from transformers import AutoTokenizer, AutoModelForCausalLM
import transformers
import torch

model_id = "Syed-Hasan-8503/openhermes-gemma-2b-it"
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 Machine Learning?" }]
prompt = tokenizer.apply_chat_template(chat, tokenize=False, add_generation_prompt=True)

After the prompt is ready, generation can be performed like this:

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 data

🏆 Evaluation

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: 23.8

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: 39.9

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: 44.7

TRUTHFULQA

Task	Version	Metric	Value	Std Err
truthfulqa_mc	1	mc1	30.11	1.61
truthfulqa_mc	1	mc2	47.69	1.61

Average:38.9

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: 1000

📝 Axolotl Configuration

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: axolotl-gemma-dpo
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: 1000
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