metadata
license: other
library_name: transformers
tags:
- chatml
- finetune
- gpt4
- synthetic data
- custom_code
- qwen2
datasets:
- Locutusque/Hercules-v3.0
license_name: tongyi-qianwen-research
license_link: https://huggingface.co/Qwen/Qwen1.5-1.8B-Chat/raw/main/LICENSE
model-index:
- name: Reyna-Mini-1.8B-v0.2
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: 36.6
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=aloobun/Reyna-Mini-1.8B-v0.2
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: 60.19
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=aloobun/Reyna-Mini-1.8B-v0.2
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: 44.75
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=aloobun/Reyna-Mini-1.8B-v0.2
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=aloobun/Reyna-Mini-1.8B-v0.2
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: 61.56
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=aloobun/Reyna-Mini-1.8B-v0.2
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: 31.31
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=aloobun/Reyna-Mini-1.8B-v0.2
name: Open LLM Leaderboard
- Finetuned Qwen/Qwen1.5-1.8B-Chat, with SFT on Hercules v3 dataset.
- This marks the third model in this series.
- Format: ChatML -
<|im_start|>system {system}<|im_end|> <|im_start|>user {prompt}<|im_end|> <|im_start|>assistant - Next step would be to do a DPO train on top.
Benchamrks:
| Avg. | Arc | HellaSwag | MMLU | TruthfulQA | Winogrande | GSM8K |
|---|---|---|---|---|---|---|
| 45.94 | 36.6 | 60.19 | 44.75 | 41.24 | 61.56 | 31.31 |
Example:
from transformers import AutoModelForCausalLM, AutoTokenizer, TextStreamer, StoppingCriteria
import torch
class MyStoppingCriteria(StoppingCriteria):
def __init__(self, target_sequence, prompt):
self.target_sequence = target_sequence
self.prompt=prompt
def __call__(self, input_ids, scores, **kwargs):
generated_text = tokenizer.decode(input_ids[0])
generated_text = generated_text.replace(self.prompt,'')
if self.target_sequence in generated_text:
return True
return False
def __len__(self):
return 1
def __iter__(self):
yield self
modelpath="aloobun/Reyna-Mini-1.8B-v0.2"
model = AutoModelForCausalLM.from_pretrained(
modelpath,
torch_dtype=torch.bfloat16,
device_map="cuda",
trust_remote_code=True,
)
tokenizer = AutoTokenizer.from_pretrained(
modelpath,
trust_remote_code=True,
use_fast=False,
)
prompt = "<|im_start|>system\nYou are a helpful assistant.<|im_end|>\n<|im_start|>user\nIs there inherent order in nature or is it all chaos and chance?<|im_end|>\n<|im_start|>assistant\n"
encoded_input = tokenizer(prompt, return_tensors='pt')
input_ids=encoded_input['input_ids'].cuda()
streamer = TextStreamer(tokenizer=tokenizer, skip_prompt=True)
op = model.generate(
input_ids,
streamer=streamer,
pad_token_id=tokenizer.eos_token_id,
do_sample=True,
temperature=0.6,
top_p=0.8,
max_new_tokens=512,
stopping_criteria=MyStoppingCriteria("<|im_end|>", prompt)
)
Output:
Nature appears to be inherently organized, with patterns and structures that can be observed across different levels of organization. However, the exact mechanisms by which these patterns emerge and evolve remain largely unknown. The universe seems to be governed by a series of laws and principles known as "laws of physics," such as Newton's laws of motion, electromagnetism, and thermodynamics. These laws govern how matter and energy interact with each other and how they behave over time. Despite our understanding of these laws, we still struggle to comprehend the underlying mechanisms that allow for the emergence of complex patterns and structures. This is because the universe operates on a scale that is too small for us to observe directly, and therefore we cannot fully understand its internal workings. In summary, while there may be some level of order and structure within the universe, the precise mechanisms governing this order remain largely unknown.<|im_end|>
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 45.94 |
| AI2 Reasoning Challenge (25-Shot) | 36.60 |
| HellaSwag (10-Shot) | 60.19 |
| MMLU (5-Shot) | 44.75 |
| TruthfulQA (0-shot) | 41.24 |
| Winogrande (5-shot) | 61.56 |
| GSM8k (5-shot) | 31.31 |