metadata
language:
- bg
- ca
- cs
- da
- de
- en
- es
- fr
- hr
- hu
- it
- nl
- pl
- pt
- ro
- ru
- sl
- sr
- sv
- uk
license: apache-2.0
library_name: transformers
datasets:
- Open-Orca/OpenOrca
- OpenAssistant/oasst_top1_2023-08-25
model-index:
- name: Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v3
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: 60.58
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v3
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: 83.34
name: normalized accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v3
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: 61.53
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v3
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: 48.21
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v3
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: 77.74
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v3
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: 36.16
name: accuracy
source:
url: >-
https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v3
name: Open LLM Leaderboard
reference-data-model:
datasets:
- OpenAssistant/oasst_top1_2023-08-25:
lang: "bg,ca,cs,da,de,en,es,fr,hr,hu,it,nl,pl,pt,ro,ru,sl,sr,sv,uk"
link: https://huggingface.co/datasets/OpenAssistant/oasst_top1_2023-08-25
model:
- Open-Orca/Mistral-7B-OpenOrca
Link:
https://huggingface.co/Open-Orca/Mistral-7B-OpenOrca
100 examples of generating:
- Link:
https://huggingface.co/NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v3/blob/main/output.xlsx
Activated training with:
- Link:
https://huggingface.co/blog/tomaarsen/attention-sinks
https://github.com/tomaarsen/attention_sinks
https://arxiv.org/abs/2309.17453
TRL:
- Link:
https://huggingface.co/docs/trl/index
https://huggingface.co/docs/trl/sft_trainer
flash-attention:
- Link:
https://github.com/Dao-AILab/flash-attention
https://arxiv.org/abs/2205.14135
Version:
- Link:
https://huggingface.co/NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v1
https://huggingface.co/NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v2
https://huggingface.co/NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v3
Eval model:
- link:
https://huggingface.co/datasets/open-llm-leaderboard/details_NickyNicky__Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v3
# attention-sinks
pip install attention_sinks
# flash-attn
!export CUDA_HOME=/usr/local/cuda-11.8
!MAX_JOBS=4 pip install flash-attn --no-build-isolation -qqq
!pip install git+"https://github.com/HazyResearch/flash-attention.git#subdirectory=csrc/rotary" -qqq
Version
import torch, transformers,torchvision
torch.__version__,transformers.__version__, torchvision.__version__
#OUTPUTS: ('2.0.1+cu118', '4.34.0', '0.15.2+cu118')
How to use
from transformers import (
AutoModelForCausalLM,
AutoTokenizer,
BitsAndBytesConfig,
HfArgumentParser,
TrainingArguments,
pipeline,
logging,
GenerationConfig,
TextIteratorStreamer,
)
from attention_sinks import AutoModelForCausalLM
import torch
# model_id = 'Open-Orca/Mistral-7B-OpenOrca'
model_id='NickyNicky/Mistral-7B-OpenOrca-oasst_top1_2023-08-25-v3'
model = AutoModelForCausalLM.from_pretrained(model_id,
device_map="auto",
trust_remote_code=True,
torch_dtype=torch.bfloat16,
load_in_4bit=True,
low_cpu_mem_usage= True,
#use_flash_attention_2=True, #GPU A100 or GPU supported
attention_sink_size=4,
attention_sink_window_size=1024, #512, # <- Low for the sake of faster generation
)
max_length=2048
print("max_length",max_length)
tokenizer = AutoTokenizer.from_pretrained(model_id,
# use_fast = False,
max_length=max_length,)
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = 'right'
#EXAMPLE #1
txt="""<|im_start|>user
I'm looking for an efficient Python script to output prime numbers. Can you help me out? I'm interested in a script that can handle large numbers and output them quickly. Also, it would be great if the script could take a range of numbers as input and output all the prime numbers within that range. Can you generate a script that fits these requirements? Thanks!<|im_end|>
<|im_start|>assistant
"""
#EXAMPLE #2
txt="""<|im_start|>user
Estoy desarrollando una REST API con Nodejs, y estoy tratando de aplicar algún sistema de seguridad, ya sea con tokens o algo similar, me puedes ayudar?<|im_end|>
<|im_start|>assistant
"""
inputs = tokenizer.encode(txt, return_tensors="pt").to("cuda")
generation_config = GenerationConfig(
max_new_tokens=max_new_tokens,
temperature=0.7,
top_p=0.9,
top_k=len_tokens,
repetition_penalty=1.11,
do_sample=True,
# pad_token_id=tokenizer.eos_token_id,
# eos_token_id=tokenizer.eos_token_id,
# use_cache=True,
# stopping_criteria= StoppingCriteriaList([stopping_criteria]),
)
outputs = model.generate(generation_config=generation_config,
input_ids=inputs,)
tokenizer.decode(outputs[0], skip_special_tokens=False) #True
Open LLM Leaderboard Evaluation Results
Detailed results can be found here
| Metric | Value |
|---|---|
| Avg. | 61.26 |
| AI2 Reasoning Challenge (25-Shot) | 60.58 |
| HellaSwag (10-Shot) | 83.34 |
| MMLU (5-Shot) | 61.53 |
| TruthfulQA (0-shot) | 48.21 |
| Winogrande (5-shot) | 77.74 |
| GSM8k (5-shot) | 36.16 |