File size: 6,757 Bytes
066f3f7 b93fc62 f6a95bf b93fc62 f1b1f9d 7dc1770 23f7316 86dcaaa 1ea5e91 e090456 1ea5e91 e090456 1ea5e91 e090456 1ea5e91 e090456 1ea5e91 e090456 1ea5e91 b93fc62 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 |
---
language:
- en
license: apache-2.0
tags:
- merge
pipeline_tag: text-generation
dtype: bfloat16
model-index:
- name: MistralTrix-v1
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: 72.27
name: normalized accuracy
source:
url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=CultriX/MistralTrix-v1
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: 88.33
name: normalized accuracy
source:
url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=CultriX/MistralTrix-v1
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: 65.24
name: accuracy
source:
url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=CultriX/MistralTrix-v1
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: 70.73
source:
url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=CultriX/MistralTrix-v1
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: 80.98
name: accuracy
source:
url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=CultriX/MistralTrix-v1
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: 62.77
name: accuracy
source:
url: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard?query=CultriX/MistralTrix-v1
name: Open LLM Leaderboard
---
# EDIT:
Always check my space for the latest benchmark results for my models!
* https://huggingface.co/spaces/CultriX/Yet_Another_LLM_Leaderboard
# Results:
T: 🟦
Model: CultriX/MistralTrix-v1 📑
Average: 73.39
ARC: 72.27
HellaSwag: 88.33
MMLU: 65.24
TruthfulQA: 70.73
Winogrande: 80.98
GSM8K: 62.77
# Edit/Disclaimer:
Currently the #1 ranked 7B LLM on the LLM Leaderboards, woah!
I did not expect that result at all and am in no way a professional when it comes to LLM's or computer science in general,
just a guy that likes to nerd about and tinker around.
For those wondering how I achieved this, the answer is that I simply attempted to apply the techniques outlined in this amazing article myself: https://towardsdatascience.com/fine-tune-a-mistral-7b-model-with-direct-preference-optimization-708042745aac
Therefore, all credit basically goes to the guy who wrote that.
He offers the exact Colab notebook I used to train this model for free, as well as a really nice GitHub page I hope he doesn't mind me sharing: https://github.com/mlabonne/llm-course/
So huge thank you to him for sharing his knowledge and learning me a thing or two in the process!
# GGUF
I attempted to quantisize the model myself, which again I pretty much have no clue about, but it seems to run fine for me when I test them:
https://huggingface.co/CultriX/MistralTrix-v1-GGUF
I'll say it one more time though:
"I am a complete beginner to all of this, so if these do end up sucking don't be surprised."
You have been warned :)
# Description:
(trained on a single Colab GPU in less than a few hours)
MistralTrix-v1 is an zyh3826/GML-Mistral-merged-v1 model that has been further fine-tuned with Direct Preference Optimization (DPO) using Intel's dataset for neural-chat-7b-v3-1.
It surpasses the original model on several benchmarks (see results).
It is directly inspired by the RLHF process described by Intel/neural-chat-7b-v3-1's authors to improve performance.
I used the same dataset and reformatted it to apply the ChatML template.
The code to train this model is available on Google Colab and GitHub.
Fine-tuning took about an hour on Google Colab A-1000 GPU with 40GB VRAM.
# TRAINING SPECIFICATIONS
> LoRA configuration
peft_config = LoraConfig(
r=16,
lora_alpha=16,
lora_dropout=0.05,
bias="none",
task_type="CAUSAL_LM",
target_modules=['k_proj', 'gate_proj', 'v_proj', 'up_proj', 'q_proj', 'o_proj', 'down_proj']
)
> Model to fine-tune
model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype=torch.float16,
load_in_4bit=True
)
model.config.use_cache = False
> Reference model
ref_model = AutoModelForCausalLM.from_pretrained(
model_name,
torch_dtype=torch.float16,
load_in_4bit=True
)
> Training arguments
training_args = TrainingArguments(
per_device_train_batch_size=4,
gradient_accumulation_steps=4,
gradient_checkpointing=True,
learning_rate=5e-5,
lr_scheduler_type="cosine",
max_steps=200,
save_strategy="no",
logging_steps=1,
output_dir=new_model,
optim="paged_adamw_32bit",
warmup_steps=100,
bf16=True,
report_to="wandb",
)
> Create DPO trainer
dpo_trainer = DPOTrainer(
model,
ref_model,
args=training_args,
train_dataset=dataset,
tokenizer=tokenizer,
peft_config=peft_config,
beta=0.1,
max_prompt_length=1024,
max_length=1536,
)
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_CultriX__MistralTrix-v1)
| Metric |Value|
|---------------------------------|----:|
|Avg. |73.39|
|AI2 Reasoning Challenge (25-Shot)|72.27|
|HellaSwag (10-Shot) |88.33|
|MMLU (5-Shot) |65.24|
|TruthfulQA (0-shot) |70.73|
|Winogrande (5-shot) |80.98|
|GSM8k (5-shot) |62.77|
|