llama-2 code

<p><h1> speechless-code-mistral-7b-v1.0 </h1></p>

Use the following dataset to fine-tune mistralai/Mistral-7B-v0.1 in order to improve the model's reasoning and planning abilities.

Total 201,981 samples.

HumanEval

Metric Value
humaneval-python 50.0

Big Code Models Leaderboard

CodeLlama-34B-Python: 53.29

CodeLlama-34B-Instruct: 50.79

CodeLlama-13B-Instruct: 50.6

CodeLlama-34B: 45.11

CodeLlama-13B-Python: 42.89

CodeLlama-13B: 35.07

lm-evaluation-harness

Open LLM Leaderboard

Metric Value
ARC 59.64
HellaSwag 82.25
MMLU 61.33
TruthfulQA 48.45
Average 62.92

Parameters

lr 2e-4
lr_scheduler_type cosine
weight_decay 0.0
optim paged_adamw_8bit
flash_attention True
rerope False
max_new_tokens 4096
num_train_epochs 2
bits 4
lora_r 64
lora_alpha 16
lora_dropout 0.05
double_quant True
quant_type nf4
dataset_format airoboros
mini_batch_size 2
grandient_accumulation_steps 32
bf16 True

A40-48G x 2

epoch 2.0
etrain_loss 0.5
etrain_runtime 1 day, 10:25:26.77
etrain_samples_per_second 3.194
etrain_steps_per_second 0.025
eeval_loss 0.5146
eeval_runtime 0:00:25.04
eeval_samples_per_second 7.985
eeval_steps_per_second