The model is fully finetuned on translated datasets using axolotl. These datasets primarily consist of translated versions sourced from teknium/OpenHermes-2.5 and the Open-Orca/SlimOrca datasets.

See axolotl config

axolotl version: 0.4.0

yaml

base_model: mistralai/Mistral-7B-v0.1
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer

load_in_8bit: false
load_in_4bit: false
strict: false

datasets:
  - path: NovusResearch/OpenHermes-2.5-Translated-TR-sharegpt-style
    type: sharegpt
    conversation: chatml
  - path: data/merged_all.json
    ds_type: json
    type: sharegpt
    conversation: chatml

dataset_prepared_path:
val_set_size: 0.05
output_dir: ./out

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true
eval_sample_packing: false

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 2
num_epochs: 2
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.000005

train_on_inputs: false
group_by_length: false
bf16: auto
fp16:
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

## Use 
wandb_project: full_finetune
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:


warmup_steps: 10
evals_per_epoch: 0
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "<|im_end|>"
  unk_token: "<unk>"
tokens:
  - "<|im_start|>"

🎯 OpenLLMTurkishLeaderboard

Metric	Value
Avg.	36.41
AI2 Reasoning Challenge	27.24
HellaSwag	33.93
MMLU	40.64
TruthfulQA	47.90
Winogrande	50.86
GSM8k	17.91

NovusResearch/Thestral-0.1-tr-chat-7B

Thestral-0.1-tr-chat-7B

🎯 OpenLLMTurkishLeaderboard

NovusResearch/
Thestral-0.1-tr-chat-7B