metadata
license: apache-2.0
language:
- ru
tags:
- generated_from_trainer
base_model: WlappaAI/Mistral-7B-wikipedia_ru_pruned-0.1_merged
model-index:
- name: dracor-ru-small-lora_merged
results: []
See axolotl config
axolotl version: 0.4.0
base_model: WlappaAI/Mistral-7B-wikipedia_ru_pruned-0.1_merged
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true
load_in_8bit: true
load_in_4bit: false
strict: false
datasets:
- path: ./datasets/ru-dracor
type: completion
field: text
dataset_prepared_path: last_run_prepared
val_set_size: 0.05
output_dir: ./models/output/dracor_ru_lora
adapter: lora
lora_model_dir:
sequence_len: 1024
sample_packing: true
pad_to_sequence_len: true
lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:
- gate_proj
- down_proj
- up_proj
- q_proj
- v_proj
- k_proj
- o_proj
wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:
gradient_accumulation_steps: 1
micro_batch_size: 6
num_epochs: 1
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 0.0002
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:
xformers_attention:
flash_attention: true
loss_watchdog_threshold: 5.0
loss_watchdog_patience: 3
warmup_steps: 10
evals_per_epoch: 1
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
dracor-ru-small-lora_merged
This model is a Q8_0 GGUF merge of WlappaAI/dracor-ru-small-lora together with WlappaAI/Mistral-7B-wikipedia_ru_pruned-0.1_merged. It's trained on Russian DraCor dataset. It achieves the following results on the evaluation set:
- Loss: 1.1876
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.0002
- train_batch_size: 6
- eval_batch_size: 6
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_steps: 10
- num_epochs: 1
Training results
Training Loss | Epoch | Step | Validation Loss |
---|---|---|---|
1.7921 | 1.0 | 1056 | 1.6606 |
Framework versions
- PEFT 0.10.0
- Transformers 4.40.0.dev0
- Pytorch 2.2.2+cu121
- Datasets 2.18.0
- Tokenizers 0.15.0
- GGUF 0.9.0