---
library_name: peft
tags:
- generated_from_trainer
model-index:
- name: qlora-yi-34b-200k-aezakmi-v2-rawrr-v1-run1
  results: []
---


[<img src="https://raw.githubusercontent.com/OpenAccess-AI-Collective/axolotl/main/image/axolotl-badge-web.png" alt="Built with Axolotl" width="200" height="32"/>](https://github.com/OpenAccess-AI-Collective/axolotl)
<details><summary>See axolotl config</summary>

axolotl version: `0.3.0`
```yaml
base_model: ./yi-34b-rawrr-dpo-1
base_model_config: ./yi-34b-rawrr-dpo-1
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: false
is_llama_derived_model: true

load_in_8bit: false
load_in_4bit: true
bnb_4bit_use_double_quant: true
buse_double_quants: true
bnb_4bit_compute_dtype: torch.bfloat16
torch_dtype: bf16
strict: false
datasets:
  - path: /run/media/..../aezakmi_v2/aezakmi_v2_draft2.jsonl
    type: alpaca_w_system2.load_open_orca_chatml
    conversation: chatml
dataset_prepared_path: last_run_prepared
val_set_size: 0.01
adapter: qlora
lora_model_dir:
sequence_len: 1400
sample_packing: true
lora_r: 16
lora_alpha: 32
lora_dropout: 0.05
lora_target_modules:
  - q_proj
  - v_proj
  - k_proj
  - o_proj
  - gate_proj
  - down_proj
  - up_proj
lora_target_linear: true
lora_fan_in_fan_out:
wandb_project:
wandb_watch:
wandb_run_id:
wandb_log_model:
output_dir: ./qlora-yi-34b-200k-aezakmi-v2-rawrr-v1-run1
pad_to_sequence_len: false
micro_batch_size: 1
gradient_accumulation_steps: 1
num_epochs: 2.4
optimizer: adamw_bnb_8bit
torchdistx_path:
lr_scheduler: constant
learning_rate: 0.00005
train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: false
bfloat16: true
flash_optimum: false
gradient_checkpointing: true
early_stopping_patience:
save_safetensors:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
save_total_limit: 10
deepspeed:
seed: 42
warmup_steps: 100
eval_steps: 5000000
save_steps: 500
eval_table_size: 
eval_table_max_new_tokens:
debug:
weight_decay:
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<|startoftext|>"
  eos_token: "<|endoftext|>"
  unk_token: "<unk>"

```

</details><br>

# qlora-yi-34b-200k-aezakmi-v2-rawrr-v1-run1

This LoRA has been trained from [adamo1139/yi-34b-200k-rawrr-dpo-1](https://huggingface.co/adamo1139/yi-34b-200k-rawrr-dpo-1).
If you want to re-create the model from this, first get yourelf Yi-34B-200K llama-fied, then merge in LoRA [adamo1139/Yi-34B-200K-rawrr1-LORA-DPO-experimental-r2](https://huggingface.co/adamo1139/Yi-34B-200K-rawrr1-LORA-DPO-experimental-r2) \
Then, merge in this lora.

It's all pretty experimental still. I think I will re-run DPO training on rawrr with maybe a bit longer context and with higher lr or epoch count. I want even stronger anti-refusal effect.
It's still better in this regard than Yi-34B-200K-AEZAKMI-v2, but it's not perfect.

## Model description

More information needed

## Intended uses & limitations

More information needed

## Training and evaluation data

More information needed

## Training procedure


The following `bitsandbytes` quantization config was used during training:
- quant_method: bitsandbytes
- load_in_8bit: False
- load_in_4bit: True
- llm_int8_threshold: 6.0
- llm_int8_skip_modules: None
- llm_int8_enable_fp32_cpu_offload: False
- llm_int8_has_fp16_weight: False
- bnb_4bit_quant_type: nf4
- bnb_4bit_use_double_quant: True
- bnb_4bit_compute_dtype: bfloat16

### Training hyperparameters

The following hyperparameters were used during training:
- learning_rate: 5e-05
- train_batch_size: 1
- eval_batch_size: 1
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: constant
- lr_scheduler_warmup_steps: 100
- num_epochs: 2.4

### Training results



### Framework versions

- PEFT 0.7.0
- Transformers 4.37.0.dev0
- Pytorch 2.0.1+cu118
- Datasets 2.15.0
- Tokenizers 0.15.0