See axolotl config

axolotl version: 0.3.0

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

load_in_8bit: true
load_in_4bit: false
strict: false

datasets:
  - path: datasets/norobots_150/norobots_150
    type: completion
  - path: datasets/separated/bloke-separate
    type: completion
  - path: datasets/separated/kcpp-separate
    type: completion
  - path: datasets/separated/kcpp-support-separate
    type: completion
  - path: datasets/separated/st-chat-separate
    type: completion
  - path: datasets/separated/exllama2_readme.txt
    type: completion
  - path: datasets/separated/koboldcpp_readme.txt
    type: completion
  - path: datasets/separated/llama_readme.txt
    type: completion
  - path: datasets/separated/ooba_readme.txt
    type: completion
  - path: datasets/separated/sillytavern_readme.txt
    type: completion
  - path: datasets/separated/sillytavern_simple_setup_guide.txt
    type: completion
  - path: datasets/transformer_article.txt
    type: completion
  - path: datasets/lmg_thread.txt
    type: completion

dataset_prepared_path: last_run_prepared
val_set_size: 0.1
output_dir: ./llmTechChat-lora

adapter: lora
lora_model_dir:

chat_template: chatml

sequence_len: 8192
sample_packing: true
pad_to_sequence_len: true

lora_r: 128
lora_alpha: 64
lora_dropout: 0.20
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: llmTechChat

gradient_accumulation_steps: 8
micro_batch_size: 1
num_epochs: 4
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 0.0003

train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: false

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

loss_watchdog_threshold: 5.0
loss_watchdog_patience: 3

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_table_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"

save_safetensors: true

llmTechChat-lora

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 1.9365

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: True
• load_in_4bit: False
• 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: fp4
• bnb_4bit_use_double_quant: False
• bnb_4bit_compute_dtype: float32

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 10
• num_epochs: 4

Training results

Training Loss	Epoch	Step	Validation Loss
4.3577	0.01	1	4.3261
2.0615	0.25	40	2.0476
1.9905	0.5	80	1.9691
1.8699	0.75	120	1.9344
1.9604	1.0	160	1.9111
1.7684	1.23	200	1.8978
1.7673	1.48	240	1.8809
1.7296	1.73	280	1.8630
1.7737	1.98	320	1.8479
1.5871	2.22	360	1.8883
1.5339	2.47	400	1.8761
1.5589	2.72	440	1.8657
1.5651	2.96	480	1.8590
1.3134	3.2	520	1.9497
1.3423	3.45	560	1.9406
1.3635	3.7	600	1.9362
1.3235	3.95	640	1.9365

Framework versions

• PEFT 0.7.0
• Transformers 4.37.0.dev0
• Pytorch 2.0.1+cu118
• Datasets 2.16.1
• Tokenizers 0.15.0